diff --git a/latency-aware/walker2d/small-policy/sample-factory-qwengr00t-rtx3090-v1/README.md b/latency-aware/walker2d/small-policy/sample-factory-qwengr00t-rtx3090-v1/README.md new file mode 100644 index 0000000000000000000000000000000000000000..de5af40ebf2f3f602e0df6b96ad088a11ae4c10e --- /dev/null +++ b/latency-aware/walker2d/small-policy/sample-factory-qwengr00t-rtx3090-v1/README.md @@ -0,0 +1,26 @@ +# walker2d / sample-factory-appo + +Training condition: `latency-aware`. Run: `walker2d_profile_appo_h1_5fps_20260917`. + +Copied from [Standard-Pipeline](https://huggingface.co/latency-sensitive-bench/Standard-Pipeline/tree/2208875f92b21e3f8ec2363a30a6705dfe56cc04/walker2d/profile_latency/small_model/GR00T) at `2208875f92b21e3f8ec2363a30a6705dfe56cc04`. + +- Checkpoint: `checkpoint.pth` +- Selection: training_best +- Checkpoint SHA256: `c0f38c72c5aa5c4a17565a7ca4dc63bf95e70a656dc446a05d2e41b1a0f50740` +- Configuration, FPS, actions, normalization and latency semantics retain their source values. +- This import does not run evaluations or change training/evaluation/acceptance status. + +Sample Factory APPO inference bundle: `config.json` and the explicit `checkpoint.pth`. +The checkpoint contains `model`, `train_step`, and `env_steps`; every exported model +tensor was reloaded and compared bit-for-bit with the selected source checkpoint. + +## Download + +```python +from huggingface_hub import snapshot_download +prefix = "latency-aware/walker2d/small-policy/sample-factory-qwengr00t-rtx3090-v1" +snapshot_download("latency-sensitive-bench/benchmark-models", revision="", + allow_patterns=[prefix + "/**"]) +``` + +Use the verified import commit listed in the repository's import report. diff --git a/latency-aware/walker2d/small-policy/sample-factory-qwengr00t-rtx3090-v1/checkpoint.pth b/latency-aware/walker2d/small-policy/sample-factory-qwengr00t-rtx3090-v1/checkpoint.pth new file mode 100644 index 0000000000000000000000000000000000000000..ed4a4dece47453f0d5734d68d93f6c670188ce6b --- /dev/null +++ b/latency-aware/walker2d/small-policy/sample-factory-qwengr00t-rtx3090-v1/checkpoint.pth @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c0f38c72c5aa5c4a17565a7ca4dc63bf95e70a656dc446a05d2e41b1a0f50740 +size 29877 diff --git a/latency-aware/walker2d/small-policy/sample-factory-qwengr00t-rtx3090-v1/config.full.yaml b/latency-aware/walker2d/small-policy/sample-factory-qwengr00t-rtx3090-v1/config.full.yaml new file mode 100644 index 0000000000000000000000000000000000000000..df19f258cbfb9a7c92af5bf042358f4aee3c48ce --- /dev/null +++ b/latency-aware/walker2d/small-policy/sample-factory-qwengr00t-rtx3090-v1/config.full.yaml @@ -0,0 +1,155 @@ +experiment: + name: walker2d_profile_appo_h1_5fps_20260917 + seed: 0 +backend: + type: sample_factory + algo: APPO + device: cuda + train_dir: /mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/teacher/checkpoints + restart_behavior: overwrite + run_mode: train + extra_args: + - --wandb_user + - dongqianyu99-zhejiang-university +executor: + mode: simulated +env: + name: gymnasium + task_name: walker2d + env_id: LatencyBench/Walker2dContinuous-v0 + registration_imports: + - latency_bench.envs.gymnasium_walker2d + make_kwargs: + base_env_id: Walker2d-v4 + render_mode: rgb_array + base_make_kwargs: + forward_reward_weight: 1.0 + ctrl_cost_weight: 0.001 + healthy_reward: 1.0 + terminate_when_unhealthy: true + reset_noise_scale: 0.005 + exclude_current_positions_from_observation: true + env_fps: 5 + obs_fps: 5 + frame_stack: 1 + action_space: + type: box + labels: + - right_thigh_torque + - right_leg_torque + - right_foot_torque + - left_thigh_torque + - left_leg_torque + - left_foot_torque + low: + - -1.0 + - -1.0 + - -1.0 + - -1.0 + - -1.0 + - -1.0 + high: + - 1.0 + - 1.0 + - 1.0 + - 1.0 + - 1.0 + - 1.0 + dtype: float32 + noop_action: + - 0.0 + - 0.0 + - 0.0 + - 0.0 + - 0.0 + - 0.0 + base_prompt: Move the Walker2d robot forward while keeping its torso upright. Predict + six continuous torques in [-1, 1] ordered as right thigh, right leg, right foot, + left thigh, left leg, and left foot. + state_space: + labels: + - torso_height + - torso_angle + - right_thigh_angle + - right_leg_angle + - right_foot_angle + - left_thigh_angle + - left_leg_angle + - left_foot_angle + - torso_x_velocity + - torso_z_velocity + - torso_angular_velocity + - right_thigh_angular_velocity + - right_leg_angular_velocity + - right_foot_angular_velocity + - left_thigh_angular_velocity + - left_leg_angular_velocity + - left_foot_angular_velocity +latency: + method: iid + profile_path: /mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/profile_3090_20260917T164637291306Z/profile.json + profile_worker_slot: 0 + seed: 0 + add_latency_info: false +scheduler: + hold_policy: hold + ordering_policy: issue_order_fifo +policy: + type: random +training: + train_for_env_steps: 10000000 + num_workers: 8 + num_envs_per_worker: 8 + worker_num_splits: 2 + num_policies: 1 + batch_size: 1024 + rollout: 64 + recurrence: 1 + num_epochs: 2 + num_batches_per_epoch: 4 + max_policy_lag: 300 + learning_rate: 0.00295 + gamma: 0.99 + gae_lambda: 0.95 + ppo_clip_ratio: 0.2 + ppo_clip_value: 0.2 + value_loss_coeff: 1.3 + max_grad_norm: 3.5 + exploration_loss_coeff: 0.0 + async_rl: false + batched_sampling: false + use_rnn: false + encoder_mlp_layers: + - 64 + - 64 + normalize_input: true + normalize_returns: true + save_every_sec: 600 + keep_checkpoints: 5 + lr_schedule: linear_decay + kl_loss_coeff: 0.1 + serial_mode: false + nonlinearity: tanh + adaptive_stddev: false + policy_initialization: torch_default + initial_stddev: 1.0 + shuffle_minibatches: false + value_bootstrap: true +evaluation: + eval_interval_steps: null + eval_episodes: 20 + eval_parallel_envs: 1 + eval_max_steps: 1000 + eval_deterministic: true + eval_latency_values: null +logging: + output_dir: /mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/teacher/training + video: + enabled: false + num_bins: 1 + save_step_records: false + save_action_records: false + save_latency_records: false + wandb_project: latency-sensitive-bench + wandb_group: gr00t-six-env-h1-5fps + wandb_job_type: profile_teacher diff --git a/latency-aware/walker2d/small-policy/sample-factory-qwengr00t-rtx3090-v1/config.json b/latency-aware/walker2d/small-policy/sample-factory-qwengr00t-rtx3090-v1/config.json new file mode 100644 index 0000000000000000000000000000000000000000..9fa492afe9006c39b207d5ab861be4e49a0d0c98 --- /dev/null +++ b/latency-aware/walker2d/small-policy/sample-factory-qwengr00t-rtx3090-v1/config.json @@ -0,0 +1,256 @@ +{ + "help": false, + "algo": "APPO", + "env": "latency_gymnasium", + "experiment": "walker2d_profile_appo_h1_5fps_20260917", + "train_dir": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/teacher/checkpoints", + "restart_behavior": "overwrite", + "device": "gpu", + "seed": 0, + "initial_model_path": null, + "num_policies": 1, + "async_rl": false, + "serial_mode": false, + "batched_sampling": false, + "num_batches_to_accumulate": 2, + "worker_num_splits": 2, + "policy_workers_per_policy": 1, + "max_policy_lag": 300, + "num_workers": 8, + "num_envs_per_worker": 8, + "batch_size": 1024, + "num_batches_per_epoch": 4, + "num_epochs": 2, + "rollout": 64, + "fasttd3_replay_capacity": 6553600, + "fasttd3_replay_batch_size": 32768, + "fasttd3_action_chunk_horizon": 1, + "fasttd3_transitions_per_update": 64, + "fasttd3_train_for_optimizer_steps": 10000000000, + "fasttd3_v_min": -250.0, + "fasttd3_v_max": 250.0, + "fasttd3_actor_action_l2": 0.0, + "fasttd3_compile": true, + "recurrence": 1, + "shuffle_minibatches": false, + "gamma": 0.99, + "reward_scale": 1.0, + "reward_clip": 1000.0, + "value_bootstrap": true, + "normalize_returns": true, + "exploration_loss_coeff": 0.0, + "value_loss_coeff": 1.3, + "kl_loss_coeff": 0.1, + "exploration_loss": "entropy", + "gae_lambda": 0.95, + "ppo_clip_ratio": 0.2, + "ppo_clip_value": 0.2, + "with_vtrace": false, + "vtrace_rho": 1.0, + "vtrace_c": 1.0, + "optimizer": "adam", + "adam_eps": 1e-06, + "adam_beta1": 0.9, + "adam_beta2": 0.999, + "max_grad_norm": 3.5, + "learning_rate": 0.00295, + "lr_schedule": "linear_decay", + "lr_schedule_kl_threshold": 0.008, + "lr_adaptive_min": 1e-06, + "lr_adaptive_max": 0.01, + "obs_subtract_mean": 0.0, + "obs_scale": 1.0, + "normalize_input": true, + "normalize_input_keys": null, + "decorrelate_experience_max_seconds": 0, + "decorrelate_envs_on_one_worker": true, + "actor_worker_gpus": [], + "set_workers_cpu_affinity": true, + "force_envs_single_thread": false, + "default_niceness": 0, + "log_to_file": true, + "experiment_summaries_interval": 10, + "flush_summaries_interval": 30, + "stats_avg": 100, + "summaries_use_frameskip": true, + "heartbeat_interval": 20, + "heartbeat_reporting_interval": 180, + "train_for_env_steps": 10000000, + "train_for_seconds": 10000000000, + "save_every_sec": 600, + "keep_checkpoints": 5, + "load_checkpoint_kind": "latest", + "save_milestones_sec": -1, + "save_best_every_sec": 5, + "save_best_metric": "reward", + "save_best_after": 100000, + "benchmark": false, + "encoder_mlp_layers": [ + 64, + 64 + ], + "encoder_conv_architecture": "convnet_simple", + "encoder_conv_mlp_layers": [ + 512 + ], + "use_rnn": false, + "rnn_size": 512, + "rnn_type": "gru", + "rnn_num_layers": 1, + "decoder_mlp_layers": [], + "nonlinearity": "tanh", + "policy_initialization": "torch_default", + "policy_init_gain": 1.0, + "actor_critic_share_weights": true, + "adaptive_stddev": false, + "continuous_tanh_scale": 0.0, + "initial_stddev": 1.0, + "use_env_info_cache": false, + "env_gpu_actions": false, + "env_gpu_observations": true, + "env_frameskip": 1, + "env_framestack": 1, + "pixel_format": "CHW", + "use_record_episode_statistics": false, + "with_wandb": true, + "wandb_user": "dongqianyu99-zhejiang-university", + "wandb_project": "latency-sensitive-bench", + "wandb_group": "gr00t-six-env-h1-5fps", + "wandb_job_type": "profile_teacher", + "wandb_tags": null, + "with_pbt": false, + "pbt_mix_policies_in_one_env": true, + "pbt_period_env_steps": 5000000, + "pbt_start_mutation": 20000000, + "pbt_replace_fraction": 0.3, + "pbt_mutation_rate": 0.15, + "pbt_replace_reward_gap": 0.1, + "pbt_replace_reward_gap_absolute": 1e-06, + "pbt_optimize_gamma": false, + "pbt_target_objective": "true_objective", + "pbt_perturb_min": 1.1, + "pbt_perturb_max": 1.5, + "gym_task_name": "walker2d", + "gym_env_id": "LatencyBench/Walker2dContinuous-v0", + "gym_make_kwargs_json": "{\"base_env_id\": \"Walker2d-v4\", \"render_mode\": \"rgb_array\", \"base_make_kwargs\": {\"forward_reward_weight\": 1.0, \"ctrl_cost_weight\": 0.001, \"healthy_reward\": 1.0, \"terminate_when_unhealthy\": true, \"reset_noise_scale\": 0.005, \"exclude_current_positions_from_observation\": true}}", + "gym_registration_imports_json": "[\"latency_bench.envs.gymnasium_walker2d\"]", + "gym_action_space_json": "{\"type\": \"box\", \"labels\": [\"right_thigh_torque\", \"right_leg_torque\", \"right_foot_torque\", \"left_thigh_torque\", \"left_leg_torque\", \"left_foot_torque\"], \"low\": [-1.0, -1.0, -1.0, -1.0, -1.0, -1.0], \"high\": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0], \"dtype\": \"float32\"}", + "gym_action_labels_json": null, + "gym_action_values_json": null, + "gym_noop_action_id": null, + "gym_noop_action_json": "[0.0, 0.0, 0.0, 0.0, 0.0, 0.0]", + "gym_base_prompt": "Move the Walker2d robot forward while keeping its torso upright. Predict six continuous torques in [-1, 1] ordered as right thigh, right leg, right foot, left thigh, left leg, and left foot.", + "gym_state_labels_json": "[\"torso_height\", \"torso_angle\", \"right_thigh_angle\", \"right_leg_angle\", \"right_foot_angle\", \"left_thigh_angle\", \"left_leg_angle\", \"left_foot_angle\", \"torso_x_velocity\", \"torso_z_velocity\", \"torso_angular_velocity\", \"right_thigh_angular_velocity\", \"right_leg_angular_velocity\", \"right_foot_angular_velocity\", \"left_thigh_angular_velocity\", \"left_leg_angular_velocity\", \"left_foot_angular_velocity\"]", + "env_fps": 5.0, + "obs_fps": 5.0, + "frame_stack": 1, + "export_env_raw_rgb_frames": false, + "mode": "train", + "latency_type": "iid", + "fixed_latency_ms": null, + "mean_latency_ms": null, + "std_latency_ms": null, + "min_latency_ms": null, + "max_latency_ms": null, + "latency_seed": 0, + "latency_profile_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/profile_3090_20260917T164637291306Z/profile.json", + "latency_profile_worker_slot": 0, + "simulated_worker_capacity": null, + "add_latency_info": false, + "max_pending_actions": null, + "hold_last_chunk_action": false, + "hold_policy": "hold", + "ordering_policy": "issue_order_fifo", + "eval_episodes": 20, + "eval_parallel_envs": 1, + "eval_latency_raw_frame_values": null, + "eval_max_steps": 1000, + "eval_deterministic": true, + "eval_raw_reward": false, + "episode_metrics_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/teacher/checkpoints/walker2d_profile_appo_h1_5fps_20260917/episode_metrics.jsonl", + "ppo": null, + "command_line": "--mode train --algo APPO --env latency_gymnasium --experiment walker2d_profile_appo_h1_5fps_20260917 --train_dir /mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/teacher/checkpoints --restart_behavior overwrite --device gpu --seed 0 --episode_metrics_path /mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/teacher/checkpoints/walker2d_profile_appo_h1_5fps_20260917/episode_metrics.jsonl --train_for_env_steps 10000000 --num_workers 8 --num_envs_per_worker 8 --num_policies 1 --batch_size 1024 --rollout 64 --recurrence 1 --num_epochs 2 --num_batches_per_epoch 4 --worker_num_splits 2 --max_policy_lag 300 --learning_rate 0.00295 --kl_loss_coeff 0.1 --nonlinearity tanh --policy_initialization torch_default --initial_stddev 1.0 --gamma 0.99 --gae_lambda 0.95 --ppo_clip_ratio 0.2 --ppo_clip_value 0.2 --exploration_loss_coeff 0.0 --value_loss_coeff 1.3 --max_grad_norm 3.5 --save_every_sec 600 --keep_checkpoints 5 --async_rl False --batched_sampling False --serial_mode False --use_rnn False --normalize_returns True --normalize_input True --adaptive_stddev False --shuffle_minibatches False --value_bootstrap True --encoder_mlp_layers 64 64 --latency-type iid --latency-seed 0 --latency-profile-path /mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/profile_3090_20260917T164637291306Z/profile.json --latency-profile-worker-slot 0 --add-latency-info False --eval-episodes 20 --eval-parallel-envs 1 --eval-max-steps 1000 --eval-deterministic True --with_wandb True --wandb_project latency-sensitive-bench --wandb_group gr00t-six-env-h1-5fps --wandb_job_type profile_teacher --wandb_user dongqianyu99-zhejiang-university --gym-task-name walker2d --gym-env-id LatencyBench/Walker2dContinuous-v0 --gym-make-kwargs-json {\"base_env_id\": \"Walker2d-v4\", \"render_mode\": \"rgb_array\", \"base_make_kwargs\": {\"forward_reward_weight\": 1.0, \"ctrl_cost_weight\": 0.001, \"healthy_reward\": 1.0, \"terminate_when_unhealthy\": true, \"reset_noise_scale\": 0.005, \"exclude_current_positions_from_observation\": true}} --gym-registration-imports-json [\"latency_bench.envs.gymnasium_walker2d\"] --gym-action-space-json {\"type\": \"box\", \"labels\": [\"right_thigh_torque\", \"right_leg_torque\", \"right_foot_torque\", \"left_thigh_torque\", \"left_leg_torque\", \"left_foot_torque\"], \"low\": [-1.0, -1.0, -1.0, -1.0, -1.0, -1.0], \"high\": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0], \"dtype\": \"float32\"} --gym-noop-action-json [0.0, 0.0, 0.0, 0.0, 0.0, 0.0] --gym-base-prompt Move the Walker2d robot forward while keeping its torso upright. Predict six continuous torques in [-1, 1] ordered as right thigh, right leg, right foot, left thigh, left leg, and left foot. --gym-state-labels-json [\"torso_height\", \"torso_angle\", \"right_thigh_angle\", \"right_leg_angle\", \"right_foot_angle\", \"left_thigh_angle\", \"left_leg_angle\", \"left_foot_angle\", \"torso_x_velocity\", \"torso_z_velocity\", \"torso_angular_velocity\", \"right_thigh_angular_velocity\", \"right_leg_angular_velocity\", \"right_foot_angular_velocity\", \"left_thigh_angular_velocity\", \"left_leg_angular_velocity\", \"left_foot_angular_velocity\"] --env-fps 5 --obs-fps 5 --frame-stack 1 --hold-policy hold --ordering-policy issue_order_fifo --use_env_info_cache False", + "cli_args": { + "algo": "APPO", + "env": "latency_gymnasium", + "experiment": "walker2d_profile_appo_h1_5fps_20260917", + "train_dir": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/teacher/checkpoints", + "restart_behavior": "overwrite", + "device": "gpu", + "seed": 0, + "num_policies": 1, + "async_rl": false, + "serial_mode": false, + "batched_sampling": false, + "worker_num_splits": 2, + "max_policy_lag": 300, + "num_workers": 8, + "num_envs_per_worker": 8, + "batch_size": 1024, + "num_batches_per_epoch": 4, + "num_epochs": 2, + "rollout": 64, + "recurrence": 1, + "shuffle_minibatches": false, + "gamma": 0.99, + "value_bootstrap": true, + "normalize_returns": true, + "exploration_loss_coeff": 0.0, + "value_loss_coeff": 1.3, + "kl_loss_coeff": 0.1, + "gae_lambda": 0.95, + "ppo_clip_ratio": 0.2, + "ppo_clip_value": 0.2, + "max_grad_norm": 3.5, + "learning_rate": 0.00295, + "normalize_input": true, + "train_for_env_steps": 10000000, + "save_every_sec": 600, + "keep_checkpoints": 5, + "encoder_mlp_layers": [ + 64, + 64 + ], + "use_rnn": false, + "nonlinearity": "tanh", + "policy_initialization": "torch_default", + "adaptive_stddev": false, + "initial_stddev": 1.0, + "use_env_info_cache": false, + "with_wandb": true, + "wandb_user": "dongqianyu99-zhejiang-university", + "wandb_project": "latency-sensitive-bench", + "wandb_group": "gr00t-six-env-h1-5fps", + "wandb_job_type": "profile_teacher", + "gym_task_name": "walker2d", + "gym_env_id": "LatencyBench/Walker2dContinuous-v0", + "gym_make_kwargs_json": "{\"base_env_id\": \"Walker2d-v4\", \"render_mode\": \"rgb_array\", \"base_make_kwargs\": {\"forward_reward_weight\": 1.0, \"ctrl_cost_weight\": 0.001, \"healthy_reward\": 1.0, \"terminate_when_unhealthy\": true, \"reset_noise_scale\": 0.005, \"exclude_current_positions_from_observation\": true}}", + "gym_registration_imports_json": "[\"latency_bench.envs.gymnasium_walker2d\"]", + "gym_action_space_json": "{\"type\": \"box\", \"labels\": [\"right_thigh_torque\", \"right_leg_torque\", \"right_foot_torque\", \"left_thigh_torque\", \"left_leg_torque\", \"left_foot_torque\"], \"low\": [-1.0, -1.0, -1.0, -1.0, -1.0, -1.0], \"high\": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0], \"dtype\": \"float32\"}", + "gym_noop_action_json": "[0.0, 0.0, 0.0, 0.0, 0.0, 0.0]", + "gym_base_prompt": "Move the Walker2d robot forward while keeping its torso upright. Predict six continuous torques in [-1, 1] ordered as right thigh, right leg, right foot, left thigh, left leg, and left foot.", + "gym_state_labels_json": "[\"torso_height\", \"torso_angle\", \"right_thigh_angle\", \"right_leg_angle\", \"right_foot_angle\", \"left_thigh_angle\", \"left_leg_angle\", \"left_foot_angle\", \"torso_x_velocity\", \"torso_z_velocity\", \"torso_angular_velocity\", \"right_thigh_angular_velocity\", \"right_leg_angular_velocity\", \"right_foot_angular_velocity\", \"left_thigh_angular_velocity\", \"left_leg_angular_velocity\", \"left_foot_angular_velocity\"]", + "env_fps": 5.0, + "obs_fps": 5.0, + "frame_stack": 1, + "mode": "train", + "latency_type": "iid", + "latency_seed": 0, + "latency_profile_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/profile_3090_20260917T164637291306Z/profile.json", + "latency_profile_worker_slot": 0, + "add_latency_info": false, + "hold_policy": "hold", + "ordering_policy": "issue_order_fifo", + "eval_episodes": 20, + "eval_parallel_envs": 1, + "eval_max_steps": 1000, + "eval_deterministic": true, + "episode_metrics_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/teacher/checkpoints/walker2d_profile_appo_h1_5fps_20260917/episode_metrics.jsonl", + "output_dir": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/teacher/training" + }, + "git_hash": "unknown", + "git_repo_name": "not a git repository", + "eval_env_frameskip": 1, + "output_dir": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/teacher/training", + "wandb_unique_id": "walker2d_profile_appo_h1_5fps_20260917" +} \ No newline at end of file diff --git a/latency-aware/walker2d/small-policy/sample-factory-qwengr00t-rtx3090-v1/profile/DONE b/latency-aware/walker2d/small-policy/sample-factory-qwengr00t-rtx3090-v1/profile/DONE new file mode 100644 index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391 diff --git a/latency-aware/walker2d/small-policy/sample-factory-qwengr00t-rtx3090-v1/profile/hardware.json b/latency-aware/walker2d/small-policy/sample-factory-qwengr00t-rtx3090-v1/profile/hardware.json new file mode 100644 index 0000000000000000000000000000000000000000..e8204e81807044975c0e0a0953d16d6835192f31 --- /dev/null +++ b/latency-aware/walker2d/small-policy/sample-factory-qwengr00t-rtx3090-v1/profile/hardware.json @@ -0,0 +1,40 @@ +{ + "driver_version": "580.173.02", + "gpu_class": "1x-rtx3090", + "gpus": [ + { + "name": "NVIDIA GeForce RTX 3090", + "slot": 0 + } + ], + "instance_id": "instance_859cf1e47bca6046", + "topology_links": [], + "torch": { + "backends": { + "cuda_cudnn_sdp_enabled": true, + "cuda_flash_sdp_enabled": true, + "cuda_math_sdp_enabled": true, + "cuda_matmul_allow_tf32": false, + "cuda_mem_efficient_sdp_enabled": true, + "cudnn_allow_tf32": true, + "cudnn_benchmark": false + }, + "cuda_available": true, + "cuda_device_count": 1, + "cuda_version": "12.8", + "current_device": 0, + "current_device_name": "NVIDIA GeForce RTX 3090", + "device_properties": [ + { + "index": 0, + "major": 8, + "minor": 6, + "multi_processor_count": 82, + "name": "NVIDIA GeForce RTX 3090", + "total_memory": 25295257600 + } + ], + "float32_matmul_precision": "highest", + "version": "2.11.0+cu128" + } +} diff --git a/latency-aware/walker2d/small-policy/sample-factory-qwengr00t-rtx3090-v1/profile/latency_burst_model.json b/latency-aware/walker2d/small-policy/sample-factory-qwengr00t-rtx3090-v1/profile/latency_burst_model.json new file mode 100644 index 0000000000000000000000000000000000000000..9fb77aea7efff37dbb5ff7955c2673f7fc9f1322 --- /dev/null +++ b/latency-aware/walker2d/small-policy/sample-factory-qwengr00t-rtx3090-v1/profile/latency_burst_model.json @@ -0,0 +1,1308 @@ +{ + "burst_dwell_distribution": { + "distribution_type": "inverse_cdf", + "latency_ms": [ + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0 + ], + "quantile_levels": [ + 0.0, + 0.0001, + 0.0005, + 0.001, + 0.002, + 0.003, + 0.004, + 0.005, + 0.006, + 0.007, + 0.008, + 0.009, + 0.01, + 0.02, + 0.03, + 0.04, + 0.05, + 0.06, + 0.07, + 0.08, + 0.09, + 0.1, + 0.11, + 0.12, + 0.13, + 0.14, + 0.15, + 0.16, + 0.17, + 0.18, + 0.19, + 0.2, + 0.21, + 0.22, + 0.23, + 0.24, + 0.25, + 0.26, + 0.27, + 0.28, + 0.29, + 0.3, + 0.31, + 0.32, + 0.33, + 0.34, + 0.35, + 0.36, + 0.37, + 0.38, + 0.39, + 0.4, + 0.41, + 0.42, + 0.43, + 0.44, + 0.45, + 0.46, + 0.47, + 0.48, + 0.49, + 0.5, + 0.51, + 0.52, + 0.53, + 0.54, + 0.55, + 0.56, + 0.57, + 0.58, + 0.59, + 0.6, + 0.61, + 0.62, + 0.63, + 0.64, + 0.65, + 0.66, + 0.67, + 0.68, + 0.69, + 0.7, + 0.71, + 0.72, + 0.73, + 0.74, + 0.75, + 0.76, + 0.77, + 0.78, + 0.79, + 0.8, + 0.81, + 0.82, + 0.83, + 0.84, + 0.85, + 0.86, + 0.87, + 0.88, + 0.89, + 0.9, + 0.91, + 0.92, + 0.93, + 0.94, + 0.95, + 0.96, + 0.97, + 0.98, + 0.99, + 0.991, + 0.992, + 0.993, + 0.994, + 0.995, + 0.996, + 0.997, + 0.998, + 0.999, + 0.9995, + 0.9999, + 1.0 + ] + }, + "burst_dwell_lengths": [ + 1 + ], + "burst_merge_gap_records": 30, + "burst_rank_processes": [ + { + "draw_count": 1, + "dwell_length_spearman_rho": 0.0, + "level_ms": { + "distribution_type": "inverse_cdf", + "latency_ms": [ + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396, + 159.24888014793396 + ], + "quantile_levels": [ + 0.0, + 0.0001, + 0.0005, + 0.001, + 0.002, + 0.003, + 0.004, + 0.005, + 0.006, + 0.007, + 0.008, + 0.009, + 0.01, + 0.02, + 0.03, + 0.04, + 0.05, + 0.06, + 0.07, + 0.08, + 0.09, + 0.1, + 0.11, + 0.12, + 0.13, + 0.14, + 0.15, + 0.16, + 0.17, + 0.18, + 0.19, + 0.2, + 0.21, + 0.22, + 0.23, + 0.24, + 0.25, + 0.26, + 0.27, + 0.28, + 0.29, + 0.3, + 0.31, + 0.32, + 0.33, + 0.34, + 0.35, + 0.36, + 0.37, + 0.38, + 0.39, + 0.4, + 0.41, + 0.42, + 0.43, + 0.44, + 0.45, + 0.46, + 0.47, + 0.48, + 0.49, + 0.5, + 0.51, + 0.52, + 0.53, + 0.54, + 0.55, + 0.56, + 0.57, + 0.58, + 0.59, + 0.6, + 0.61, + 0.62, + 0.63, + 0.64, + 0.65, + 0.66, + 0.67, + 0.68, + 0.69, + 0.7, + 0.71, + 0.72, + 0.73, + 0.74, + 0.75, + 0.76, + 0.77, + 0.78, + 0.79, + 0.8, + 0.81, + 0.82, + 0.83, + 0.84, + 0.85, + 0.86, + 0.87, + 0.88, + 0.89, + 0.9, + 0.91, + 0.92, + 0.93, + 0.94, + 0.95, + 0.96, + 0.97, + 0.98, + 0.99, + 0.991, + 0.992, + 0.993, + 0.994, + 0.995, + 0.996, + 0.997, + 0.998, + 0.999, + 0.9995, + 0.9999, + 1.0 + ] + }, + "level_residual_ms": { + "distribution_type": "inverse_cdf", + "latency_ms": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ], + "quantile_levels": [ + 0.0, + 0.0001, + 0.0005, + 0.001, + 0.002, + 0.003, + 0.004, + 0.005, + 0.006, + 0.007, + 0.008, + 0.009, + 0.01, + 0.02, + 0.03, + 0.04, + 0.05, + 0.06, + 0.07, + 0.08, + 0.09, + 0.1, + 0.11, + 0.12, + 0.13, + 0.14, + 0.15, + 0.16, + 0.17, + 0.18, + 0.19, + 0.2, + 0.21, + 0.22, + 0.23, + 0.24, + 0.25, + 0.26, + 0.27, + 0.28, + 0.29, + 0.3, + 0.31, + 0.32, + 0.33, + 0.34, + 0.35, + 0.36, + 0.37, + 0.38, + 0.39, + 0.4, + 0.41, + 0.42, + 0.43, + 0.44, + 0.45, + 0.46, + 0.47, + 0.48, + 0.49, + 0.5, + 0.51, + 0.52, + 0.53, + 0.54, + 0.55, + 0.56, + 0.57, + 0.58, + 0.59, + 0.6, + 0.61, + 0.62, + 0.63, + 0.64, + 0.65, + 0.66, + 0.67, + 0.68, + 0.69, + 0.7, + 0.71, + 0.72, + 0.73, + 0.74, + 0.75, + 0.76, + 0.77, + 0.78, + 0.79, + 0.8, + 0.81, + 0.82, + 0.83, + 0.84, + 0.85, + 0.86, + 0.87, + 0.88, + 0.89, + 0.9, + 0.91, + 0.92, + 0.93, + 0.94, + 0.95, + 0.96, + 0.97, + 0.98, + 0.99, + 0.991, + 0.992, + 0.993, + 0.994, + 0.995, + 0.996, + 0.997, + 0.998, + 0.999, + 0.9995, + 0.9999, + 1.0 + ] + }, + "severity": { + "distribution_type": "inverse_cdf", + "latency_ms": [ + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0 + ], + "quantile_levels": [ + 0.0, + 0.0001, + 0.0005, + 0.001, + 0.002, + 0.003, + 0.004, + 0.005, + 0.006, + 0.007, + 0.008, + 0.009, + 0.01, + 0.02, + 0.03, + 0.04, + 0.05, + 0.06, + 0.07, + 0.08, + 0.09, + 0.1, + 0.11, + 0.12, + 0.13, + 0.14, + 0.15, + 0.16, + 0.17, + 0.18, + 0.19, + 0.2, + 0.21, + 0.22, + 0.23, + 0.24, + 0.25, + 0.26, + 0.27, + 0.28, + 0.29, + 0.3, + 0.31, + 0.32, + 0.33, + 0.34, + 0.35, + 0.36, + 0.37, + 0.38, + 0.39, + 0.4, + 0.41, + 0.42, + 0.43, + 0.44, + 0.45, + 0.46, + 0.47, + 0.48, + 0.49, + 0.5, + 0.51, + 0.52, + 0.53, + 0.54, + 0.55, + 0.56, + 0.57, + 0.58, + 0.59, + 0.6, + 0.61, + 0.62, + 0.63, + 0.64, + 0.65, + 0.66, + 0.67, + 0.68, + 0.69, + 0.7, + 0.71, + 0.72, + 0.73, + 0.74, + 0.75, + 0.76, + 0.77, + 0.78, + 0.79, + 0.8, + 0.81, + 0.82, + 0.83, + 0.84, + 0.85, + 0.86, + 0.87, + 0.88, + 0.89, + 0.9, + 0.91, + 0.92, + 0.93, + 0.94, + 0.95, + 0.96, + 0.97, + 0.98, + 0.99, + 0.991, + 0.992, + 0.993, + 0.994, + 0.995, + 0.996, + 0.997, + 0.998, + 0.999, + 0.9995, + 0.9999, + 1.0 + ] + }, + "spike_count": 1 + } + ], + "burst_slot_rank_templates": [ + { + "dwell_length": 1, + "slot_to_rank": { + "0": 0 + } + } + ], + "model_type": "hidden_regime", + "pre_worker_time_ms": { + "distribution_type": "inverse_cdf", + "latency_ms": [ + 0.32036399841308594, + 0.32036399841308594, + 0.32723311603069305, + 0.33272179365158083, + 0.34245591163635253, + 0.3487426149845123, + 0.35062161445617673, + 0.3524500966072083, + 0.35448283672332764, + 0.3576815378665924, + 0.3607575559616089, + 0.36199055671691893, + 0.3643078088760376, + 0.3815332889556885, + 0.394054114818573, + 0.4045921802520752, + 0.4232257008552551, + 0.588078451156616, + 0.6740153789520265, + 0.7390505075454712, + 1.1690887093544007, + 1.1805009841918945, + 1.1854936599731445, + 1.190834927558899, + 1.1936466217041015, + 1.196754789352417, + 1.2016456127166748, + 1.2052122354507446, + 1.2098347425460816, + 1.2136710166931153, + 1.2186432838439942, + 1.2227401733398438, + 1.2264718174934388, + 1.2304659366607666, + 1.2377743244171142, + 1.242726945877075, + 1.2481653690338135, + 1.2525454998016357, + 1.2580512285232544, + 1.265002679824829, + 1.2748169422149658, + 1.2793519496917725, + 1.2833239436149597, + 1.2864039421081543, + 1.288594913482666, + 1.2904037475585939, + 1.2922656536102295, + 1.2936686992645263, + 1.2952197790145874, + 1.2970149040222168, + 1.2982266306877137, + 1.2992517948150635, + 1.3002004027366638, + 1.3013105869293213, + 1.3023244500160218, + 1.3032267332077025, + 1.3042846322059631, + 1.3051421642303467, + 1.3060412645339965, + 1.3069873332977295, + 1.3077953457832336, + 1.3088939189910889, + 1.30997976064682, + 1.3111885070800782, + 1.3117629051208497, + 1.3126658916473388, + 1.313631296157837, + 1.3146527767181397, + 1.3157230734825134, + 1.3167790412902831, + 1.3174383997917176, + 1.3183507919311523, + 1.3192538976669312, + 1.3202075004577636, + 1.321112072467804, + 1.3222339153289795, + 1.3233216404914856, + 1.3245097160339356, + 1.3256892442703248, + 1.3268343687057496, + 1.327985405921936, + 1.3291511535644531, + 1.3307852745056152, + 1.332238245010376, + 1.3338016510009765, + 1.3351406574249267, + 1.336811363697052, + 1.3386332511901855, + 1.3400704741477967, + 1.3418533325195312, + 1.3438435316085815, + 1.3454315662384033, + 1.3476183533668518, + 1.3502938270568847, + 1.3530936598777772, + 1.3553231954574585, + 1.3583070039749146, + 1.3610630989074708, + 1.3649658799171447, + 1.3685470581054688, + 1.373920977115631, + 1.3793320655822754, + 1.3871524214744568, + 1.3926500082015991, + 1.3995159149169922, + 1.4119826316833495, + 1.4287474155426025, + 1.4381157636642456, + 1.4493753790855408, + 1.4671900272369385, + 1.4888291239738467, + 1.4907235157489775, + 1.4924510049819946, + 1.4941827166080475, + 1.4968405318260192, + 1.5028127133846283, + 1.5108259582519525, + 1.518246532678605, + 1.5366208744049068, + 1.6046243298053764, + 1.6402621507644752, + 1.778925895690918, + 1.778925895690918 + ], + "quantile_levels": [ + 0.0, + 0.0001, + 0.0005, + 0.001, + 0.002, + 0.003, + 0.004, + 0.005, + 0.006, + 0.007, + 0.008, + 0.009, + 0.01, + 0.02, + 0.03, + 0.04, + 0.05, + 0.06, + 0.07, + 0.08, + 0.09, + 0.1, + 0.11, + 0.12, + 0.13, + 0.14, + 0.15, + 0.16, + 0.17, + 0.18, + 0.19, + 0.2, + 0.21, + 0.22, + 0.23, + 0.24, + 0.25, + 0.26, + 0.27, + 0.28, + 0.29, + 0.3, + 0.31, + 0.32, + 0.33, + 0.34, + 0.35, + 0.36, + 0.37, + 0.38, + 0.39, + 0.4, + 0.41, + 0.42, + 0.43, + 0.44, + 0.45, + 0.46, + 0.47, + 0.48, + 0.49, + 0.5, + 0.51, + 0.52, + 0.53, + 0.54, + 0.55, + 0.56, + 0.57, + 0.58, + 0.59, + 0.6, + 0.61, + 0.62, + 0.63, + 0.64, + 0.65, + 0.66, + 0.67, + 0.68, + 0.69, + 0.7, + 0.71, + 0.72, + 0.73, + 0.74, + 0.75, + 0.76, + 0.77, + 0.78, + 0.79, + 0.8, + 0.81, + 0.82, + 0.83, + 0.84, + 0.85, + 0.86, + 0.87, + 0.88, + 0.89, + 0.9, + 0.91, + 0.92, + 0.93, + 0.94, + 0.95, + 0.96, + 0.97, + 0.98, + 0.99, + 0.991, + 0.992, + 0.993, + 0.994, + 0.995, + 0.996, + 0.997, + 0.998, + 0.999, + 0.9995, + 0.9999, + 1.0 + ] + }, + "regime_step_counts": { + "burst": 1, + "calm": 3094 + }, + "regime_transition_counts": { + "burst": { + "burst": 0, + "calm": 1 + }, + "calm": { + "burst": 1, + "calm": 3088 + } + }, + "reset_scope": "session", + "schema_version": 12, + "spike_median_multiplier": 1.25, + "spike_threshold_ms_by_worker_slot": { + "0": 135.06769254803658 + }, + "worker_count": 1 +} diff --git a/latency-aware/walker2d/small-policy/sample-factory-qwengr00t-rtx3090-v1/profile/latency_distribution.json b/latency-aware/walker2d/small-policy/sample-factory-qwengr00t-rtx3090-v1/profile/latency_distribution.json new file mode 100644 index 0000000000000000000000000000000000000000..5e63d9ce5275743c1183763f2ee38d615e61c26c --- /dev/null +++ b/latency-aware/walker2d/small-policy/sample-factory-qwengr00t-rtx3090-v1/profile/latency_distribution.json @@ -0,0 +1,1027 @@ +{ + "schema_version": 3, + "worker_slots": { + "0": { + "all": { + "count": 3095, + "observation_to_action_latency_ms": { + "distribution_type": "inverse_cdf", + "latency_ms": [ + 99.22320699691772, + 99.22320699691772, + 99.42916572034359, + 99.58709894537925, + 99.93997703313828, + 100.03245514512062, + 100.10576476097107, + 100.27338070869446, + 100.43154655694961, + 100.54514897346496, + 100.58786373138427, + 100.73122339248657, + 100.93288924694062, + 101.83683857917785, + 102.17618277072907, + 102.78488368988037, + 103.20275801420212, + 103.53771271705628, + 103.85141879320145, + 104.17247416973115, + 104.4531753540039, + 104.67428803443909, + 104.9984158873558, + 105.19882352352143, + 105.36013214588165, + 105.530988073349, + 105.69296550750732, + 105.84973270893097, + 105.99273630380631, + 106.12834568023682, + 106.27271245718002, + 106.45480394363403, + 106.58185195922852, + 106.70940933227538, + 106.84196033477784, + 106.97506012916566, + 107.14249867200851, + 107.33465428352356, + 107.54035266637803, + 107.81159157752991, + 108.04246901273727, + 108.23629212379456, + 108.52575784921646, + 108.8793125629425, + 109.29595267772675, + 109.5309287071228, + 109.81568521261215, + 110.03037614822388, + 110.22245434522628, + 110.44733695983886, + 110.7103401184082, + 110.85259914398193, + 110.97723712921143, + 111.13601436614991, + 111.27977132797241, + 111.44822533130646, + 111.55709743499756, + 111.66463627815247, + 111.7913564801216, + 111.89587769508361, + 111.99657224416733, + 112.08105206489563, + 112.17835522890091, + 112.24816064834594, + 112.3046331524849, + 112.37526683807373, + 112.48177295923233, + 112.56383666992187, + 112.66144201755523, + 112.80137748718262, + 112.87896525859833, + 112.9791659116745, + 113.06426136493683, + 113.15538420677186, + 113.25501419305802, + 113.35301032066346, + 113.45775592327118, + 113.59696660041809, + 113.73112956285476, + 113.86885089874268, + 113.97324702739715, + 114.07272100448608, + 114.18734756708145, + 114.33366961479187, + 114.4551468372345, + 114.60303611755371, + 114.74404376745224, + 114.87333765029908, + 115.02998926639557, + 115.21705346107483, + 115.35996603965759, + 115.51110208034515, + 115.73145149946212, + 115.8619243621826, + 116.05849715471268, + 116.29468376636504, + 116.46762311458588, + 116.60534596443176, + 116.71262007951736, + 116.87904567718506, + 117.05468183755875, + 117.31163811683655, + 117.53008776903152, + 117.71672832965851, + 117.88815916776657, + 118.11798558235168, + 118.31915485858917, + 118.52074434757233, + 118.7206486582756, + 119.12539472579955, + 119.64151037931443, + 119.72379802703857, + 119.82497741699218, + 119.89392696619034, + 119.95471127033234, + 120.0299242734909, + 120.2847660255432, + 120.95733683109297, + 121.91578974008559, + 124.49807678222673, + 132.0581759798551, + 159.91820216178894, + 159.91820216178894 + ], + "quantile_levels": [ + 0.0, + 0.0001, + 0.0005, + 0.001, + 0.002, + 0.003, + 0.004, + 0.005, + 0.006, + 0.007, + 0.008, + 0.009, + 0.01, + 0.02, + 0.03, + 0.04, + 0.05, + 0.06, + 0.07, + 0.08, + 0.09, + 0.1, + 0.11, + 0.12, + 0.13, + 0.14, + 0.15, + 0.16, + 0.17, + 0.18, + 0.19, + 0.2, + 0.21, + 0.22, + 0.23, + 0.24, + 0.25, + 0.26, + 0.27, + 0.28, + 0.29, + 0.3, + 0.31, + 0.32, + 0.33, + 0.34, + 0.35, + 0.36, + 0.37, + 0.38, + 0.39, + 0.4, + 0.41, + 0.42, + 0.43, + 0.44, + 0.45, + 0.46, + 0.47, + 0.48, + 0.49, + 0.5, + 0.51, + 0.52, + 0.53, + 0.54, + 0.55, + 0.56, + 0.57, + 0.58, + 0.59, + 0.6, + 0.61, + 0.62, + 0.63, + 0.64, + 0.65, + 0.66, + 0.67, + 0.68, + 0.69, + 0.7, + 0.71, + 0.72, + 0.73, + 0.74, + 0.75, + 0.76, + 0.77, + 0.78, + 0.79, + 0.8, + 0.81, + 0.82, + 0.83, + 0.84, + 0.85, + 0.86, + 0.87, + 0.88, + 0.89, + 0.9, + 0.91, + 0.92, + 0.93, + 0.94, + 0.95, + 0.96, + 0.97, + 0.98, + 0.99, + 0.991, + 0.992, + 0.993, + 0.994, + 0.995, + 0.996, + 0.997, + 0.998, + 0.999, + 0.9995, + 0.9999, + 1.0 + ] + }, + "spearman_rho": 0.9995485054067693, + "worker_service_time_ms": { + "distribution_type": "inverse_cdf", + "latency_ms": [ + 98.75452613830566, + 98.75452613830566, + 98.85438686907291, + 99.02571328401565, + 99.45779752254487, + 99.59049253582954, + 99.65989756584167, + 99.70000897049904, + 99.91803278923035, + 100.05549708247185, + 100.16672918319702, + 100.28120213747025, + 100.35547195672989, + 101.29415669441224, + 101.74761251211166, + 102.14199221134186, + 102.41365730762482, + 102.65487484931946, + 102.93827275037765, + 103.12642040252686, + 103.3542575955391, + 103.63435292243958, + 103.81708480119705, + 103.98665144443513, + 104.14164572954178, + 104.28317828178406, + 104.47517436742783, + 104.60352334976196, + 104.76213237047196, + 104.87410960197448, + 105.04031020402908, + 105.19616520404816, + 105.32718687057495, + 105.45814895629883, + 105.57121027708054, + 105.70132374763489, + 105.86968612670898, + 106.07166357040406, + 106.30302008390427, + 106.52139294147491, + 106.76226021051407, + 106.99370408058167, + 107.2647524356842, + 107.57935655117035, + 107.97918523550034, + 108.25467529296876, + 108.51256901025772, + 108.72574026584626, + 108.9212993979454, + 109.13033485412598, + 109.40382331609726, + 109.52390909194946, + 109.67240022420883, + 109.81641945838928, + 109.98103808164596, + 110.12876727581025, + 110.24520576000214, + 110.36289238929749, + 110.47831959724427, + 110.58496022224426, + 110.66886518001556, + 110.75457000732422, + 110.8484365940094, + 110.92572956085205, + 110.98994392156601, + 111.0608488559723, + 111.15075767040253, + 111.25024995803832, + 111.34399528503418, + 111.45287566184997, + 111.56470843553544, + 111.6521270275116, + 111.76101347208024, + 111.87006192207336, + 111.97028158903122, + 112.040327334404, + 112.16831284761429, + 112.26242628097535, + 112.4190979361534, + 112.55684206485749, + 112.65307712554932, + 112.76242399215698, + 112.86989911794662, + 113.01847443580627, + 113.14175227880477, + 113.29016342163087, + 113.41451501846313, + 113.55460324287415, + 113.72017403841019, + 113.88516240119934, + 114.0360654592514, + 114.20723950862885, + 114.40640790462494, + 114.56653847694398, + 114.7652627468109, + 114.98923602104186, + 115.16235131025314, + 115.29728622436524, + 115.41436080932617, + 115.55528964996338, + 115.75665469169617, + 116.0090708732605, + 116.20922166109085, + 116.39691112041473, + 116.58660510778427, + 116.81812901496887, + 117.00732922554016, + 117.21111283302307, + 117.43425514698029, + 117.8212960243225, + 118.36746326684951, + 118.45590266108513, + 118.54080009937286, + 118.6620908510685, + 118.7169402194023, + 118.84759024977686, + 119.11940640449524, + 120.22933041453364, + 120.82420707464213, + 123.00159268975266, + 131.4062336653499, + 159.24888014793396, + 159.24888014793396 + ], + "quantile_levels": [ + 0.0, + 0.0001, + 0.0005, + 0.001, + 0.002, + 0.003, + 0.004, + 0.005, + 0.006, + 0.007, + 0.008, + 0.009, + 0.01, + 0.02, + 0.03, + 0.04, + 0.05, + 0.06, + 0.07, + 0.08, + 0.09, + 0.1, + 0.11, + 0.12, + 0.13, + 0.14, + 0.15, + 0.16, + 0.17, + 0.18, + 0.19, + 0.2, + 0.21, + 0.22, + 0.23, + 0.24, + 0.25, + 0.26, + 0.27, + 0.28, + 0.29, + 0.3, + 0.31, + 0.32, + 0.33, + 0.34, + 0.35, + 0.36, + 0.37, + 0.38, + 0.39, + 0.4, + 0.41, + 0.42, + 0.43, + 0.44, + 0.45, + 0.46, + 0.47, + 0.48, + 0.49, + 0.5, + 0.51, + 0.52, + 0.53, + 0.54, + 0.55, + 0.56, + 0.57, + 0.58, + 0.59, + 0.6, + 0.61, + 0.62, + 0.63, + 0.64, + 0.65, + 0.66, + 0.67, + 0.68, + 0.69, + 0.7, + 0.71, + 0.72, + 0.73, + 0.74, + 0.75, + 0.76, + 0.77, + 0.78, + 0.79, + 0.8, + 0.81, + 0.82, + 0.83, + 0.84, + 0.85, + 0.86, + 0.87, + 0.88, + 0.89, + 0.9, + 0.91, + 0.92, + 0.93, + 0.94, + 0.95, + 0.96, + 0.97, + 0.98, + 0.99, + 0.991, + 0.992, + 0.993, + 0.994, + 0.995, + 0.996, + 0.997, + 0.998, + 0.999, + 0.9995, + 0.9999, + 1.0 + ] + } + }, + "steady": { + "count": 3094, + "observation_to_action_latency_ms": { + "distribution_type": "inverse_cdf", + "latency_ms": [ + 99.22320699691772, + 99.22320699691772, + 99.4291475265026, + 99.58689176225663, + 99.9398753490448, + 100.03237170553207, + 100.10574906158448, + 100.27301820755005, + 100.43145009040832, + 100.54510977363586, + 100.58768297195435, + 100.73106924057006, + 100.93229277610779, + 101.83682395935058, + 102.17598546028137, + 102.78485960006714, + 103.2027048110962, + 103.53631356239319, + 103.84930500984191, + 104.17240930080413, + 104.45295521736145, + 104.67423934936524, + 104.99618706703185, + 105.19881187915801, + 105.35895367622375, + 105.5295336675644, + 105.69243893623351, + 105.84872452259064, + 105.99248383522034, + 106.12704199790954, + 106.27097979545593, + 106.45084190368652, + 106.57948274612427, + 106.70563902378082, + 106.84049203872681, + 106.97161992073059, + 107.14191794395447, + 107.33255678653717, + 107.53844790458679, + 107.80959053993224, + 108.04170256614685, + 108.23570296764373, + 108.52342710494995, + 108.87580090522766, + 109.28260650634766, + 109.53078701019287, + 109.8145525932312, + 110.0303715133667, + 110.21772505760192, + 110.44604891300202, + 110.7100734424591, + 110.85057506561279, + 110.976790599823, + 111.13558660030365, + 111.27604258537292, + 111.44789675712586, + 111.55351853370667, + 111.6602259683609, + 111.79104159355164, + 111.89472816944122, + 111.9902446269989, + 112.08060002326965, + 112.17793950080872, + 112.24794021606445, + 112.30365852355958, + 112.37482343673706, + 112.47858295440673, + 112.56282970428467, + 112.66075355529784, + 112.80043536663055, + 112.87583395004272, + 112.97757012844086, + 113.06352269172669, + 113.15406580924987, + 113.25393080711365, + 113.35285932064056, + 113.45304865837097, + 113.5951464176178, + 113.72895416259766, + 113.86850547790527, + 113.97013641357422, + 114.07239472866058, + 114.18514239311219, + 114.3329379940033, + 114.45335556030274, + 114.59983550548553, + 114.73854184150696, + 114.86755680561066, + 115.02933004379273, + 115.21143441200256, + 115.35923015594483, + 115.50828003883362, + 115.73064454078674, + 115.84734886169434, + 116.05421135902405, + 116.28132198810577, + 116.4657564163208, + 116.60340900421143, + 116.71045592308045, + 116.87004503250122, + 117.04853163719177, + 117.3057951927185, + 117.52893462181092, + 117.71306704998017, + 117.86754645347595, + 118.1131922674179, + 118.31688604354858, + 118.51486089229583, + 118.70966945648193, + 119.11284963607788, + 119.62059361457824, + 119.70220011806488, + 119.80803647613526, + 119.88963515996933, + 119.94102578544617, + 120.0157650923729, + 120.16904916000364, + 120.5495177865028, + 121.73250816345214, + 123.09343721055976, + 124.95696385955809, + 132.41034197807312, + 132.41034197807312 + ], + "quantile_levels": [ + 0.0, + 0.0001, + 0.0005, + 0.001, + 0.002, + 0.003, + 0.004, + 0.005, + 0.006, + 0.007, + 0.008, + 0.009, + 0.01, + 0.02, + 0.03, + 0.04, + 0.05, + 0.06, + 0.07, + 0.08, + 0.09, + 0.1, + 0.11, + 0.12, + 0.13, + 0.14, + 0.15, + 0.16, + 0.17, + 0.18, + 0.19, + 0.2, + 0.21, + 0.22, + 0.23, + 0.24, + 0.25, + 0.26, + 0.27, + 0.28, + 0.29, + 0.3, + 0.31, + 0.32, + 0.33, + 0.34, + 0.35, + 0.36, + 0.37, + 0.38, + 0.39, + 0.4, + 0.41, + 0.42, + 0.43, + 0.44, + 0.45, + 0.46, + 0.47, + 0.48, + 0.49, + 0.5, + 0.51, + 0.52, + 0.53, + 0.54, + 0.55, + 0.56, + 0.57, + 0.58, + 0.59, + 0.6, + 0.61, + 0.62, + 0.63, + 0.64, + 0.65, + 0.66, + 0.67, + 0.68, + 0.69, + 0.7, + 0.71, + 0.72, + 0.73, + 0.74, + 0.75, + 0.76, + 0.77, + 0.78, + 0.79, + 0.8, + 0.81, + 0.82, + 0.83, + 0.84, + 0.85, + 0.86, + 0.87, + 0.88, + 0.89, + 0.9, + 0.91, + 0.92, + 0.93, + 0.94, + 0.95, + 0.96, + 0.97, + 0.98, + 0.99, + 0.991, + 0.992, + 0.993, + 0.994, + 0.995, + 0.996, + 0.997, + 0.998, + 0.999, + 0.9995, + 0.9999, + 1.0 + ] + }, + "spearman_rho": 0.9995480674876682, + "worker_service_time_ms": { + "distribution_type": "inverse_cdf", + "latency_ms": [ + 98.75452613830566, + 98.75452613830566, + 98.8543078815937, + 99.02567823314666, + 99.45775291824341, + 99.59041699552536, + 99.65989508628846, + 99.70000818014145, + 99.91795456886291, + 100.0552305226326, + 100.1665827217102, + 100.28072974681854, + 100.35500584602356, + 101.2941440153122, + 101.74742962837219, + 102.14179589748383, + 102.413609457016, + 102.65296932220458, + 102.93745662689209, + 103.12594020843505, + 103.3539143371582, + 103.63286294937134, + 103.81621097564697, + 103.98584180355073, + 104.14075045585632, + 104.28279706478119, + 104.47343454360961, + 104.602457447052, + 104.7621068763733, + 104.87407166481017, + 105.03943796157837, + 105.19612519741058, + 105.32668438911438, + 105.45810830593109, + 105.57101434707641, + 105.7006136417389, + 105.86729383468628, + 106.06924108982086, + 106.30191291809082, + 106.52006046295166, + 106.76016472816467, + 106.98808867931366, + 107.26399293899536, + 107.57933251857757, + 107.97763031959533, + 108.25383613586426, + 108.50995388031006, + 108.7242898130417, + 108.92072467803955, + 109.12973146438598, + 109.40061864852905, + 109.52376661300659, + 109.67168751716613, + 109.81073287963868, + 109.97755250930786, + 110.12792809963226, + 110.24433798789978, + 110.35856191158295, + 110.47724398612976, + 110.58463953971862, + 110.66824443817139, + 110.7544914484024, + 110.8479015827179, + 110.9249896621704, + 110.988516664505, + 111.06074045181275, + 111.15060529708862, + 111.24939653396606, + 111.34373544692993, + 111.44646321296692, + 111.56446446418762, + 111.65086011886596, + 111.75772401809692, + 111.86346021175385, + 111.96789213180541, + 112.03838314056397, + 112.16716780662537, + 112.26036375045777, + 112.41615811347961, + 112.55308448314666, + 112.65031149864197, + 112.75928657054901, + 112.86168020248412, + 113.01474835395813, + 113.138128824234, + 113.28559089183807, + 113.41284990310669, + 113.5493541765213, + 113.70897500991822, + 113.88268896102905, + 114.0343407535553, + 114.20164093971253, + 114.40476523399353, + 114.56448984622955, + 114.75662749290467, + 114.98597189426422, + 115.16058959960938, + 115.29675541877747, + 115.41023595809936, + 115.55201926231385, + 115.75309795379638, + 115.99993162155151, + 116.20096955299377, + 116.39203874588013, + 116.58490141868592, + 116.8105647611618, + 117.00442953109741, + 117.21098148822784, + 117.4254603767395, + 117.8143440246582, + 118.34197818756103, + 118.4202447347641, + 118.50866367816926, + 118.61377112197876, + 118.69165815544129, + 118.79300454854966, + 119.0119661836624, + 119.6289191842078, + 120.43082016181944, + 122.17210936594003, + 123.20034849286078, + 131.81460094451904, + 131.81460094451904 + ], + "quantile_levels": [ + 0.0, + 0.0001, + 0.0005, + 0.001, + 0.002, + 0.003, + 0.004, + 0.005, + 0.006, + 0.007, + 0.008, + 0.009, + 0.01, + 0.02, + 0.03, + 0.04, + 0.05, + 0.06, + 0.07, + 0.08, + 0.09, + 0.1, + 0.11, + 0.12, + 0.13, + 0.14, + 0.15, + 0.16, + 0.17, + 0.18, + 0.19, + 0.2, + 0.21, + 0.22, + 0.23, + 0.24, + 0.25, + 0.26, + 0.27, + 0.28, + 0.29, + 0.3, + 0.31, + 0.32, + 0.33, + 0.34, + 0.35, + 0.36, + 0.37, + 0.38, + 0.39, + 0.4, + 0.41, + 0.42, + 0.43, + 0.44, + 0.45, + 0.46, + 0.47, + 0.48, + 0.49, + 0.5, + 0.51, + 0.52, + 0.53, + 0.54, + 0.55, + 0.56, + 0.57, + 0.58, + 0.59, + 0.6, + 0.61, + 0.62, + 0.63, + 0.64, + 0.65, + 0.66, + 0.67, + 0.68, + 0.69, + 0.7, + 0.71, + 0.72, + 0.73, + 0.74, + 0.75, + 0.76, + 0.77, + 0.78, + 0.79, + 0.8, + 0.81, + 0.82, + 0.83, + 0.84, + 0.85, + 0.86, + 0.87, + 0.88, + 0.89, + 0.9, + 0.91, + 0.92, + 0.93, + 0.94, + 0.95, + 0.96, + 0.97, + 0.98, + 0.99, + 0.991, + 0.992, + 0.993, + 0.994, + 0.995, + 0.996, + 0.997, + 0.998, + 0.999, + 0.9995, + 0.9999, + 1.0 + ] + } + } + } + } +} diff --git a/latency-aware/walker2d/small-policy/sample-factory-qwengr00t-rtx3090-v1/profile/latency_profile.png b/latency-aware/walker2d/small-policy/sample-factory-qwengr00t-rtx3090-v1/profile/latency_profile.png new file mode 100644 index 0000000000000000000000000000000000000000..b9a809b7c9b345ae7bfd0c6caf9d6aace40d7916 Binary files /dev/null and b/latency-aware/walker2d/small-policy/sample-factory-qwengr00t-rtx3090-v1/profile/latency_profile.png differ diff --git a/latency-aware/walker2d/small-policy/sample-factory-qwengr00t-rtx3090-v1/profile/profile.json b/latency-aware/walker2d/small-policy/sample-factory-qwengr00t-rtx3090-v1/profile/profile.json new file mode 100644 index 0000000000000000000000000000000000000000..d3e8799b9858188febc70f53168e7969431d069a --- /dev/null +++ b/latency-aware/walker2d/small-policy/sample-factory-qwengr00t-rtx3090-v1/profile/profile.json @@ -0,0 +1,71 @@ +{ + "burst_model_path": "latency_burst_model.json", + "distribution_path": "latency_distribution.json", + "env_fps": 5, + "frame_ms": 200.0, + "gpu_class": "1x-rtx3090", + "instance_id": "instance_859cf1e47bca6046", + "latency_kind": "observation_to_action_latency", + "latency_method": "temporal", + "model_id": "gr00t", + "n_admitted_observations": 3095, + "n_capacity_drops": 0, + "n_observation_attempts": 3095, + "per_slot_summary": { + "0": { + "admitted_count": 3095, + "mean_observation_to_action_latency_ms": 111.28701411079321, + "mean_worker_service_time_ms": 110.04660924409241, + "p95_observation_to_action_latency_ms": 118.31860404014587, + "p95_worker_service_time_ms": 117.0066906452179, + "p99_worker_service_time_ms": 118.36731338024138 + } + }, + "provenance": { + "base_config": "/workspace/lzj/latency-sensitive-bench/runs/walker2d/gr00t_h1_5fps_20260917/P.yaml", + "checkpoint_kind": "latest", + "model_artifact": { + "action_horizon": 1, + "checkpoint_sha256": "5df7bf29e8cc0bc984cb21ad4f8db673f16b17ceb438f4f4c91be65ee52c4de1", + "clock_fps": 5, + "hf_prefix": "walker2d/zero_latency/GR00T", + "hf_repo": "latency-sensitive-bench/Standard-Pipeline", + "hf_revision": "9ea50f42a1e93a9035c8df6a392e5e9f141493de", + "published": true, + "source": "huggingface", + "training_condition": "zero_latency", + "training_updates": 5000 + }, + "session_ids": [ + 0, + 1, + 2, + 3, + 4 + ] + }, + "sample_model_type": "hidden_regime", + "source_run_id": "20260917T164637291306Z", + "summary": { + "frame_ms": 200.0, + "max_ms": 159.91820216178894, + "mean_effective_frames": 0.556435070553966, + "mean_ms": 111.28701411079321, + "min_ms": 99.22320699691772, + "n_samples": 3095, + "p50_frames": 0.5604052603244781, + "p50_ms": 112.08105206489563, + "p90_frames": 0.5865452063083648, + "p90_ms": 117.30904126167297, + "p95_frames": 0.5915930202007293, + "p95_ms": 118.31860404014587, + "p99_frames": 0.5981799455404282, + "p99_ms": 119.63598910808564, + "prob_latency_gt_1_frame": 0.0, + "prob_latency_gt_2_frames": 0.0, + "prob_latency_gt_3_frames": 0.0, + "std_ms": 4.800312930327855 + }, + "visualization_path": "latency_profile.png", + "workload_id": "walker2d" +} diff --git a/latency-aware/walker2d/small-policy/sample-factory-qwengr00t-rtx3090-v1/provenance.json b/latency-aware/walker2d/small-policy/sample-factory-qwengr00t-rtx3090-v1/provenance.json new file mode 100644 index 0000000000000000000000000000000000000000..11a00fbf1f3e4e0c5e79eae5434b30ccdf275689 --- /dev/null +++ b/latency-aware/walker2d/small-policy/sample-factory-qwengr00t-rtx3090-v1/provenance.json @@ -0,0 +1,44 @@ +{ + "task": "walker2d", + "model": "sample-factory-appo", + "training_condition": "latency-aware", + "training_run_id": "walker2d_profile_appo_h1_5fps_20260917", + "source": { + "repo_id": "latency-sensitive-bench/Standard-Pipeline", + "revision": "2208875f92b21e3f8ec2363a30a6705dfe56cc04", + "path": "walker2d/profile_latency/small_model/GR00T", + "url": "https://huggingface.co/latency-sensitive-bench/Standard-Pipeline/tree/2208875f92b21e3f8ec2363a30a6705dfe56cc04/walker2d/profile_latency/small_model/GR00T" + }, + "hf_repo_id": "latency-sensitive-bench/benchmark-models", + "hf_path": "latency-aware/walker2d/small-policy/sample-factory-qwengr00t-rtx3090-v1", + "checkpoint": { + "source_file": "walker2d/profile_latency/small_model/GR00T/checkpoints/selected.pth", + "source_sha256": "a9f0e63f8094fe10f82d2395fc7b59958dc22ac1abc59a6da52772eb1fd944ab", + "source_bytes": 83973, + "file": "checkpoint.pth", + "selection_rule": "training_best", + "method": "inference_export", + "sha256": "c0f38c72c5aa5c4a17565a7ca4dc63bf95e70a656dc446a05d2e41b1a0f50740", + "bytes": 29877, + "train_step": 14536, + "env_steps": 7442432, + "tensor_count": 15, + "tensor_equality_verified": true, + "removed_fields": [ + "best_performance", + "curr_lr", + "optimizer" + ] + }, + "config_source": "walker2d/profile_latency/small_model/GR00T/config.json", + "task_contract_file": null, + "missing_source_files": [], + "migration": { + "operation": "copy_selected_inference_bundle", + "source_modified": false, + "evaluation_performed": false, + "experiment_status": "preserved_in_source_evidence" + }, + "algorithm": "APPO", + "latency_model": "qwengr00t" +} diff --git a/latency-aware/walker2d/small-policy/sample-factory-qwengr00t-rtx3090-v1/selection.json b/latency-aware/walker2d/small-policy/sample-factory-qwengr00t-rtx3090-v1/selection.json new file mode 100644 index 0000000000000000000000000000000000000000..d3d89d9816d4d9aa5f39ab819798c690d5bcd44d --- /dev/null +++ b/latency-aware/walker2d/small-policy/sample-factory-qwengr00t-rtx3090-v1/selection.json @@ -0,0 +1,66 @@ +{ + "selected": "training_best", + "candidates": { + "final": { + "checkpoint": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/teacher/checkpoints/walker2d_profile_appo_h1_5fps_20260917/checkpoint_p0/checkpoint_000019536_10002432.pth", + "sha256": "dcfe291db668f89db2fead0028a76fb5a005965b9daaab7a29bd19eb13c43ca0", + "episodes": 20, + "returns": [ + 4299.208001651971, + 4253.381546607346, + 4274.26267070416, + 4281.764485481581, + 4316.5965023405115, + 4185.72463542741, + 4268.629567678589, + 4265.304288469715, + 4331.873574180083, + 3269.8377559505593, + 3583.941712492686, + 4264.461568188047, + 4268.998666838904, + 4293.114110604841, + 4274.87183914279, + 4358.281606522924, + 4247.960230740107, + 4363.387419962692, + 4258.273723341316, + 4286.0125421364155 + ], + "mean": 4197.294322423132, + "std": 264.3552584808844, + "strict_gt3000": 20 + }, + "training_best": { + "checkpoint": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/teacher/checkpoints/walker2d_profile_appo_h1_5fps_20260917/checkpoint_p0/best_000014536_7442432_reward_4186.861.pth", + "sha256": "a9f0e63f8094fe10f82d2395fc7b59958dc22ac1abc59a6da52772eb1fd944ab", + "episodes": 20, + "returns": [ + 4232.421844289257, + 4211.834128554937, + 4235.009681429353, + 4216.192126194387, + 4225.968774742952, + 4246.130242600228, + 4259.369933944325, + 4211.69597436189, + 4227.7128439529915, + 4219.107550141977, + 4176.009289189152, + 4216.315331519501, + 4233.440001863116, + 4210.597842410642, + 4216.913958972969, + 4236.856116118128, + 4241.158612744575, + 4224.4414913018245, + 4191.601768644205, + 4233.529106106846 + ], + "mean": 4223.315330954163, + "std": 18.175464884559204, + "strict_gt3000": 20 + } + }, + "completed_at": "2026-09-17T17:24:56.075104+00:00" +} diff --git a/latency-aware/walker2d/small-policy/sample-factory-qwengr00t-rtx3090-v1/teacher-preparation.json b/latency-aware/walker2d/small-policy/sample-factory-qwengr00t-rtx3090-v1/teacher-preparation.json new file mode 100644 index 0000000000000000000000000000000000000000..494dbc7876a062156d54c4b6ca62b3605bdf9517 --- /dev/null +++ b/latency-aware/walker2d/small-policy/sample-factory-qwengr00t-rtx3090-v1/teacher-preparation.json @@ -0,0 +1,16 @@ +{ + "state": "PROFILE_TEACHER_CONFIG_READY", + "profile": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/profile_3090_20260917T164637291306Z/profile.json", + "profile_sha256": "acf0a997fa7866b5a101714ec1479b8cc19b36ae1262f11081ab828de3e79661", + "source_recipe": "/mnt/local/lzj/latency-sensitive-bench/code/h1-5fps-20260917-v2/configs/examples/gymnasium/walker2d/small_model_train_sf_official_10m.yaml", + "source_recipe_sha256": "ef0abc795b6c33d152dbdeaa4a5a817df2546f400c17ce27ac9159e01bb92901", + "config": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/teacher.yaml", + "algo": "APPO", + "budget_env_steps": 10000000, + "seed": 0, + "fps": 5, + "latency": "iid", + "initialization": "fresh", + "metadata_derivation": "current task_name walker2d and state17 labels from wrapper; original official10m recipe seed0/hyperparameters/native physics preserved", + "next": "Run only after immutable new P is verified; Final/best20 tieFinal, E10, bounded12probe, strictreturn>3000." +} diff --git a/latency-aware/walker2d/small-policy/sample-factory-qwengr00t-rtx3090-v1/validation/E10-summary.json b/latency-aware/walker2d/small-policy/sample-factory-qwengr00t-rtx3090-v1/validation/E10-summary.json new file mode 100644 index 0000000000000000000000000000000000000000..80f49dbfa7e679118647ffd97c49088515180b46 --- /dev/null +++ b/latency-aware/walker2d/small-policy/sample-factory-qwengr00t-rtx3090-v1/validation/E10-summary.json @@ -0,0 +1,20 @@ +{ + "checkpoint": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/teacher/checkpoints/walker2d_profile_appo_h1_5fps_20260917/checkpoint_p0/best_000014536_7442432_reward_4186.861.pth", + "sha256": "a9f0e63f8094fe10f82d2395fc7b59958dc22ac1abc59a6da52772eb1fd944ab", + "episodes": 10, + "returns": [ + 4232.421844289257, + 4211.834128554937, + 4235.009681429353, + 4216.192126194387, + 4225.968774742952, + 4246.130242600228, + 4259.369933944325, + 4211.69597436189, + 4227.7128439529915, + 4219.107550141977 + ], + "mean": 4228.54431002123, + "std": 14.588936196978617, + "strict_gt3000": 10 +} diff --git a/latency-aware/walker2d/small-policy/sample-factory-qwengr00t-rtx3090-v1/validation/E10/episode_metrics.jsonl b/latency-aware/walker2d/small-policy/sample-factory-qwengr00t-rtx3090-v1/validation/E10/episode_metrics.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..3ddfc03373dfb3a4e04be24c258fdd5488386ee3 --- /dev/null +++ b/latency-aware/walker2d/small-policy/sample-factory-qwengr00t-rtx3090-v1/validation/E10/episode_metrics.jsonl @@ -0,0 +1,10 @@ +{"episode_id": 0, "episode_return": 4232.421844289257, "episode_return_env": 4232.421844289257, "game_score": null, "mean_latency_ms": 111.69516596958135, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/teacher/checkpoints/walker2d_profile_appo_h1_5fps_20260917/checkpoint_p0/best_000014536_7442432_reward_4186.861.pth", "config_name": "walker2d_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Walker2dContinuous-v0", "episode_seed": 104771, "frame_ms": 200.0, "gpu_class": "1x-rtx3090", "idle_worker_count": 0, "in_flight_count": 1, "inference_worker_count": 1, "instance_id": "instance_859cf1e47bca6046", "latency_type": "profile_sample", "mode": "simulated", "model_id": "gr00t", "obs_fps": 5.0, "output_dir": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/E10", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "walker2d_profile_appo_h1_5fps_20260917", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_profile_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/profile_3090_20260917T164637291306Z/profile.json", "source_run_id": "20260917T164637291306Z", "submitted_observation_frames": 1000, "workload_id": "walker2d"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 117.74894213442445, "p99_latency_ms": 119.81063013225, "return_raw": null, "survival_steps": 1000} +{"episode_id": 1, "episode_return": 4211.834128554937, "episode_return_env": 4211.834128554937, "game_score": null, "mean_latency_ms": 111.30146882528804, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/teacher/checkpoints/walker2d_profile_appo_h1_5fps_20260917/checkpoint_p0/best_000014536_7442432_reward_4186.861.pth", "config_name": "walker2d_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Walker2dContinuous-v0", "episode_seed": 104772, "frame_ms": 200.0, "gpu_class": "1x-rtx3090", "idle_worker_count": 0, "in_flight_count": 1, "inference_worker_count": 1, "instance_id": "instance_859cf1e47bca6046", "latency_type": "profile_sample", "mode": "simulated", "model_id": "gr00t", "obs_fps": 5.0, "output_dir": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/E10", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "walker2d_profile_appo_h1_5fps_20260917", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_profile_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/profile_3090_20260917T164637291306Z/profile.json", "source_run_id": "20260917T164637291306Z", "submitted_observation_frames": 1000, "workload_id": "walker2d"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 117.56241994943518, "p99_latency_ms": 119.35226772995084, "return_raw": null, "survival_steps": 1000} +{"episode_id": 2, "episode_return": 4235.009681429353, "episode_return_env": 4235.009681429353, "game_score": null, "mean_latency_ms": 111.1356948439972, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/teacher/checkpoints/walker2d_profile_appo_h1_5fps_20260917/checkpoint_p0/best_000014536_7442432_reward_4186.861.pth", "config_name": "walker2d_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Walker2dContinuous-v0", "episode_seed": 104773, "frame_ms": 200.0, "gpu_class": "1x-rtx3090", "idle_worker_count": 0, "in_flight_count": 1, "inference_worker_count": 1, "instance_id": "instance_859cf1e47bca6046", "latency_type": "profile_sample", "mode": "simulated", "model_id": "gr00t", "obs_fps": 5.0, "output_dir": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/E10", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "walker2d_profile_appo_h1_5fps_20260917", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_profile_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/profile_3090_20260917T164637291306Z/profile.json", "source_run_id": "20260917T164637291306Z", "submitted_observation_frames": 1000, "workload_id": "walker2d"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 117.05222681816122, "p99_latency_ms": 119.47251970056475, "return_raw": null, "survival_steps": 1000} +{"episode_id": 3, "episode_return": 4216.192126194387, "episode_return_env": 4216.192126194387, "game_score": null, "mean_latency_ms": 111.39095322995061, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/teacher/checkpoints/walker2d_profile_appo_h1_5fps_20260917/checkpoint_p0/best_000014536_7442432_reward_4186.861.pth", "config_name": "walker2d_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Walker2dContinuous-v0", "episode_seed": 104774, "frame_ms": 200.0, "gpu_class": "1x-rtx3090", "idle_worker_count": 0, "in_flight_count": 1, "inference_worker_count": 1, "instance_id": "instance_859cf1e47bca6046", "latency_type": "profile_sample", "mode": "simulated", "model_id": "gr00t", "obs_fps": 5.0, "output_dir": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/E10", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "walker2d_profile_appo_h1_5fps_20260917", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_profile_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/profile_3090_20260917T164637291306Z/profile.json", "source_run_id": "20260917T164637291306Z", "submitted_observation_frames": 1000, "workload_id": "walker2d"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 117.0485143536838, "p99_latency_ms": 119.56701831294635, "return_raw": null, "survival_steps": 1000} +{"episode_id": 4, "episode_return": 4225.968774742952, "episode_return_env": 4225.968774742952, "game_score": null, "mean_latency_ms": 111.4434478616093, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/teacher/checkpoints/walker2d_profile_appo_h1_5fps_20260917/checkpoint_p0/best_000014536_7442432_reward_4186.861.pth", "config_name": "walker2d_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Walker2dContinuous-v0", "episode_seed": 104775, "frame_ms": 200.0, "gpu_class": "1x-rtx3090", "idle_worker_count": 0, "in_flight_count": 1, "inference_worker_count": 1, "instance_id": "instance_859cf1e47bca6046", "latency_type": "profile_sample", "mode": "simulated", "model_id": "gr00t", "obs_fps": 5.0, "output_dir": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/E10", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "walker2d_profile_appo_h1_5fps_20260917", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_profile_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/profile_3090_20260917T164637291306Z/profile.json", "source_run_id": "20260917T164637291306Z", "submitted_observation_frames": 1000, "workload_id": "walker2d"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 117.03228350329478, "p99_latency_ms": 119.5999014787056, "return_raw": null, "survival_steps": 1000} +{"episode_id": 5, "episode_return": 4246.130242600228, "episode_return_env": 4246.130242600228, "game_score": null, "mean_latency_ms": 111.27118913304983, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/teacher/checkpoints/walker2d_profile_appo_h1_5fps_20260917/checkpoint_p0/best_000014536_7442432_reward_4186.861.pth", "config_name": "walker2d_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Walker2dContinuous-v0", "episode_seed": 104776, "frame_ms": 200.0, "gpu_class": "1x-rtx3090", "idle_worker_count": 0, "in_flight_count": 1, "inference_worker_count": 1, "instance_id": "instance_859cf1e47bca6046", "latency_type": "profile_sample", "mode": "simulated", "model_id": "gr00t", "obs_fps": 5.0, "output_dir": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/E10", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "walker2d_profile_appo_h1_5fps_20260917", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_profile_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/profile_3090_20260917T164637291306Z/profile.json", "source_run_id": "20260917T164637291306Z", "submitted_observation_frames": 1000, "workload_id": "walker2d"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 116.84997029595598, "p99_latency_ms": 119.51571891712796, "return_raw": null, "survival_steps": 1000} +{"episode_id": 6, "episode_return": 4259.369933944325, "episode_return_env": 4259.369933944325, "game_score": null, "mean_latency_ms": 111.35248938694566, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/teacher/checkpoints/walker2d_profile_appo_h1_5fps_20260917/checkpoint_p0/best_000014536_7442432_reward_4186.861.pth", "config_name": "walker2d_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Walker2dContinuous-v0", "episode_seed": 104777, "frame_ms": 200.0, "gpu_class": "1x-rtx3090", "idle_worker_count": 0, "in_flight_count": 1, "inference_worker_count": 1, "instance_id": "instance_859cf1e47bca6046", "latency_type": "profile_sample", "mode": "simulated", "model_id": "gr00t", "obs_fps": 5.0, "output_dir": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/E10", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "walker2d_profile_appo_h1_5fps_20260917", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_profile_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/profile_3090_20260917T164637291306Z/profile.json", "source_run_id": "20260917T164637291306Z", "submitted_observation_frames": 1000, "workload_id": "walker2d"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 117.24712708483169, "p99_latency_ms": 119.48416732772776, "return_raw": null, "survival_steps": 1000} +{"episode_id": 7, "episode_return": 4211.69597436189, "episode_return_env": 4211.69597436189, "game_score": null, "mean_latency_ms": 111.32393599620043, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/teacher/checkpoints/walker2d_profile_appo_h1_5fps_20260917/checkpoint_p0/best_000014536_7442432_reward_4186.861.pth", "config_name": "walker2d_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Walker2dContinuous-v0", "episode_seed": 104778, "frame_ms": 200.0, "gpu_class": "1x-rtx3090", "idle_worker_count": 0, "in_flight_count": 1, "inference_worker_count": 1, "instance_id": "instance_859cf1e47bca6046", "latency_type": "profile_sample", "mode": "simulated", "model_id": "gr00t", "obs_fps": 5.0, "output_dir": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/E10", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "walker2d_profile_appo_h1_5fps_20260917", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_profile_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/profile_3090_20260917T164637291306Z/profile.json", "source_run_id": "20260917T164637291306Z", "submitted_observation_frames": 1000, "workload_id": "walker2d"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 117.6012124114765, "p99_latency_ms": 119.98007001875979, "return_raw": null, "survival_steps": 1000} +{"episode_id": 8, "episode_return": 4227.7128439529915, "episode_return_env": 4227.7128439529915, "game_score": null, "mean_latency_ms": 111.39142950446052, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/teacher/checkpoints/walker2d_profile_appo_h1_5fps_20260917/checkpoint_p0/best_000014536_7442432_reward_4186.861.pth", "config_name": "walker2d_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Walker2dContinuous-v0", "episode_seed": 104779, "frame_ms": 200.0, "gpu_class": "1x-rtx3090", "idle_worker_count": 0, "in_flight_count": 1, "inference_worker_count": 1, "instance_id": "instance_859cf1e47bca6046", "latency_type": "profile_sample", "mode": "simulated", "model_id": "gr00t", "obs_fps": 5.0, "output_dir": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/E10", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "walker2d_profile_appo_h1_5fps_20260917", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_profile_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/profile_3090_20260917T164637291306Z/profile.json", "source_run_id": "20260917T164637291306Z", "submitted_observation_frames": 1000, "workload_id": "walker2d"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 117.43350990206312, "p99_latency_ms": 119.84356332305674, "return_raw": null, "survival_steps": 1000} +{"episode_id": 9, "episode_return": 4219.107550141977, "episode_return_env": 4219.107550141977, "game_score": null, "mean_latency_ms": 111.14170808097057, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/teacher/checkpoints/walker2d_profile_appo_h1_5fps_20260917/checkpoint_p0/best_000014536_7442432_reward_4186.861.pth", "config_name": "walker2d_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Walker2dContinuous-v0", "episode_seed": 104780, "frame_ms": 200.0, "gpu_class": "1x-rtx3090", "idle_worker_count": 0, "in_flight_count": 1, "inference_worker_count": 1, "instance_id": "instance_859cf1e47bca6046", "latency_type": "profile_sample", "mode": "simulated", "model_id": "gr00t", "obs_fps": 5.0, "output_dir": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/E10", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "walker2d_profile_appo_h1_5fps_20260917", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_profile_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/profile_3090_20260917T164637291306Z/profile.json", "source_run_id": "20260917T164637291306Z", "submitted_observation_frames": 1000, "workload_id": "walker2d"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 117.19818841002535, "p99_latency_ms": 119.46448607839513, "return_raw": null, "survival_steps": 1000} diff --git a/latency-aware/walker2d/small-policy/sample-factory-qwengr00t-rtx3090-v1/validation/probe_state.json b/latency-aware/walker2d/small-policy/sample-factory-qwengr00t-rtx3090-v1/validation/probe_state.json new file mode 100644 index 0000000000000000000000000000000000000000..2184ad078fc10b67dc87f25b0672cfc73f5f6521 --- /dev/null +++ b/latency-aware/walker2d/small-policy/sample-factory-qwengr00t-rtx3090-v1/validation/probe_state.json @@ -0,0 +1,121 @@ +{ + "accepted_specs": [ + { + "attempt_idx": 0, + "episode_idx": 0, + "episode_length": 1000, + "episode_raw_frames": 1000, + "episode_raw_return": 4247.287119686604, + "seed": 0, + "split": "train" + }, + { + "attempt_idx": 1, + "episode_idx": 1, + "episode_length": 1000, + "episode_raw_frames": 1000, + "episode_raw_return": 4201.434518292546, + "seed": 1, + "split": "val" + }, + { + "attempt_idx": 2, + "episode_idx": 2, + "episode_length": 1000, + "episode_raw_frames": 1000, + "episode_raw_return": 4227.306919425726, + "seed": 2, + "split": "train" + }, + { + "attempt_idx": 3, + "episode_idx": 3, + "episode_length": 1000, + "episode_raw_frames": 1000, + "episode_raw_return": 4201.307677522302, + "seed": 3, + "split": "val" + }, + { + "attempt_idx": 4, + "episode_idx": 4, + "episode_length": 1000, + "episode_raw_frames": 1000, + "episode_raw_return": 4218.184621155262, + "seed": 4, + "split": "train" + }, + { + "attempt_idx": 5, + "episode_idx": 5, + "episode_length": 1000, + "episode_raw_frames": 1000, + "episode_raw_return": 4216.688547462225, + "seed": 5, + "split": "train" + }, + { + "attempt_idx": 6, + "episode_idx": 6, + "episode_length": 1000, + "episode_raw_frames": 1000, + "episode_raw_return": 4225.2133866250515, + "seed": 6, + "split": "train" + }, + { + "attempt_idx": 7, + "episode_idx": 7, + "episode_length": 1000, + "episode_raw_frames": 1000, + "episode_raw_return": 4231.253190428019, + "seed": 7, + "split": "val" + }, + { + "attempt_idx": 8, + "episode_idx": 8, + "episode_length": 1000, + "episode_raw_frames": 1000, + "episode_raw_return": 4215.274691671133, + "seed": 8, + "split": "train" + }, + { + "attempt_idx": 9, + "episode_idx": 9, + "episode_length": 1000, + "episode_raw_frames": 1000, + "episode_raw_return": 4244.544062376022, + "seed": 9, + "split": "train" + }, + { + "attempt_idx": 10, + "episode_idx": 10, + "episode_length": 1000, + "episode_raw_frames": 1000, + "episode_raw_return": 4246.81555891037, + "seed": 10, + "split": "train" + }, + { + "attempt_idx": 11, + "episode_idx": 11, + "episode_length": 1000, + "episode_raw_frames": 1000, + "episode_raw_return": 4235.734463244677, + "seed": 11, + "split": "train" + } + ], + "attempted_episodes": 12, + "next_attempt_idx": 12, + "rejected_episodes": 0, + "runtime_metadata": { + "checkpoint_train_step": 14536, + "env_fps": 5.0, + "frame_stack": 1, + "obs_fps": 5.0 + } +} diff --git a/latency-aware/walker2d/small-policy/sample-factory-qwengr00t-rtx3090-v1/validation/selection_final/episode_metrics.jsonl b/latency-aware/walker2d/small-policy/sample-factory-qwengr00t-rtx3090-v1/validation/selection_final/episode_metrics.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..3565272c7a601f21832272367dea1779bc147b7a --- /dev/null +++ b/latency-aware/walker2d/small-policy/sample-factory-qwengr00t-rtx3090-v1/validation/selection_final/episode_metrics.jsonl @@ -0,0 +1,20 @@ +{"episode_id": 0, "episode_return": 4299.208001651971, "episode_return_env": 4299.208001651971, "game_score": null, "mean_latency_ms": 111.69516596958135, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/teacher/checkpoints/walker2d_profile_appo_h1_5fps_20260917/checkpoint_p0/checkpoint_000019536_10002432.pth", "config_name": "walker2d_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Walker2dContinuous-v0", "episode_seed": 104771, "frame_ms": 200.0, "gpu_class": "1x-rtx3090", "idle_worker_count": 0, "in_flight_count": 1, "inference_worker_count": 1, "instance_id": "instance_859cf1e47bca6046", "latency_type": "profile_sample", "mode": "simulated", "model_id": "gr00t", "obs_fps": 5.0, "output_dir": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/selection_final", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "walker2d_profile_appo_h1_5fps_20260917", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_profile_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/profile_3090_20260917T164637291306Z/profile.json", "source_run_id": "20260917T164637291306Z", "submitted_observation_frames": 1000, "workload_id": "walker2d"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 117.74894213442445, "p99_latency_ms": 119.81063013225, "return_raw": null, "survival_steps": 1000} +{"episode_id": 1, "episode_return": 4253.381546607346, "episode_return_env": 4253.381546607346, "game_score": null, "mean_latency_ms": 111.30146882528804, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/teacher/checkpoints/walker2d_profile_appo_h1_5fps_20260917/checkpoint_p0/checkpoint_000019536_10002432.pth", "config_name": "walker2d_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Walker2dContinuous-v0", "episode_seed": 104772, "frame_ms": 200.0, "gpu_class": "1x-rtx3090", "idle_worker_count": 0, "in_flight_count": 1, "inference_worker_count": 1, "instance_id": "instance_859cf1e47bca6046", "latency_type": "profile_sample", "mode": "simulated", "model_id": "gr00t", "obs_fps": 5.0, "output_dir": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/selection_final", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "walker2d_profile_appo_h1_5fps_20260917", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_profile_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/profile_3090_20260917T164637291306Z/profile.json", "source_run_id": "20260917T164637291306Z", "submitted_observation_frames": 1000, "workload_id": "walker2d"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 117.56241994943518, "p99_latency_ms": 119.35226772995084, "return_raw": null, "survival_steps": 1000} +{"episode_id": 2, "episode_return": 4274.26267070416, "episode_return_env": 4274.26267070416, "game_score": null, "mean_latency_ms": 111.1356948439972, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/teacher/checkpoints/walker2d_profile_appo_h1_5fps_20260917/checkpoint_p0/checkpoint_000019536_10002432.pth", "config_name": "walker2d_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Walker2dContinuous-v0", "episode_seed": 104773, "frame_ms": 200.0, "gpu_class": "1x-rtx3090", "idle_worker_count": 0, "in_flight_count": 1, "inference_worker_count": 1, "instance_id": "instance_859cf1e47bca6046", "latency_type": "profile_sample", "mode": "simulated", "model_id": "gr00t", "obs_fps": 5.0, "output_dir": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/selection_final", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "walker2d_profile_appo_h1_5fps_20260917", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_profile_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/profile_3090_20260917T164637291306Z/profile.json", "source_run_id": "20260917T164637291306Z", "submitted_observation_frames": 1000, "workload_id": "walker2d"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 117.05222681816122, "p99_latency_ms": 119.47251970056475, "return_raw": null, "survival_steps": 1000} +{"episode_id": 3, "episode_return": 4281.764485481581, "episode_return_env": 4281.764485481581, "game_score": null, "mean_latency_ms": 111.39095322995061, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/teacher/checkpoints/walker2d_profile_appo_h1_5fps_20260917/checkpoint_p0/checkpoint_000019536_10002432.pth", "config_name": "walker2d_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Walker2dContinuous-v0", "episode_seed": 104774, "frame_ms": 200.0, "gpu_class": "1x-rtx3090", "idle_worker_count": 0, "in_flight_count": 1, "inference_worker_count": 1, "instance_id": "instance_859cf1e47bca6046", "latency_type": "profile_sample", "mode": "simulated", "model_id": "gr00t", "obs_fps": 5.0, "output_dir": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/selection_final", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "walker2d_profile_appo_h1_5fps_20260917", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_profile_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/profile_3090_20260917T164637291306Z/profile.json", "source_run_id": "20260917T164637291306Z", "submitted_observation_frames": 1000, "workload_id": "walker2d"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 117.0485143536838, "p99_latency_ms": 119.56701831294635, "return_raw": null, "survival_steps": 1000} +{"episode_id": 4, "episode_return": 4316.5965023405115, "episode_return_env": 4316.5965023405115, "game_score": null, "mean_latency_ms": 111.4434478616093, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/teacher/checkpoints/walker2d_profile_appo_h1_5fps_20260917/checkpoint_p0/checkpoint_000019536_10002432.pth", "config_name": "walker2d_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Walker2dContinuous-v0", "episode_seed": 104775, "frame_ms": 200.0, "gpu_class": "1x-rtx3090", "idle_worker_count": 0, "in_flight_count": 1, "inference_worker_count": 1, "instance_id": "instance_859cf1e47bca6046", "latency_type": "profile_sample", "mode": "simulated", "model_id": "gr00t", "obs_fps": 5.0, "output_dir": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/selection_final", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "walker2d_profile_appo_h1_5fps_20260917", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_profile_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/profile_3090_20260917T164637291306Z/profile.json", "source_run_id": "20260917T164637291306Z", "submitted_observation_frames": 1000, "workload_id": "walker2d"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 117.03228350329478, "p99_latency_ms": 119.5999014787056, "return_raw": null, "survival_steps": 1000} +{"episode_id": 5, "episode_return": 4185.72463542741, "episode_return_env": 4185.72463542741, "game_score": null, "mean_latency_ms": 111.27118913304983, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/teacher/checkpoints/walker2d_profile_appo_h1_5fps_20260917/checkpoint_p0/checkpoint_000019536_10002432.pth", "config_name": "walker2d_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Walker2dContinuous-v0", "episode_seed": 104776, "frame_ms": 200.0, "gpu_class": "1x-rtx3090", "idle_worker_count": 0, "in_flight_count": 1, "inference_worker_count": 1, "instance_id": "instance_859cf1e47bca6046", "latency_type": "profile_sample", "mode": "simulated", "model_id": "gr00t", "obs_fps": 5.0, "output_dir": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/selection_final", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "walker2d_profile_appo_h1_5fps_20260917", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_profile_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/profile_3090_20260917T164637291306Z/profile.json", "source_run_id": "20260917T164637291306Z", "submitted_observation_frames": 1000, "workload_id": "walker2d"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 116.84997029595598, "p99_latency_ms": 119.51571891712796, "return_raw": null, "survival_steps": 1000} +{"episode_id": 6, "episode_return": 4268.629567678589, "episode_return_env": 4268.629567678589, "game_score": null, "mean_latency_ms": 111.35248938694566, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/teacher/checkpoints/walker2d_profile_appo_h1_5fps_20260917/checkpoint_p0/checkpoint_000019536_10002432.pth", "config_name": "walker2d_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Walker2dContinuous-v0", "episode_seed": 104777, "frame_ms": 200.0, "gpu_class": "1x-rtx3090", "idle_worker_count": 0, "in_flight_count": 1, "inference_worker_count": 1, "instance_id": "instance_859cf1e47bca6046", "latency_type": "profile_sample", "mode": "simulated", "model_id": "gr00t", "obs_fps": 5.0, "output_dir": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/selection_final", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "walker2d_profile_appo_h1_5fps_20260917", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_profile_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/profile_3090_20260917T164637291306Z/profile.json", "source_run_id": "20260917T164637291306Z", "submitted_observation_frames": 1000, "workload_id": "walker2d"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 117.24712708483169, "p99_latency_ms": 119.48416732772776, "return_raw": null, "survival_steps": 1000} +{"episode_id": 7, "episode_return": 4265.304288469715, "episode_return_env": 4265.304288469715, "game_score": null, "mean_latency_ms": 111.32393599620043, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/teacher/checkpoints/walker2d_profile_appo_h1_5fps_20260917/checkpoint_p0/checkpoint_000019536_10002432.pth", "config_name": "walker2d_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Walker2dContinuous-v0", "episode_seed": 104778, "frame_ms": 200.0, "gpu_class": "1x-rtx3090", "idle_worker_count": 0, "in_flight_count": 1, "inference_worker_count": 1, "instance_id": "instance_859cf1e47bca6046", "latency_type": "profile_sample", "mode": "simulated", "model_id": "gr00t", "obs_fps": 5.0, "output_dir": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/selection_final", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "walker2d_profile_appo_h1_5fps_20260917", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_profile_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/profile_3090_20260917T164637291306Z/profile.json", "source_run_id": "20260917T164637291306Z", "submitted_observation_frames": 1000, "workload_id": "walker2d"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 117.6012124114765, "p99_latency_ms": 119.98007001875979, "return_raw": null, "survival_steps": 1000} +{"episode_id": 8, "episode_return": 4331.873574180083, "episode_return_env": 4331.873574180083, "game_score": null, "mean_latency_ms": 111.39142950446052, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/teacher/checkpoints/walker2d_profile_appo_h1_5fps_20260917/checkpoint_p0/checkpoint_000019536_10002432.pth", "config_name": "walker2d_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Walker2dContinuous-v0", "episode_seed": 104779, "frame_ms": 200.0, "gpu_class": "1x-rtx3090", "idle_worker_count": 0, "in_flight_count": 1, "inference_worker_count": 1, "instance_id": "instance_859cf1e47bca6046", "latency_type": "profile_sample", "mode": "simulated", "model_id": "gr00t", "obs_fps": 5.0, "output_dir": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/selection_final", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "walker2d_profile_appo_h1_5fps_20260917", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_profile_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/profile_3090_20260917T164637291306Z/profile.json", "source_run_id": "20260917T164637291306Z", "submitted_observation_frames": 1000, "workload_id": "walker2d"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 117.43350990206312, "p99_latency_ms": 119.84356332305674, "return_raw": null, "survival_steps": 1000} +{"episode_id": 9, "episode_return": 3269.8377559505593, "episode_return_env": 3269.8377559505593, "game_score": null, "mean_latency_ms": 111.37005565526842, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/teacher/checkpoints/walker2d_profile_appo_h1_5fps_20260917/checkpoint_p0/checkpoint_000019536_10002432.pth", "config_name": "walker2d_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Walker2dContinuous-v0", "episode_seed": 104780, "frame_ms": 200.0, "gpu_class": "1x-rtx3090", "idle_worker_count": 0, "in_flight_count": 1, "inference_worker_count": 1, "instance_id": "instance_859cf1e47bca6046", "latency_type": "profile_sample", "mode": "simulated", "model_id": "gr00t", "obs_fps": 5.0, "output_dir": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/selection_final", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "walker2d_profile_appo_h1_5fps_20260917", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_profile_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/profile_3090_20260917T164637291306Z/profile.json", "source_run_id": "20260917T164637291306Z", "submitted_observation_frames": 775, "workload_id": "walker2d"}, "num_actions": 775, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 117.25017212362863, "p99_latency_ms": 119.5234373815378, "return_raw": null, "survival_steps": 775} +{"episode_id": 10, "episode_return": 3583.941712492686, "episode_return_env": 3583.941712492686, "game_score": null, "mean_latency_ms": 111.50627686282196, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/teacher/checkpoints/walker2d_profile_appo_h1_5fps_20260917/checkpoint_p0/checkpoint_000019536_10002432.pth", "config_name": "walker2d_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Walker2dContinuous-v0", "episode_seed": 104781, "frame_ms": 200.0, "gpu_class": "1x-rtx3090", "idle_worker_count": 0, "in_flight_count": 1, "inference_worker_count": 1, "instance_id": "instance_859cf1e47bca6046", "latency_type": "profile_sample", "mode": "simulated", "model_id": "gr00t", "obs_fps": 5.0, "output_dir": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/selection_final", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "walker2d_profile_appo_h1_5fps_20260917", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_profile_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/profile_3090_20260917T164637291306Z/profile.json", "source_run_id": "20260917T164637291306Z", "submitted_observation_frames": 828, "workload_id": "walker2d"}, "num_actions": 828, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 117.28877687609, "p99_latency_ms": 119.6550234179746, "return_raw": null, "survival_steps": 828} +{"episode_id": 11, "episode_return": 4264.461568188047, "episode_return_env": 4264.461568188047, "game_score": null, "mean_latency_ms": 111.42724624710532, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/teacher/checkpoints/walker2d_profile_appo_h1_5fps_20260917/checkpoint_p0/checkpoint_000019536_10002432.pth", "config_name": "walker2d_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Walker2dContinuous-v0", "episode_seed": 104782, "frame_ms": 200.0, "gpu_class": "1x-rtx3090", "idle_worker_count": 0, "in_flight_count": 1, "inference_worker_count": 1, "instance_id": "instance_859cf1e47bca6046", "latency_type": "profile_sample", "mode": "simulated", "model_id": "gr00t", "obs_fps": 5.0, "output_dir": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/selection_final", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "walker2d_profile_appo_h1_5fps_20260917", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_profile_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/profile_3090_20260917T164637291306Z/profile.json", "source_run_id": "20260917T164637291306Z", "submitted_observation_frames": 1000, "workload_id": "walker2d"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 117.3355416029389, "p99_latency_ms": 119.41730918430277, "return_raw": null, "survival_steps": 1000} +{"episode_id": 12, "episode_return": 4268.998666838904, "episode_return_env": 4268.998666838904, "game_score": null, "mean_latency_ms": 111.16619680837387, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/teacher/checkpoints/walker2d_profile_appo_h1_5fps_20260917/checkpoint_p0/checkpoint_000019536_10002432.pth", "config_name": "walker2d_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Walker2dContinuous-v0", "episode_seed": 104783, "frame_ms": 200.0, "gpu_class": "1x-rtx3090", "idle_worker_count": 0, "in_flight_count": 1, "inference_worker_count": 1, "instance_id": "instance_859cf1e47bca6046", "latency_type": "profile_sample", "mode": "simulated", "model_id": "gr00t", "obs_fps": 5.0, "output_dir": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/selection_final", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "walker2d_profile_appo_h1_5fps_20260917", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_profile_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/profile_3090_20260917T164637291306Z/profile.json", "source_run_id": "20260917T164637291306Z", "submitted_observation_frames": 1000, "workload_id": "walker2d"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 117.00745232569294, "p99_latency_ms": 119.69535709077327, "return_raw": null, "survival_steps": 1000} +{"episode_id": 13, "episode_return": 4293.114110604841, "episode_return_env": 4293.114110604841, "game_score": null, "mean_latency_ms": 111.36298419765151, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/teacher/checkpoints/walker2d_profile_appo_h1_5fps_20260917/checkpoint_p0/checkpoint_000019536_10002432.pth", "config_name": "walker2d_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Walker2dContinuous-v0", "episode_seed": 104784, "frame_ms": 200.0, "gpu_class": "1x-rtx3090", "idle_worker_count": 0, "in_flight_count": 1, "inference_worker_count": 1, "instance_id": "instance_859cf1e47bca6046", "latency_type": "profile_sample", "mode": "simulated", "model_id": "gr00t", "obs_fps": 5.0, "output_dir": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/selection_final", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "walker2d_profile_appo_h1_5fps_20260917", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_profile_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/profile_3090_20260917T164637291306Z/profile.json", "source_run_id": "20260917T164637291306Z", "submitted_observation_frames": 1000, "workload_id": "walker2d"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 117.35314006167931, "p99_latency_ms": 119.29840408104478, "return_raw": null, "survival_steps": 1000} +{"episode_id": 14, "episode_return": 4274.87183914279, "episode_return_env": 4274.87183914279, "game_score": null, "mean_latency_ms": 111.28983514762457, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/teacher/checkpoints/walker2d_profile_appo_h1_5fps_20260917/checkpoint_p0/checkpoint_000019536_10002432.pth", "config_name": "walker2d_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Walker2dContinuous-v0", "episode_seed": 104785, "frame_ms": 200.0, "gpu_class": "1x-rtx3090", "idle_worker_count": 0, "in_flight_count": 1, "inference_worker_count": 1, "instance_id": "instance_859cf1e47bca6046", "latency_type": "profile_sample", "mode": "simulated", "model_id": "gr00t", "obs_fps": 5.0, "output_dir": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/selection_final", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "walker2d_profile_appo_h1_5fps_20260917", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_profile_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/profile_3090_20260917T164637291306Z/profile.json", "source_run_id": "20260917T164637291306Z", "submitted_observation_frames": 1000, "workload_id": "walker2d"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 117.38174978611063, "p99_latency_ms": 119.67898319498183, "return_raw": null, "survival_steps": 1000} +{"episode_id": 15, "episode_return": 4358.281606522924, "episode_return_env": 4358.281606522924, "game_score": null, "mean_latency_ms": 111.32582834871468, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/teacher/checkpoints/walker2d_profile_appo_h1_5fps_20260917/checkpoint_p0/checkpoint_000019536_10002432.pth", "config_name": "walker2d_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Walker2dContinuous-v0", "episode_seed": 104786, "frame_ms": 200.0, "gpu_class": "1x-rtx3090", "idle_worker_count": 0, "in_flight_count": 1, "inference_worker_count": 1, "instance_id": "instance_859cf1e47bca6046", "latency_type": "profile_sample", "mode": "simulated", "model_id": "gr00t", "obs_fps": 5.0, "output_dir": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/selection_final", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "walker2d_profile_appo_h1_5fps_20260917", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_profile_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/profile_3090_20260917T164637291306Z/profile.json", "source_run_id": "20260917T164637291306Z", "submitted_observation_frames": 1000, "workload_id": "walker2d"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 117.75776583437825, "p99_latency_ms": 119.8977188545467, "return_raw": null, "survival_steps": 1000} +{"episode_id": 16, "episode_return": 4247.960230740107, "episode_return_env": 4247.960230740107, "game_score": null, "mean_latency_ms": 111.22765929385554, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/teacher/checkpoints/walker2d_profile_appo_h1_5fps_20260917/checkpoint_p0/checkpoint_000019536_10002432.pth", "config_name": "walker2d_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Walker2dContinuous-v0", "episode_seed": 104787, "frame_ms": 200.0, "gpu_class": "1x-rtx3090", "idle_worker_count": 0, "in_flight_count": 1, "inference_worker_count": 1, "instance_id": "instance_859cf1e47bca6046", "latency_type": "profile_sample", "mode": "simulated", "model_id": "gr00t", "obs_fps": 5.0, "output_dir": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/selection_final", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "walker2d_profile_appo_h1_5fps_20260917", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_profile_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/profile_3090_20260917T164637291306Z/profile.json", "source_run_id": "20260917T164637291306Z", "submitted_observation_frames": 1000, "workload_id": "walker2d"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 117.21577074582231, "p99_latency_ms": 120.12690715830806, "return_raw": null, "survival_steps": 1000} +{"episode_id": 17, "episode_return": 4363.387419962692, "episode_return_env": 4363.387419962692, "game_score": null, "mean_latency_ms": 111.16506995305689, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/teacher/checkpoints/walker2d_profile_appo_h1_5fps_20260917/checkpoint_p0/checkpoint_000019536_10002432.pth", "config_name": "walker2d_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Walker2dContinuous-v0", "episode_seed": 104788, "frame_ms": 200.0, "gpu_class": "1x-rtx3090", "idle_worker_count": 0, "in_flight_count": 1, "inference_worker_count": 1, "instance_id": "instance_859cf1e47bca6046", "latency_type": "profile_sample", "mode": "simulated", "model_id": "gr00t", "obs_fps": 5.0, "output_dir": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/selection_final", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "walker2d_profile_appo_h1_5fps_20260917", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_profile_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/profile_3090_20260917T164637291306Z/profile.json", "source_run_id": "20260917T164637291306Z", "submitted_observation_frames": 1000, "workload_id": "walker2d"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 116.99913787830096, "p99_latency_ms": 119.5112123304007, "return_raw": null, "survival_steps": 1000} +{"episode_id": 18, "episode_return": 4258.273723341316, "episode_return_env": 4258.273723341316, "game_score": null, "mean_latency_ms": 111.1902288601307, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/teacher/checkpoints/walker2d_profile_appo_h1_5fps_20260917/checkpoint_p0/checkpoint_000019536_10002432.pth", "config_name": "walker2d_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Walker2dContinuous-v0", "episode_seed": 104789, "frame_ms": 200.0, "gpu_class": "1x-rtx3090", "idle_worker_count": 0, "in_flight_count": 1, "inference_worker_count": 1, "instance_id": "instance_859cf1e47bca6046", "latency_type": "profile_sample", "mode": "simulated", "model_id": "gr00t", "obs_fps": 5.0, "output_dir": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/selection_final", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "walker2d_profile_appo_h1_5fps_20260917", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_profile_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/profile_3090_20260917T164637291306Z/profile.json", "source_run_id": "20260917T164637291306Z", "submitted_observation_frames": 1000, "workload_id": "walker2d"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 117.34100575042588, "p99_latency_ms": 119.57563536830767, "return_raw": null, "survival_steps": 1000} +{"episode_id": 19, "episode_return": 4286.0125421364155, "episode_return_env": 4286.0125421364155, "game_score": null, "mean_latency_ms": 111.45914887004557, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/teacher/checkpoints/walker2d_profile_appo_h1_5fps_20260917/checkpoint_p0/checkpoint_000019536_10002432.pth", "config_name": "walker2d_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Walker2dContinuous-v0", "episode_seed": 104790, "frame_ms": 200.0, "gpu_class": "1x-rtx3090", "idle_worker_count": 0, "in_flight_count": 1, "inference_worker_count": 1, "instance_id": "instance_859cf1e47bca6046", "latency_type": "profile_sample", "mode": "simulated", "model_id": "gr00t", "obs_fps": 5.0, "output_dir": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/selection_final", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "walker2d_profile_appo_h1_5fps_20260917", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_profile_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/profile_3090_20260917T164637291306Z/profile.json", "source_run_id": "20260917T164637291306Z", "submitted_observation_frames": 1000, "workload_id": "walker2d"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 117.09138053808518, "p99_latency_ms": 119.74910231216302, "return_raw": null, "survival_steps": 1000} diff --git a/latency-aware/walker2d/small-policy/sample-factory-qwengr00t-rtx3090-v1/validation/selection_training_best/episode_metrics.jsonl b/latency-aware/walker2d/small-policy/sample-factory-qwengr00t-rtx3090-v1/validation/selection_training_best/episode_metrics.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..1939b81bbcc95f57adfcbedfb31cc7c0814c0de9 --- /dev/null +++ b/latency-aware/walker2d/small-policy/sample-factory-qwengr00t-rtx3090-v1/validation/selection_training_best/episode_metrics.jsonl @@ -0,0 +1,20 @@ +{"episode_id": 0, "episode_return": 4232.421844289257, "episode_return_env": 4232.421844289257, "game_score": null, "mean_latency_ms": 111.69516596958135, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/teacher/checkpoints/walker2d_profile_appo_h1_5fps_20260917/checkpoint_p0/best_000014536_7442432_reward_4186.861.pth", "config_name": "walker2d_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Walker2dContinuous-v0", "episode_seed": 104771, "frame_ms": 200.0, "gpu_class": "1x-rtx3090", "idle_worker_count": 0, "in_flight_count": 1, "inference_worker_count": 1, "instance_id": "instance_859cf1e47bca6046", "latency_type": "profile_sample", "mode": "simulated", "model_id": "gr00t", "obs_fps": 5.0, "output_dir": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/selection_training_best", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "walker2d_profile_appo_h1_5fps_20260917", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_profile_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/profile_3090_20260917T164637291306Z/profile.json", "source_run_id": "20260917T164637291306Z", "submitted_observation_frames": 1000, "workload_id": "walker2d"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 117.74894213442445, "p99_latency_ms": 119.81063013225, "return_raw": null, "survival_steps": 1000} +{"episode_id": 1, "episode_return": 4211.834128554937, "episode_return_env": 4211.834128554937, "game_score": null, "mean_latency_ms": 111.30146882528804, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/teacher/checkpoints/walker2d_profile_appo_h1_5fps_20260917/checkpoint_p0/best_000014536_7442432_reward_4186.861.pth", "config_name": "walker2d_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Walker2dContinuous-v0", "episode_seed": 104772, "frame_ms": 200.0, "gpu_class": "1x-rtx3090", "idle_worker_count": 0, "in_flight_count": 1, "inference_worker_count": 1, "instance_id": "instance_859cf1e47bca6046", "latency_type": "profile_sample", "mode": "simulated", "model_id": "gr00t", "obs_fps": 5.0, "output_dir": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/selection_training_best", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "walker2d_profile_appo_h1_5fps_20260917", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_profile_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/profile_3090_20260917T164637291306Z/profile.json", "source_run_id": "20260917T164637291306Z", "submitted_observation_frames": 1000, "workload_id": "walker2d"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 117.56241994943518, "p99_latency_ms": 119.35226772995084, "return_raw": null, "survival_steps": 1000} +{"episode_id": 2, "episode_return": 4235.009681429353, "episode_return_env": 4235.009681429353, "game_score": null, "mean_latency_ms": 111.1356948439972, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/teacher/checkpoints/walker2d_profile_appo_h1_5fps_20260917/checkpoint_p0/best_000014536_7442432_reward_4186.861.pth", "config_name": "walker2d_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Walker2dContinuous-v0", "episode_seed": 104773, "frame_ms": 200.0, "gpu_class": "1x-rtx3090", "idle_worker_count": 0, "in_flight_count": 1, "inference_worker_count": 1, "instance_id": "instance_859cf1e47bca6046", "latency_type": "profile_sample", "mode": "simulated", "model_id": "gr00t", "obs_fps": 5.0, "output_dir": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/selection_training_best", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "walker2d_profile_appo_h1_5fps_20260917", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_profile_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/profile_3090_20260917T164637291306Z/profile.json", "source_run_id": "20260917T164637291306Z", "submitted_observation_frames": 1000, "workload_id": "walker2d"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 117.05222681816122, "p99_latency_ms": 119.47251970056475, "return_raw": null, "survival_steps": 1000} +{"episode_id": 3, "episode_return": 4216.192126194387, "episode_return_env": 4216.192126194387, "game_score": null, "mean_latency_ms": 111.39095322995061, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/teacher/checkpoints/walker2d_profile_appo_h1_5fps_20260917/checkpoint_p0/best_000014536_7442432_reward_4186.861.pth", "config_name": "walker2d_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Walker2dContinuous-v0", "episode_seed": 104774, "frame_ms": 200.0, "gpu_class": "1x-rtx3090", "idle_worker_count": 0, "in_flight_count": 1, "inference_worker_count": 1, "instance_id": "instance_859cf1e47bca6046", "latency_type": "profile_sample", "mode": "simulated", "model_id": "gr00t", "obs_fps": 5.0, "output_dir": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/selection_training_best", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "walker2d_profile_appo_h1_5fps_20260917", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_profile_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/profile_3090_20260917T164637291306Z/profile.json", "source_run_id": "20260917T164637291306Z", "submitted_observation_frames": 1000, "workload_id": "walker2d"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 117.0485143536838, "p99_latency_ms": 119.56701831294635, "return_raw": null, "survival_steps": 1000} +{"episode_id": 4, "episode_return": 4225.968774742952, "episode_return_env": 4225.968774742952, "game_score": null, "mean_latency_ms": 111.4434478616093, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/teacher/checkpoints/walker2d_profile_appo_h1_5fps_20260917/checkpoint_p0/best_000014536_7442432_reward_4186.861.pth", "config_name": "walker2d_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Walker2dContinuous-v0", "episode_seed": 104775, "frame_ms": 200.0, "gpu_class": "1x-rtx3090", "idle_worker_count": 0, "in_flight_count": 1, "inference_worker_count": 1, "instance_id": "instance_859cf1e47bca6046", "latency_type": "profile_sample", "mode": "simulated", "model_id": "gr00t", "obs_fps": 5.0, "output_dir": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/selection_training_best", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "walker2d_profile_appo_h1_5fps_20260917", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_profile_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/profile_3090_20260917T164637291306Z/profile.json", "source_run_id": "20260917T164637291306Z", "submitted_observation_frames": 1000, "workload_id": "walker2d"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 117.03228350329478, "p99_latency_ms": 119.5999014787056, "return_raw": null, "survival_steps": 1000} +{"episode_id": 5, "episode_return": 4246.130242600228, "episode_return_env": 4246.130242600228, "game_score": null, "mean_latency_ms": 111.27118913304983, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/teacher/checkpoints/walker2d_profile_appo_h1_5fps_20260917/checkpoint_p0/best_000014536_7442432_reward_4186.861.pth", "config_name": "walker2d_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Walker2dContinuous-v0", "episode_seed": 104776, "frame_ms": 200.0, "gpu_class": "1x-rtx3090", "idle_worker_count": 0, "in_flight_count": 1, "inference_worker_count": 1, "instance_id": "instance_859cf1e47bca6046", "latency_type": "profile_sample", "mode": "simulated", "model_id": "gr00t", "obs_fps": 5.0, "output_dir": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/selection_training_best", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "walker2d_profile_appo_h1_5fps_20260917", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_profile_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/profile_3090_20260917T164637291306Z/profile.json", "source_run_id": "20260917T164637291306Z", "submitted_observation_frames": 1000, "workload_id": "walker2d"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 116.84997029595598, "p99_latency_ms": 119.51571891712796, "return_raw": null, "survival_steps": 1000} +{"episode_id": 6, "episode_return": 4259.369933944325, "episode_return_env": 4259.369933944325, "game_score": null, "mean_latency_ms": 111.35248938694566, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/teacher/checkpoints/walker2d_profile_appo_h1_5fps_20260917/checkpoint_p0/best_000014536_7442432_reward_4186.861.pth", "config_name": "walker2d_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Walker2dContinuous-v0", "episode_seed": 104777, "frame_ms": 200.0, "gpu_class": "1x-rtx3090", "idle_worker_count": 0, "in_flight_count": 1, "inference_worker_count": 1, "instance_id": "instance_859cf1e47bca6046", "latency_type": "profile_sample", "mode": "simulated", "model_id": "gr00t", "obs_fps": 5.0, "output_dir": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/selection_training_best", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "walker2d_profile_appo_h1_5fps_20260917", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_profile_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/profile_3090_20260917T164637291306Z/profile.json", "source_run_id": "20260917T164637291306Z", "submitted_observation_frames": 1000, "workload_id": "walker2d"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 117.24712708483169, "p99_latency_ms": 119.48416732772776, "return_raw": null, "survival_steps": 1000} +{"episode_id": 7, "episode_return": 4211.69597436189, "episode_return_env": 4211.69597436189, "game_score": null, "mean_latency_ms": 111.32393599620043, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/teacher/checkpoints/walker2d_profile_appo_h1_5fps_20260917/checkpoint_p0/best_000014536_7442432_reward_4186.861.pth", "config_name": "walker2d_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Walker2dContinuous-v0", "episode_seed": 104778, "frame_ms": 200.0, "gpu_class": "1x-rtx3090", "idle_worker_count": 0, "in_flight_count": 1, "inference_worker_count": 1, "instance_id": "instance_859cf1e47bca6046", "latency_type": "profile_sample", "mode": "simulated", "model_id": "gr00t", "obs_fps": 5.0, "output_dir": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/selection_training_best", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "walker2d_profile_appo_h1_5fps_20260917", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_profile_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/profile_3090_20260917T164637291306Z/profile.json", "source_run_id": "20260917T164637291306Z", "submitted_observation_frames": 1000, "workload_id": "walker2d"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 117.6012124114765, "p99_latency_ms": 119.98007001875979, "return_raw": null, "survival_steps": 1000} +{"episode_id": 8, "episode_return": 4227.7128439529915, "episode_return_env": 4227.7128439529915, "game_score": null, "mean_latency_ms": 111.39142950446052, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/teacher/checkpoints/walker2d_profile_appo_h1_5fps_20260917/checkpoint_p0/best_000014536_7442432_reward_4186.861.pth", "config_name": "walker2d_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Walker2dContinuous-v0", "episode_seed": 104779, "frame_ms": 200.0, "gpu_class": "1x-rtx3090", "idle_worker_count": 0, "in_flight_count": 1, "inference_worker_count": 1, "instance_id": "instance_859cf1e47bca6046", "latency_type": "profile_sample", "mode": "simulated", "model_id": "gr00t", "obs_fps": 5.0, "output_dir": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/selection_training_best", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "walker2d_profile_appo_h1_5fps_20260917", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_profile_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/profile_3090_20260917T164637291306Z/profile.json", "source_run_id": "20260917T164637291306Z", "submitted_observation_frames": 1000, "workload_id": "walker2d"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 117.43350990206312, "p99_latency_ms": 119.84356332305674, "return_raw": null, "survival_steps": 1000} +{"episode_id": 9, "episode_return": 4219.107550141977, "episode_return_env": 4219.107550141977, "game_score": null, "mean_latency_ms": 111.14170808097057, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/teacher/checkpoints/walker2d_profile_appo_h1_5fps_20260917/checkpoint_p0/best_000014536_7442432_reward_4186.861.pth", "config_name": "walker2d_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Walker2dContinuous-v0", "episode_seed": 104780, "frame_ms": 200.0, "gpu_class": "1x-rtx3090", "idle_worker_count": 0, "in_flight_count": 1, "inference_worker_count": 1, "instance_id": "instance_859cf1e47bca6046", "latency_type": "profile_sample", "mode": "simulated", "model_id": "gr00t", "obs_fps": 5.0, "output_dir": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/selection_training_best", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "walker2d_profile_appo_h1_5fps_20260917", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_profile_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/profile_3090_20260917T164637291306Z/profile.json", "source_run_id": "20260917T164637291306Z", "submitted_observation_frames": 1000, "workload_id": "walker2d"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 117.19818841002535, "p99_latency_ms": 119.46448607839513, "return_raw": null, "survival_steps": 1000} +{"episode_id": 10, "episode_return": 4176.009289189152, "episode_return_env": 4176.009289189152, "game_score": null, "mean_latency_ms": 111.38809908827963, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/teacher/checkpoints/walker2d_profile_appo_h1_5fps_20260917/checkpoint_p0/best_000014536_7442432_reward_4186.861.pth", "config_name": "walker2d_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Walker2dContinuous-v0", "episode_seed": 104781, "frame_ms": 200.0, "gpu_class": "1x-rtx3090", "idle_worker_count": 0, "in_flight_count": 1, "inference_worker_count": 1, "instance_id": "instance_859cf1e47bca6046", "latency_type": "profile_sample", "mode": "simulated", "model_id": "gr00t", "obs_fps": 5.0, "output_dir": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/selection_training_best", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "walker2d_profile_appo_h1_5fps_20260917", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_profile_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/profile_3090_20260917T164637291306Z/profile.json", "source_run_id": "20260917T164637291306Z", "submitted_observation_frames": 1000, "workload_id": "walker2d"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 117.10050262266324, "p99_latency_ms": 119.68479310513533, "return_raw": null, "survival_steps": 1000} +{"episode_id": 11, "episode_return": 4216.315331519501, "episode_return_env": 4216.315331519501, "game_score": null, "mean_latency_ms": 111.42724624710532, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/teacher/checkpoints/walker2d_profile_appo_h1_5fps_20260917/checkpoint_p0/best_000014536_7442432_reward_4186.861.pth", "config_name": "walker2d_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Walker2dContinuous-v0", "episode_seed": 104782, "frame_ms": 200.0, "gpu_class": "1x-rtx3090", "idle_worker_count": 0, "in_flight_count": 1, "inference_worker_count": 1, "instance_id": "instance_859cf1e47bca6046", "latency_type": "profile_sample", "mode": "simulated", "model_id": "gr00t", "obs_fps": 5.0, "output_dir": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/selection_training_best", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "walker2d_profile_appo_h1_5fps_20260917", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_profile_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/profile_3090_20260917T164637291306Z/profile.json", "source_run_id": "20260917T164637291306Z", "submitted_observation_frames": 1000, "workload_id": "walker2d"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 117.3355416029389, "p99_latency_ms": 119.41730918430277, "return_raw": null, "survival_steps": 1000} +{"episode_id": 12, "episode_return": 4233.440001863116, "episode_return_env": 4233.440001863116, "game_score": null, "mean_latency_ms": 111.16619680837387, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/teacher/checkpoints/walker2d_profile_appo_h1_5fps_20260917/checkpoint_p0/best_000014536_7442432_reward_4186.861.pth", "config_name": "walker2d_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Walker2dContinuous-v0", "episode_seed": 104783, "frame_ms": 200.0, "gpu_class": "1x-rtx3090", "idle_worker_count": 0, "in_flight_count": 1, "inference_worker_count": 1, "instance_id": "instance_859cf1e47bca6046", "latency_type": "profile_sample", "mode": "simulated", "model_id": "gr00t", "obs_fps": 5.0, "output_dir": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/selection_training_best", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "walker2d_profile_appo_h1_5fps_20260917", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_profile_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/profile_3090_20260917T164637291306Z/profile.json", "source_run_id": "20260917T164637291306Z", "submitted_observation_frames": 1000, "workload_id": "walker2d"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 117.00745232569294, "p99_latency_ms": 119.69535709077327, "return_raw": null, "survival_steps": 1000} +{"episode_id": 13, "episode_return": 4210.597842410642, "episode_return_env": 4210.597842410642, "game_score": null, "mean_latency_ms": 111.36298419765151, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/teacher/checkpoints/walker2d_profile_appo_h1_5fps_20260917/checkpoint_p0/best_000014536_7442432_reward_4186.861.pth", "config_name": "walker2d_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Walker2dContinuous-v0", "episode_seed": 104784, "frame_ms": 200.0, "gpu_class": "1x-rtx3090", "idle_worker_count": 0, "in_flight_count": 1, "inference_worker_count": 1, "instance_id": "instance_859cf1e47bca6046", "latency_type": "profile_sample", "mode": "simulated", "model_id": "gr00t", "obs_fps": 5.0, "output_dir": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/selection_training_best", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "walker2d_profile_appo_h1_5fps_20260917", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_profile_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/profile_3090_20260917T164637291306Z/profile.json", "source_run_id": "20260917T164637291306Z", "submitted_observation_frames": 1000, "workload_id": "walker2d"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 117.35314006167931, "p99_latency_ms": 119.29840408104478, "return_raw": null, "survival_steps": 1000} +{"episode_id": 14, "episode_return": 4216.913958972969, "episode_return_env": 4216.913958972969, "game_score": null, "mean_latency_ms": 111.28983514762457, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/teacher/checkpoints/walker2d_profile_appo_h1_5fps_20260917/checkpoint_p0/best_000014536_7442432_reward_4186.861.pth", "config_name": "walker2d_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Walker2dContinuous-v0", "episode_seed": 104785, "frame_ms": 200.0, "gpu_class": "1x-rtx3090", "idle_worker_count": 0, "in_flight_count": 1, "inference_worker_count": 1, "instance_id": "instance_859cf1e47bca6046", "latency_type": "profile_sample", "mode": "simulated", "model_id": "gr00t", "obs_fps": 5.0, "output_dir": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/selection_training_best", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "walker2d_profile_appo_h1_5fps_20260917", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_profile_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/profile_3090_20260917T164637291306Z/profile.json", "source_run_id": "20260917T164637291306Z", "submitted_observation_frames": 1000, "workload_id": "walker2d"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 117.38174978611063, "p99_latency_ms": 119.67898319498183, "return_raw": null, "survival_steps": 1000} +{"episode_id": 15, "episode_return": 4236.856116118128, "episode_return_env": 4236.856116118128, "game_score": null, "mean_latency_ms": 111.32582834871468, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/teacher/checkpoints/walker2d_profile_appo_h1_5fps_20260917/checkpoint_p0/best_000014536_7442432_reward_4186.861.pth", "config_name": "walker2d_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Walker2dContinuous-v0", "episode_seed": 104786, "frame_ms": 200.0, "gpu_class": "1x-rtx3090", "idle_worker_count": 0, "in_flight_count": 1, "inference_worker_count": 1, "instance_id": "instance_859cf1e47bca6046", "latency_type": "profile_sample", "mode": "simulated", "model_id": "gr00t", "obs_fps": 5.0, "output_dir": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/selection_training_best", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "walker2d_profile_appo_h1_5fps_20260917", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_profile_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/profile_3090_20260917T164637291306Z/profile.json", "source_run_id": "20260917T164637291306Z", "submitted_observation_frames": 1000, "workload_id": "walker2d"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 117.75776583437825, "p99_latency_ms": 119.8977188545467, "return_raw": null, "survival_steps": 1000} +{"episode_id": 16, "episode_return": 4241.158612744575, "episode_return_env": 4241.158612744575, "game_score": null, "mean_latency_ms": 111.22765929385554, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/teacher/checkpoints/walker2d_profile_appo_h1_5fps_20260917/checkpoint_p0/best_000014536_7442432_reward_4186.861.pth", "config_name": "walker2d_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Walker2dContinuous-v0", "episode_seed": 104787, "frame_ms": 200.0, "gpu_class": "1x-rtx3090", "idle_worker_count": 0, "in_flight_count": 1, "inference_worker_count": 1, "instance_id": "instance_859cf1e47bca6046", "latency_type": "profile_sample", "mode": "simulated", "model_id": "gr00t", "obs_fps": 5.0, "output_dir": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/selection_training_best", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "walker2d_profile_appo_h1_5fps_20260917", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_profile_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/profile_3090_20260917T164637291306Z/profile.json", "source_run_id": "20260917T164637291306Z", "submitted_observation_frames": 1000, "workload_id": "walker2d"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 117.21577074582231, "p99_latency_ms": 120.12690715830806, "return_raw": null, "survival_steps": 1000} +{"episode_id": 17, "episode_return": 4224.4414913018245, "episode_return_env": 4224.4414913018245, "game_score": null, "mean_latency_ms": 111.16506995305689, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/teacher/checkpoints/walker2d_profile_appo_h1_5fps_20260917/checkpoint_p0/best_000014536_7442432_reward_4186.861.pth", "config_name": "walker2d_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Walker2dContinuous-v0", "episode_seed": 104788, "frame_ms": 200.0, "gpu_class": "1x-rtx3090", "idle_worker_count": 0, "in_flight_count": 1, "inference_worker_count": 1, "instance_id": "instance_859cf1e47bca6046", "latency_type": "profile_sample", "mode": "simulated", "model_id": "gr00t", "obs_fps": 5.0, "output_dir": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/selection_training_best", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "walker2d_profile_appo_h1_5fps_20260917", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_profile_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/profile_3090_20260917T164637291306Z/profile.json", "source_run_id": "20260917T164637291306Z", "submitted_observation_frames": 1000, "workload_id": "walker2d"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 116.99913787830096, "p99_latency_ms": 119.5112123304007, "return_raw": null, "survival_steps": 1000} +{"episode_id": 18, "episode_return": 4191.601768644205, "episode_return_env": 4191.601768644205, "game_score": null, "mean_latency_ms": 111.1902288601307, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/teacher/checkpoints/walker2d_profile_appo_h1_5fps_20260917/checkpoint_p0/best_000014536_7442432_reward_4186.861.pth", "config_name": "walker2d_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Walker2dContinuous-v0", "episode_seed": 104789, "frame_ms": 200.0, "gpu_class": "1x-rtx3090", "idle_worker_count": 0, "in_flight_count": 1, "inference_worker_count": 1, "instance_id": "instance_859cf1e47bca6046", "latency_type": "profile_sample", "mode": "simulated", "model_id": "gr00t", "obs_fps": 5.0, "output_dir": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/selection_training_best", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "walker2d_profile_appo_h1_5fps_20260917", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_profile_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/profile_3090_20260917T164637291306Z/profile.json", "source_run_id": "20260917T164637291306Z", "submitted_observation_frames": 1000, "workload_id": "walker2d"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 117.34100575042588, "p99_latency_ms": 119.57563536830767, "return_raw": null, "survival_steps": 1000} +{"episode_id": 19, "episode_return": 4233.529106106846, "episode_return_env": 4233.529106106846, "game_score": null, "mean_latency_ms": 111.45914887004557, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/teacher/checkpoints/walker2d_profile_appo_h1_5fps_20260917/checkpoint_p0/best_000014536_7442432_reward_4186.861.pth", "config_name": "walker2d_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Walker2dContinuous-v0", "episode_seed": 104790, "frame_ms": 200.0, "gpu_class": "1x-rtx3090", "idle_worker_count": 0, "in_flight_count": 1, "inference_worker_count": 1, "instance_id": "instance_859cf1e47bca6046", "latency_type": "profile_sample", "mode": "simulated", "model_id": "gr00t", "obs_fps": 5.0, "output_dir": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/selection_training_best", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "walker2d_profile_appo_h1_5fps_20260917", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_profile_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/profile_3090_20260917T164637291306Z/profile.json", "source_run_id": "20260917T164637291306Z", "submitted_observation_frames": 1000, "workload_id": "walker2d"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 117.09138053808518, "p99_latency_ms": 119.74910231216302, "return_raw": null, "survival_steps": 1000} diff --git a/latency-aware/walker2d/small-policy/sample-factory-qwengr00t-rtx3090-v1/validation/teacher_probe_audit.json b/latency-aware/walker2d/small-policy/sample-factory-qwengr00t-rtx3090-v1/validation/teacher_probe_audit.json new file mode 100644 index 0000000000000000000000000000000000000000..8d3e3e7bf178f6320b62069dc048437b83a41635 --- /dev/null +++ b/latency-aware/walker2d/small-policy/sample-factory-qwengr00t-rtx3090-v1/validation/teacher_probe_audit.json @@ -0,0 +1,324 @@ +{ + "evaluation50": { + "selection_final": { + "episodes": 20, + "seeds": [ + 104771, + 104772, + 104773, + 104774, + 104775, + 104776, + 104777, + 104778, + 104779, + 104780, + 104781, + 104782, + 104783, + 104784, + 104785, + 104786, + 104787, + 104788, + 104789, + 104790 + ], + "raw_steps": 19603, + "returns": [ + 4299.208001651971, + 4253.381546607346, + 4274.26267070416, + 4281.764485481581, + 4316.5965023405115, + 4185.72463542741, + 4268.629567678589, + 4265.304288469715, + 4331.873574180083, + 3269.8377559505593, + 3583.941712492686, + 4264.461568188047, + 4268.998666838904, + 4293.114110604841, + 4274.87183914279, + 4358.281606522924, + 4247.960230740107, + 4363.387419962692, + 4258.273723341316, + 4286.0125421364155 + ], + "clock_ms": 200, + "drop_invalid": 0 + }, + "selection_training_best": { + "episodes": 20, + "seeds": [ + 104771, + 104772, + 104773, + 104774, + 104775, + 104776, + 104777, + 104778, + 104779, + 104780, + 104781, + 104782, + 104783, + 104784, + 104785, + 104786, + 104787, + 104788, + 104789, + 104790 + ], + "raw_steps": 20000, + "returns": [ + 4232.421844289257, + 4211.834128554937, + 4235.009681429353, + 4216.192126194387, + 4225.968774742952, + 4246.130242600228, + 4259.369933944325, + 4211.69597436189, + 4227.7128439529915, + 4219.107550141977, + 4176.009289189152, + 4216.315331519501, + 4233.440001863116, + 4210.597842410642, + 4216.913958972969, + 4236.856116118128, + 4241.158612744575, + 4224.4414913018245, + 4191.601768644205, + 4233.529106106846 + ], + "clock_ms": 200, + "drop_invalid": 0 + }, + "E10": { + "episodes": 10, + "seeds": [ + 104771, + 104772, + 104773, + 104774, + 104775, + 104776, + 104777, + 104778, + 104779, + 104780 + ], + "raw_steps": 10000, + "returns": [ + 4232.421844289257, + 4211.834128554937, + 4235.009681429353, + 4216.192126194387, + 4225.968774742952, + 4246.130242600228, + 4259.369933944325, + 4211.69597436189, + 4227.7128439529915, + 4219.107550141977 + ], + "clock_ms": 200, + "drop_invalid": 0 + } + }, + "probe": { + "accepted_specs": [ + { + "attempt_idx": 0, + "episode_idx": 0, + "episode_length": 1000, + "episode_raw_frames": 1000, + "episode_raw_return": 4247.287119686604, + "seed": 0, + "split": "train" + }, + { + "attempt_idx": 1, + "episode_idx": 1, + "episode_length": 1000, + "episode_raw_frames": 1000, + "episode_raw_return": 4201.434518292546, + "seed": 1, + "split": "val" + }, + { + "attempt_idx": 2, + "episode_idx": 2, + "episode_length": 1000, + "episode_raw_frames": 1000, + "episode_raw_return": 4227.306919425726, + "seed": 2, + "split": "train" + }, + { + "attempt_idx": 3, + "episode_idx": 3, + "episode_length": 1000, + "episode_raw_frames": 1000, + "episode_raw_return": 4201.307677522302, + "seed": 3, + "split": "val" + }, + { + "attempt_idx": 4, + "episode_idx": 4, + "episode_length": 1000, + "episode_raw_frames": 1000, + "episode_raw_return": 4218.184621155262, + "seed": 4, + "split": "train" + }, + { + "attempt_idx": 5, + "episode_idx": 5, + "episode_length": 1000, + "episode_raw_frames": 1000, + "episode_raw_return": 4216.688547462225, + "seed": 5, + "split": "train" + }, + { + "attempt_idx": 6, + "episode_idx": 6, + "episode_length": 1000, + "episode_raw_frames": 1000, + "episode_raw_return": 4225.2133866250515, + "seed": 6, + "split": "train" + }, + { + "attempt_idx": 7, + "episode_idx": 7, + "episode_length": 1000, + "episode_raw_frames": 1000, + "episode_raw_return": 4231.253190428019, + "seed": 7, + "split": "val" + }, + { + "attempt_idx": 8, + "episode_idx": 8, + "episode_length": 1000, + "episode_raw_frames": 1000, + "episode_raw_return": 4215.274691671133, + "seed": 8, + "split": "train" + }, + { + "attempt_idx": 9, + "episode_idx": 9, + "episode_length": 1000, + "episode_raw_frames": 1000, + "episode_raw_return": 4244.544062376022, + "seed": 9, + "split": "train" + }, + { + "attempt_idx": 10, + "episode_idx": 10, + "episode_length": 1000, + "episode_raw_frames": 1000, + "episode_raw_return": 4246.81555891037, + "seed": 10, + "split": "train" + }, + { + "attempt_idx": 11, + "episode_idx": 11, + "episode_length": 1000, + "episode_raw_frames": 1000, + "episode_raw_return": 4235.734463244677, + "seed": 11, + "split": "train" + } + ], + "attempted_episodes": 12, + "next_attempt_idx": 12, + "rejected_episodes": 0, + "runtime_metadata": { + "checkpoint_train_step": 14536, + "env_fps": 5.0, + "frame_stack": 1, + "obs_fps": 5.0 + } + }, + "selection": { + "selected": "training_best", + "candidates": { + "final": { + "checkpoint": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/teacher/checkpoints/walker2d_profile_appo_h1_5fps_20260917/checkpoint_p0/checkpoint_000019536_10002432.pth", + "sha256": "dcfe291db668f89db2fead0028a76fb5a005965b9daaab7a29bd19eb13c43ca0", + "episodes": 20, + "returns": [ + 4299.208001651971, + 4253.381546607346, + 4274.26267070416, + 4281.764485481581, + 4316.5965023405115, + 4185.72463542741, + 4268.629567678589, + 4265.304288469715, + 4331.873574180083, + 3269.8377559505593, + 3583.941712492686, + 4264.461568188047, + 4268.998666838904, + 4293.114110604841, + 4274.87183914279, + 4358.281606522924, + 4247.960230740107, + 4363.387419962692, + 4258.273723341316, + 4286.0125421364155 + ], + "mean": 4197.294322423132, + "std": 264.3552584808844, + "strict_gt3000": 20 + }, + "training_best": { + "checkpoint": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/teacher/checkpoints/walker2d_profile_appo_h1_5fps_20260917/checkpoint_p0/best_000014536_7442432_reward_4186.861.pth", + "sha256": "a9f0e63f8094fe10f82d2395fc7b59958dc22ac1abc59a6da52772eb1fd944ab", + "episodes": 20, + "returns": [ + 4232.421844289257, + 4211.834128554937, + 4235.009681429353, + 4216.192126194387, + 4225.968774742952, + 4246.130242600228, + 4259.369933944325, + 4211.69597436189, + 4227.7128439529915, + 4219.107550141977, + 4176.009289189152, + 4216.315331519501, + 4233.440001863116, + 4210.597842410642, + 4216.913958972969, + 4236.856116118128, + 4241.158612744575, + 4224.4414913018245, + 4191.601768644205, + 4233.529106106846 + ], + "mean": 4223.315330954163, + "std": 18.175464884559204, + "strict_gt3000": 20 + } + }, + "completed_at": "2026-09-17T17:24:56.075104+00:00" + }, + "recipe_seed": 0, + "profile_sha256": "acf0a997fa7866b5a101714ec1479b8cc19b36ae1262f11081ab828de3e79661", + "note": "E10 reuses first10 selection heldoutseeds" +} \ No newline at end of file diff --git a/latency-aware/walker2d/small-policy/sample-factory-qwenoft-rtx3090-v1/README.md b/latency-aware/walker2d/small-policy/sample-factory-qwenoft-rtx3090-v1/README.md new file mode 100644 index 0000000000000000000000000000000000000000..689d224304d33595258c48e6ca5d548c8cb6ea22 --- /dev/null +++ b/latency-aware/walker2d/small-policy/sample-factory-qwenoft-rtx3090-v1/README.md @@ -0,0 +1,26 @@ +# walker2d / sample-factory-appo + +Training condition: `latency-aware`. Run: `walker2d_profile_20260912T092949Z`. + +Copied from [Standard-Pipeline](https://huggingface.co/latency-sensitive-bench/Standard-Pipeline/tree/2208875f92b21e3f8ec2363a30a6705dfe56cc04/walker2d/profile_latency/small_model) at `2208875f92b21e3f8ec2363a30a6705dfe56cc04`. + +- Checkpoint: `checkpoint.pth` +- Selection: best_reward_after_full_training_budget +- Checkpoint SHA256: `c90dca9e5eb3ef609daec95bf2334e5d8637ab9d4454afffda26826ec22f71b8` +- Configuration, FPS, actions, normalization and latency semantics retain their source values. +- This import does not run evaluations or change training/evaluation/acceptance status. + +Sample Factory APPO inference bundle: `config.json` and the explicit `checkpoint.pth`. +The checkpoint contains `model`, `train_step`, and `env_steps`; every exported model +tensor was reloaded and compared bit-for-bit with the selected source checkpoint. + +## Download + +```python +from huggingface_hub import snapshot_download +prefix = "latency-aware/walker2d/small-policy/sample-factory-qwenoft-rtx3090-v1" +snapshot_download("latency-sensitive-bench/benchmark-models", revision="", + allow_patterns=[prefix + "/**"]) +``` + +Use the verified import commit listed in the repository's import report. diff --git a/latency-aware/walker2d/small-policy/sample-factory-qwenoft-rtx3090-v1/checkpoint.pth b/latency-aware/walker2d/small-policy/sample-factory-qwenoft-rtx3090-v1/checkpoint.pth new file mode 100644 index 0000000000000000000000000000000000000000..e79315651ee29620803378fe7224c4c80ee4e684 --- /dev/null +++ b/latency-aware/walker2d/small-policy/sample-factory-qwenoft-rtx3090-v1/checkpoint.pth @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c90dca9e5eb3ef609daec95bf2334e5d8637ab9d4454afffda26826ec22f71b8 +size 29877 diff --git a/latency-aware/walker2d/small-policy/sample-factory-qwenoft-rtx3090-v1/config.json b/latency-aware/walker2d/small-policy/sample-factory-qwenoft-rtx3090-v1/config.json new file mode 100644 index 0000000000000000000000000000000000000000..a4079893d8974a4a5751d13b0d7f1c6002113e27 --- /dev/null +++ b/latency-aware/walker2d/small-policy/sample-factory-qwenoft-rtx3090-v1/config.json @@ -0,0 +1,276 @@ +{ + "help": false, + "algo": "APPO", + "env": "latency_gymnasium", + "experiment": "walker2d_profile_20260912T092949Z", + "train_dir": "/mnt/results/latency-sensitive-bench/profile-latency-new/walker2d/20260912T092949Z-walker2d/small_models", + "restart_behavior": "overwrite", + "device": "gpu", + "seed": 0, + "initial_model_path": null, + "num_policies": 1, + "async_rl": false, + "serial_mode": false, + "batched_sampling": false, + "num_batches_to_accumulate": 2, + "worker_num_splits": 2, + "policy_workers_per_policy": 1, + "max_policy_lag": 300, + "num_workers": 8, + "num_envs_per_worker": 8, + "batch_size": 1024, + "num_batches_per_epoch": 4, + "num_epochs": 2, + "rollout": 64, + "fasttd3_replay_capacity": 6553600, + "fasttd3_replay_batch_size": 32768, + "fasttd3_transitions_per_update": 64, + "fasttd3_v_min": -250.0, + "fasttd3_v_max": 250.0, + "fasttd3_actor_action_l2": 0.0, + "fasttd3_compile": true, + "fasttd3_sonic_decoder_path": null, + "recurrence": 1, + "shuffle_minibatches": false, + "gamma": 0.99, + "reward_scale": 1.0, + "reward_clip": 1000.0, + "value_bootstrap": true, + "normalize_returns": true, + "exploration_loss_coeff": 0.0, + "value_loss_coeff": 1.3, + "kl_loss_coeff": 0.1, + "exploration_loss": "entropy", + "gae_lambda": 0.95, + "ppo_clip_ratio": 0.2, + "ppo_clip_value": 0.2, + "with_vtrace": false, + "vtrace_rho": 1.0, + "vtrace_c": 1.0, + "optimizer": "adam", + "adam_eps": 1e-06, + "adam_beta1": 0.9, + "adam_beta2": 0.999, + "max_grad_norm": 3.5, + "learning_rate": 0.00295, + "lr_schedule": "linear_decay", + "lr_schedule_kl_threshold": 0.008, + "lr_adaptive_min": 1e-06, + "lr_adaptive_max": 0.01, + "obs_subtract_mean": 0.0, + "obs_scale": 1.0, + "normalize_input": true, + "normalize_input_keys": null, + "decorrelate_experience_max_seconds": 0, + "decorrelate_envs_on_one_worker": true, + "actor_worker_gpus": [], + "set_workers_cpu_affinity": true, + "force_envs_single_thread": false, + "default_niceness": 0, + "log_to_file": true, + "experiment_summaries_interval": 10, + "flush_summaries_interval": 30, + "stats_avg": 100, + "summaries_use_frameskip": true, + "heartbeat_interval": 20, + "heartbeat_reporting_interval": 180, + "train_for_env_steps": 10000000, + "train_for_seconds": 10000000000, + "save_every_sec": 600, + "keep_checkpoints": 5, + "load_checkpoint_kind": "latest", + "save_milestones_sec": -1, + "save_best_every_sec": 5, + "save_best_metric": "reward", + "save_best_after": 100000, + "benchmark": false, + "encoder_mlp_layers": [ + 64, + 64 + ], + "encoder_conv_architecture": "convnet_simple", + "encoder_conv_mlp_layers": [ + 512 + ], + "use_rnn": false, + "rnn_size": 512, + "rnn_type": "gru", + "rnn_num_layers": 1, + "decoder_mlp_layers": [], + "nonlinearity": "tanh", + "policy_initialization": "torch_default", + "policy_init_gain": 1.0, + "actor_critic_share_weights": true, + "adaptive_stddev": false, + "continuous_tanh_scale": 0.0, + "initial_stddev": 1.0, + "use_env_info_cache": false, + "env_gpu_actions": false, + "env_gpu_observations": true, + "env_frameskip": 1, + "env_framestack": 1, + "pixel_format": "CHW", + "use_record_episode_statistics": false, + "with_wandb": false, + "wandb_user": null, + "wandb_project": "sample_factory", + "wandb_group": null, + "wandb_job_type": "SF", + "wandb_tags": [], + "with_pbt": false, + "pbt_mix_policies_in_one_env": true, + "pbt_period_env_steps": 5000000, + "pbt_start_mutation": 20000000, + "pbt_replace_fraction": 0.3, + "pbt_mutation_rate": 0.15, + "pbt_replace_reward_gap": 0.1, + "pbt_replace_reward_gap_absolute": 1e-06, + "pbt_optimize_gamma": false, + "pbt_target_objective": "true_objective", + "pbt_perturb_min": 1.1, + "pbt_perturb_max": 1.5, + "gym_task_name": "walker2d_rgb_state", + "gym_env_id": "LatencyBench/Walker2dContinuous-v0", + "gym_make_kwargs_json": "{\"base_env_id\": \"Walker2d-v4\", \"base_make_kwargs\": {\"ctrl_cost_weight\": 0.001, \"exclude_current_positions_from_observation\": true, \"forward_reward_weight\": 1.0, \"healthy_reward\": 1.0, \"reset_noise_scale\": 0.005, \"terminate_when_unhealthy\": true}, \"render_mode\": \"rgb_array\"}", + "gym_registration_imports_json": "[\"latency_bench.envs.gymnasium_walker2d\"]", + "gym_action_space_json": "{\"dtype\": \"float32\", \"high\": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0], \"labels\": [\"right_thigh_torque\", \"right_leg_torque\", \"right_foot_torque\", \"left_thigh_torque\", \"left_leg_torque\", \"left_foot_torque\"], \"low\": [-1.0, -1.0, -1.0, -1.0, -1.0, -1.0], \"type\": \"box\"}", + "gym_action_labels_json": null, + "gym_action_values_json": null, + "gym_noop_action_id": null, + "gym_noop_action_json": "[0.0, 0.0, 0.0, 0.0, 0.0, 0.0]", + "gym_base_prompt": "Move the Walker2d robot forward while keeping its torso upright. Predict six continuous torques in [-1, 1] ordered as right thigh, right leg, right foot, left thigh, left leg, and left foot.", + "gym_state_labels_json": "[\"torso_height\", \"torso_angle\", \"right_thigh_angle\", \"right_leg_angle\", \"right_foot_angle\", \"left_thigh_angle\", \"left_leg_angle\", \"left_foot_angle\", \"torso_x_velocity\", \"torso_z_velocity\", \"torso_angular_velocity\", \"right_thigh_angular_velocity\", \"right_leg_angular_velocity\", \"right_foot_angular_velocity\", \"left_thigh_angular_velocity\", \"left_leg_angular_velocity\", \"left_foot_angular_velocity\"]", + "env_fps": 10.0, + "obs_fps": 10.0, + "frame_stack": 1, + "export_env_raw_rgb_frames": false, + "mode": "train", + "latency_type": "iid", + "fixed_latency_ms": null, + "mean_latency_ms": null, + "std_latency_ms": null, + "min_latency_ms": null, + "max_latency_ms": null, + "latency_seed": 0, + "latency_profile_path": "/mnt/results/latency-sensitive-bench/profile-latency-new/walker2d/20260912T092949Z-walker2d/profile/profile.json", + "latency_profile_worker_slot": 0, + "simulated_worker_capacity": null, + "add_latency_info": false, + "max_pending_actions": null, + "hold_policy": "hold", + "ordering_policy": "issue_order_fifo", + "eval_episodes": 5, + "eval_parallel_envs": 1, + "eval_latency_raw_frame_values": null, + "eval_max_steps": 1000, + "eval_deterministic": true, + "eval_raw_reward": false, + "episode_metrics_path": "/mnt/results/latency-sensitive-bench/profile-latency-new/walker2d/20260912T092949Z-walker2d/small_models/walker2d_profile_20260912T092949Z/episode_metrics.jsonl", + "command_line": "--mode train --algo APPO --env latency_gymnasium --experiment walker2d_profile_20260912T092949Z --train_dir /mnt/results/latency-sensitive-bench/profile-latency-new/walker2d/20260912T092949Z-walker2d/small_models --restart_behavior overwrite --device gpu --seed 0 --episode_metrics_path /mnt/results/latency-sensitive-bench/profile-latency-new/walker2d/20260912T092949Z-walker2d/small_models/walker2d_profile_20260912T092949Z/episode_metrics.jsonl --train_for_env_steps 10000000 --num_workers 8 --num_envs_per_worker 8 --num_policies 1 --batch_size 1024 --rollout 64 --recurrence 1 --num_epochs 2 --num_batches_per_epoch 4 --worker_num_splits 2 --max_policy_lag 300 --num_batches_to_accumulate 2 --policy_workers_per_policy 1 --learning_rate 0.00295 --kl_loss_coeff 0.1 --lr_schedule_kl_threshold 0.008 --nonlinearity tanh --policy_initialization torch_default --continuous_tanh_scale 0.0 --initial_stddev 1.0 --env_framestack 1 --gamma 0.99 --gae_lambda 0.95 --ppo_clip_ratio 0.2 --ppo_clip_value 0.2 --exploration_loss entropy --exploration_loss_coeff 0.0 --reward_scale 1.0 --reward_clip 1000.0 --value_loss_coeff 1.3 --max_grad_norm 3.5 --optimizer adam --adam_eps 1e-06 --adam_beta1 0.9 --adam_beta2 0.999 --obs_subtract_mean 0.0 --obs_scale 1.0 --decorrelate_experience_max_seconds 0 --default_niceness 0 --rnn_type gru --rnn_size 512 --save_every_sec 600 --keep_checkpoints 5 --save_milestones_sec -1 --save_best_every_sec 5 --save_best_after 100000 --stats_avg 100 --experiment_summaries_interval 10 --async_rl False --batched_sampling False --serial_mode False --use_rnn False --normalize_returns True --normalize_input True --adaptive_stddev False --shuffle_minibatches False --value_bootstrap True --with_vtrace False --decorrelate_envs_on_one_worker True --set_workers_cpu_affinity True --force_envs_single_thread False --actor_critic_share_weights True --with_wandb False --encoder_mlp_layers 64 64 --latency-type iid --latency-seed 0 --latency-profile-path /mnt/results/latency-sensitive-bench/profile-latency-new/walker2d/20260912T092949Z-walker2d/profile/profile.json --latency-profile-worker-slot 0 --add-latency-info False --eval-episodes 5 --eval-parallel-envs 1 --eval-max-steps 1000 --eval-deterministic True --gym-task-name walker2d_rgb_state --gym-env-id LatencyBench/Walker2dContinuous-v0 --gym-make-kwargs-json {\"base_env_id\": \"Walker2d-v4\", \"base_make_kwargs\": {\"ctrl_cost_weight\": 0.001, \"exclude_current_positions_from_observation\": true, \"forward_reward_weight\": 1.0, \"healthy_reward\": 1.0, \"reset_noise_scale\": 0.005, \"terminate_when_unhealthy\": true}, \"render_mode\": \"rgb_array\"} --gym-registration-imports-json [\"latency_bench.envs.gymnasium_walker2d\"] --gym-action-space-json {\"dtype\": \"float32\", \"high\": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0], \"labels\": [\"right_thigh_torque\", \"right_leg_torque\", \"right_foot_torque\", \"left_thigh_torque\", \"left_leg_torque\", \"left_foot_torque\"], \"low\": [-1.0, -1.0, -1.0, -1.0, -1.0, -1.0], \"type\": \"box\"} --gym-noop-action-json [0.0, 0.0, 0.0, 0.0, 0.0, 0.0] --gym-base-prompt Move the Walker2d robot forward while keeping its torso upright. Predict six continuous torques in [-1, 1] ordered as right thigh, right leg, right foot, left thigh, left leg, and left foot. --gym-state-labels-json [\"torso_height\", \"torso_angle\", \"right_thigh_angle\", \"right_leg_angle\", \"right_foot_angle\", \"left_thigh_angle\", \"left_leg_angle\", \"left_foot_angle\", \"torso_x_velocity\", \"torso_z_velocity\", \"torso_angular_velocity\", \"right_thigh_angular_velocity\", \"right_leg_angular_velocity\", \"right_foot_angular_velocity\", \"left_thigh_angular_velocity\", \"left_leg_angular_velocity\", \"left_foot_angular_velocity\"] --env-fps 10.0 --obs-fps 10.0 --frame-stack 1 --hold-policy hold --ordering-policy issue_order_fifo --use_env_info_cache False", + "cli_args": { + "algo": "APPO", + "env": "latency_gymnasium", + "experiment": "walker2d_profile_20260912T092949Z", + "train_dir": "/mnt/results/latency-sensitive-bench/profile-latency-new/walker2d/20260912T092949Z-walker2d/small_models", + "restart_behavior": "overwrite", + "device": "gpu", + "seed": 0, + "num_policies": 1, + "async_rl": false, + "serial_mode": false, + "batched_sampling": false, + "num_batches_to_accumulate": 2, + "worker_num_splits": 2, + "policy_workers_per_policy": 1, + "max_policy_lag": 300, + "num_workers": 8, + "num_envs_per_worker": 8, + "batch_size": 1024, + "num_batches_per_epoch": 4, + "num_epochs": 2, + "rollout": 64, + "recurrence": 1, + "shuffle_minibatches": false, + "gamma": 0.99, + "reward_scale": 1.0, + "reward_clip": 1000.0, + "value_bootstrap": true, + "normalize_returns": true, + "exploration_loss_coeff": 0.0, + "value_loss_coeff": 1.3, + "kl_loss_coeff": 0.1, + "exploration_loss": "entropy", + "gae_lambda": 0.95, + "ppo_clip_ratio": 0.2, + "ppo_clip_value": 0.2, + "with_vtrace": false, + "optimizer": "adam", + "adam_eps": 1e-06, + "adam_beta1": 0.9, + "adam_beta2": 0.999, + "max_grad_norm": 3.5, + "learning_rate": 0.00295, + "lr_schedule_kl_threshold": 0.008, + "obs_subtract_mean": 0.0, + "obs_scale": 1.0, + "normalize_input": true, + "decorrelate_experience_max_seconds": 0, + "decorrelate_envs_on_one_worker": true, + "set_workers_cpu_affinity": true, + "force_envs_single_thread": false, + "default_niceness": 0, + "experiment_summaries_interval": 10, + "stats_avg": 100, + "train_for_env_steps": 10000000, + "save_every_sec": 600, + "keep_checkpoints": 5, + "save_milestones_sec": -1, + "save_best_every_sec": 5, + "save_best_after": 100000, + "encoder_mlp_layers": [ + 64, + 64 + ], + "use_rnn": false, + "rnn_size": 512, + "rnn_type": "gru", + "nonlinearity": "tanh", + "policy_initialization": "torch_default", + "actor_critic_share_weights": true, + "adaptive_stddev": false, + "continuous_tanh_scale": 0.0, + "initial_stddev": 1.0, + "use_env_info_cache": false, + "env_framestack": 1, + "with_wandb": false, + "gym_task_name": "walker2d_rgb_state", + "gym_env_id": "LatencyBench/Walker2dContinuous-v0", + "gym_make_kwargs_json": "{\"base_env_id\": \"Walker2d-v4\", \"base_make_kwargs\": {\"ctrl_cost_weight\": 0.001, \"exclude_current_positions_from_observation\": true, \"forward_reward_weight\": 1.0, \"healthy_reward\": 1.0, \"reset_noise_scale\": 0.005, \"terminate_when_unhealthy\": true}, \"render_mode\": \"rgb_array\"}", + "gym_registration_imports_json": "[\"latency_bench.envs.gymnasium_walker2d\"]", + "gym_action_space_json": "{\"dtype\": \"float32\", \"high\": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0], \"labels\": [\"right_thigh_torque\", \"right_leg_torque\", \"right_foot_torque\", \"left_thigh_torque\", \"left_leg_torque\", \"left_foot_torque\"], \"low\": [-1.0, -1.0, -1.0, -1.0, -1.0, -1.0], \"type\": \"box\"}", + "gym_noop_action_json": "[0.0, 0.0, 0.0, 0.0, 0.0, 0.0]", + "gym_base_prompt": "Move the Walker2d robot forward while keeping its torso upright. Predict six continuous torques in [-1, 1] ordered as right thigh, right leg, right foot, left thigh, left leg, and left foot.", + "gym_state_labels_json": "[\"torso_height\", \"torso_angle\", \"right_thigh_angle\", \"right_leg_angle\", \"right_foot_angle\", \"left_thigh_angle\", \"left_leg_angle\", \"left_foot_angle\", \"torso_x_velocity\", \"torso_z_velocity\", \"torso_angular_velocity\", \"right_thigh_angular_velocity\", \"right_leg_angular_velocity\", \"right_foot_angular_velocity\", \"left_thigh_angular_velocity\", \"left_leg_angular_velocity\", \"left_foot_angular_velocity\"]", + "env_fps": 10.0, + "obs_fps": 10.0, + "frame_stack": 1, + "mode": "train", + "latency_type": "iid", + "latency_seed": 0, + "latency_profile_path": "/mnt/results/latency-sensitive-bench/profile-latency-new/walker2d/20260912T092949Z-walker2d/profile/profile.json", + "latency_profile_worker_slot": 0, + "add_latency_info": false, + "hold_policy": "hold", + "ordering_policy": "issue_order_fifo", + "eval_episodes": 5, + "eval_parallel_envs": 1, + "eval_max_steps": 1000, + "eval_deterministic": true, + "episode_metrics_path": "/mnt/results/latency-sensitive-bench/profile-latency-new/walker2d/20260912T092949Z-walker2d/small_models/walker2d_profile_20260912T092949Z/episode_metrics.jsonl", + "output_dir": "/mnt/results/latency-sensitive-bench/profile-latency-new/walker2d/20260912T092949Z-walker2d/small_train" + }, + "git_hash": "unknown", + "git_repo_name": "not a git repository", + "eval_env_frameskip": 1, + "output_dir": "/mnt/results/latency-sensitive-bench/profile-latency-new/walker2d/20260912T092949Z-walker2d/small_train" +} \ No newline at end of file diff --git a/latency-aware/walker2d/small-policy/sample-factory-qwenoft-rtx3090-v1/profile/DONE b/latency-aware/walker2d/small-policy/sample-factory-qwenoft-rtx3090-v1/profile/DONE new file mode 100644 index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391 diff --git a/latency-aware/walker2d/small-policy/sample-factory-qwenoft-rtx3090-v1/profile/hardware.json b/latency-aware/walker2d/small-policy/sample-factory-qwenoft-rtx3090-v1/profile/hardware.json new file mode 100644 index 0000000000000000000000000000000000000000..e8204e81807044975c0e0a0953d16d6835192f31 --- /dev/null +++ b/latency-aware/walker2d/small-policy/sample-factory-qwenoft-rtx3090-v1/profile/hardware.json @@ -0,0 +1,40 @@ +{ + "driver_version": "580.173.02", + "gpu_class": "1x-rtx3090", + "gpus": [ + { + "name": "NVIDIA GeForce RTX 3090", + "slot": 0 + } + ], + "instance_id": "instance_859cf1e47bca6046", + "topology_links": [], + "torch": { + "backends": { + "cuda_cudnn_sdp_enabled": true, + "cuda_flash_sdp_enabled": true, + "cuda_math_sdp_enabled": true, + "cuda_matmul_allow_tf32": false, + "cuda_mem_efficient_sdp_enabled": true, + "cudnn_allow_tf32": true, + "cudnn_benchmark": false + }, + "cuda_available": true, + "cuda_device_count": 1, + "cuda_version": "12.8", + "current_device": 0, + "current_device_name": "NVIDIA GeForce RTX 3090", + "device_properties": [ + { + "index": 0, + "major": 8, + "minor": 6, + "multi_processor_count": 82, + "name": "NVIDIA GeForce RTX 3090", + "total_memory": 25295257600 + } + ], + "float32_matmul_precision": "highest", + "version": "2.11.0+cu128" + } +} diff --git a/latency-aware/walker2d/small-policy/sample-factory-qwenoft-rtx3090-v1/profile/latency_burst_model.json b/latency-aware/walker2d/small-policy/sample-factory-qwenoft-rtx3090-v1/profile/latency_burst_model.json new file mode 100644 index 0000000000000000000000000000000000000000..e1e6d9259f92d6d826441e270750e7712bf1e60a --- /dev/null +++ b/latency-aware/walker2d/small-policy/sample-factory-qwenoft-rtx3090-v1/profile/latency_burst_model.json @@ -0,0 +1,1322 @@ +{ + "burst_dwell_distribution": { + "distribution_type": "inverse_cdf", + "latency_ms": [ + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0 + ], + "quantile_levels": [ + 0.0, + 0.0001, + 0.0005, + 0.001, + 0.002, + 0.003, + 0.004, + 0.005, + 0.006, + 0.007, + 0.008, + 0.009, + 0.01, + 0.02, + 0.03, + 0.04, + 0.05, + 0.06, + 0.07, + 0.08, + 0.09, + 0.1, + 0.11, + 0.12, + 0.13, + 0.14, + 0.15, + 0.16, + 0.17, + 0.18, + 0.19, + 0.2, + 0.21, + 0.22, + 0.23, + 0.24, + 0.25, + 0.26, + 0.27, + 0.28, + 0.29, + 0.3, + 0.31, + 0.32, + 0.33, + 0.34, + 0.35, + 0.36, + 0.37, + 0.38, + 0.39, + 0.4, + 0.41, + 0.42, + 0.43, + 0.44, + 0.45, + 0.46, + 0.47, + 0.48, + 0.49, + 0.5, + 0.51, + 0.52, + 0.53, + 0.54, + 0.55, + 0.56, + 0.57, + 0.58, + 0.59, + 0.6, + 0.61, + 0.62, + 0.63, + 0.64, + 0.65, + 0.66, + 0.67, + 0.68, + 0.69, + 0.7, + 0.71, + 0.72, + 0.73, + 0.74, + 0.75, + 0.76, + 0.77, + 0.78, + 0.79, + 0.8, + 0.81, + 0.82, + 0.83, + 0.84, + 0.85, + 0.86, + 0.87, + 0.88, + 0.89, + 0.9, + 0.91, + 0.92, + 0.93, + 0.94, + 0.95, + 0.96, + 0.97, + 0.98, + 0.99, + 0.991, + 0.992, + 0.993, + 0.994, + 0.995, + 0.996, + 0.997, + 0.998, + 0.999, + 0.9995, + 0.9999, + 1.0 + ] + }, + "burst_dwell_lengths": [ + 1, + 1, + 1 + ], + "burst_merge_gap_records": 30, + "burst_rank_processes": [ + { + "draw_count": 3, + "dwell_length_spearman_rho": 0.0, + "level_ms": { + "distribution_type": "inverse_cdf", + "latency_ms": [ + 86.64467597007751, + 86.64467597007751, + 86.64467597007751, + 86.64467597007751, + 86.64467597007751, + 86.64467597007751, + 86.64467597007751, + 86.64467597007751, + 86.64467597007751, + 86.64467597007751, + 86.64467597007751, + 86.64467597007751, + 86.64467597007751, + 86.64467597007751, + 86.64467597007751, + 86.64467597007751, + 86.64467597007751, + 86.64467597007751, + 86.64467597007751, + 86.64467597007751, + 86.64467597007751, + 86.64467597007751, + 86.64467597007751, + 86.64467597007751, + 86.64467597007751, + 86.64467597007751, + 86.64467597007751, + 86.64467597007751, + 86.734783680439, + 87.00510681152343, + 87.27542994260789, + 87.54575307369232, + 87.81607620477676, + 88.0863993358612, + 88.35672246694566, + 88.6270455980301, + 88.89736872911453, + 89.16769186019897, + 89.43801499128341, + 89.70833812236786, + 89.9786612534523, + 90.24898438453674, + 90.51930751562118, + 90.78963064670563, + 91.05995377779007, + 91.33027690887451, + 91.60060003995895, + 91.8709231710434, + 92.14124630212784, + 92.41156943321228, + 92.68189256429672, + 92.95221569538117, + 93.22253882646561, + 93.49286195755005, + 93.76318508863449, + 94.03350821971894, + 94.30383135080338, + 94.57415448188782, + 94.84447761297226, + 95.11480074405671, + 95.38512387514115, + 95.65544700622559, + 95.80195389032365, + 95.94846077442169, + 96.09496765851975, + 96.2414745426178, + 96.38798142671585, + 96.5344883108139, + 96.68099519491196, + 96.82750207901, + 96.97400896310806, + 97.12051584720612, + 97.26702273130417, + 97.41352961540223, + 97.56003649950027, + 97.70654338359833, + 97.85305026769637, + 97.99955715179443, + 98.1460640358925, + 98.29257091999054, + 98.4390778040886, + 98.58558468818664, + 98.7320915722847, + 98.87859845638275, + 99.0251053404808, + 99.17161222457885, + 99.31811910867691, + 99.46462599277497, + 99.61113287687301, + 99.75763976097107, + 99.90414664506912, + 100.05065352916718, + 100.19716041326522, + 100.34366729736328, + 100.49017418146133, + 100.53900980949402, + 100.53900980949402, + 100.53900980949402, + 100.53900980949402, + 100.53900980949402, + 100.53900980949402, + 100.53900980949402, + 100.53900980949402, + 100.53900980949402, + 100.53900980949402, + 100.53900980949402, + 100.53900980949402, + 100.53900980949402, + 100.53900980949402, + 100.53900980949402, + 100.53900980949402, + 100.53900980949402, + 100.53900980949402, + 100.53900980949402, + 100.53900980949402, + 100.53900980949402, + 100.53900980949402, + 100.53900980949402, + 100.53900980949402, + 100.53900980949402, + 100.53900980949402, + 100.53900980949402, + 100.53900980949402 + ], + "quantile_levels": [ + 0.0, + 0.0001, + 0.0005, + 0.001, + 0.002, + 0.003, + 0.004, + 0.005, + 0.006, + 0.007, + 0.008, + 0.009, + 0.01, + 0.02, + 0.03, + 0.04, + 0.05, + 0.06, + 0.07, + 0.08, + 0.09, + 0.1, + 0.11, + 0.12, + 0.13, + 0.14, + 0.15, + 0.16, + 0.17, + 0.18, + 0.19, + 0.2, + 0.21, + 0.22, + 0.23, + 0.24, + 0.25, + 0.26, + 0.27, + 0.28, + 0.29, + 0.3, + 0.31, + 0.32, + 0.33, + 0.34, + 0.35, + 0.36, + 0.37, + 0.38, + 0.39, + 0.4, + 0.41, + 0.42, + 0.43, + 0.44, + 0.45, + 0.46, + 0.47, + 0.48, + 0.49, + 0.5, + 0.51, + 0.52, + 0.53, + 0.54, + 0.55, + 0.56, + 0.57, + 0.58, + 0.59, + 0.6, + 0.61, + 0.62, + 0.63, + 0.64, + 0.65, + 0.66, + 0.67, + 0.68, + 0.69, + 0.7, + 0.71, + 0.72, + 0.73, + 0.74, + 0.75, + 0.76, + 0.77, + 0.78, + 0.79, + 0.8, + 0.81, + 0.82, + 0.83, + 0.84, + 0.85, + 0.86, + 0.87, + 0.88, + 0.89, + 0.9, + 0.91, + 0.92, + 0.93, + 0.94, + 0.95, + 0.96, + 0.97, + 0.98, + 0.99, + 0.991, + 0.992, + 0.993, + 0.994, + 0.995, + 0.996, + 0.997, + 0.998, + 0.999, + 0.9995, + 0.9999, + 1.0 + ] + }, + "level_residual_ms": { + "distribution_type": "inverse_cdf", + "latency_ms": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ], + "quantile_levels": [ + 0.0, + 0.0001, + 0.0005, + 0.001, + 0.002, + 0.003, + 0.004, + 0.005, + 0.006, + 0.007, + 0.008, + 0.009, + 0.01, + 0.02, + 0.03, + 0.04, + 0.05, + 0.06, + 0.07, + 0.08, + 0.09, + 0.1, + 0.11, + 0.12, + 0.13, + 0.14, + 0.15, + 0.16, + 0.17, + 0.18, + 0.19, + 0.2, + 0.21, + 0.22, + 0.23, + 0.24, + 0.25, + 0.26, + 0.27, + 0.28, + 0.29, + 0.3, + 0.31, + 0.32, + 0.33, + 0.34, + 0.35, + 0.36, + 0.37, + 0.38, + 0.39, + 0.4, + 0.41, + 0.42, + 0.43, + 0.44, + 0.45, + 0.46, + 0.47, + 0.48, + 0.49, + 0.5, + 0.51, + 0.52, + 0.53, + 0.54, + 0.55, + 0.56, + 0.57, + 0.58, + 0.59, + 0.6, + 0.61, + 0.62, + 0.63, + 0.64, + 0.65, + 0.66, + 0.67, + 0.68, + 0.69, + 0.7, + 0.71, + 0.72, + 0.73, + 0.74, + 0.75, + 0.76, + 0.77, + 0.78, + 0.79, + 0.8, + 0.81, + 0.82, + 0.83, + 0.84, + 0.85, + 0.86, + 0.87, + 0.88, + 0.89, + 0.9, + 0.91, + 0.92, + 0.93, + 0.94, + 0.95, + 0.96, + 0.97, + 0.98, + 0.99, + 0.991, + 0.992, + 0.993, + 0.994, + 0.995, + 0.996, + 0.997, + 0.998, + 0.999, + 0.9995, + 0.9999, + 1.0 + ] + }, + "severity": { + "distribution_type": "inverse_cdf", + "latency_ms": [ + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0 + ], + "quantile_levels": [ + 0.0, + 0.0001, + 0.0005, + 0.001, + 0.002, + 0.003, + 0.004, + 0.005, + 0.006, + 0.007, + 0.008, + 0.009, + 0.01, + 0.02, + 0.03, + 0.04, + 0.05, + 0.06, + 0.07, + 0.08, + 0.09, + 0.1, + 0.11, + 0.12, + 0.13, + 0.14, + 0.15, + 0.16, + 0.17, + 0.18, + 0.19, + 0.2, + 0.21, + 0.22, + 0.23, + 0.24, + 0.25, + 0.26, + 0.27, + 0.28, + 0.29, + 0.3, + 0.31, + 0.32, + 0.33, + 0.34, + 0.35, + 0.36, + 0.37, + 0.38, + 0.39, + 0.4, + 0.41, + 0.42, + 0.43, + 0.44, + 0.45, + 0.46, + 0.47, + 0.48, + 0.49, + 0.5, + 0.51, + 0.52, + 0.53, + 0.54, + 0.55, + 0.56, + 0.57, + 0.58, + 0.59, + 0.6, + 0.61, + 0.62, + 0.63, + 0.64, + 0.65, + 0.66, + 0.67, + 0.68, + 0.69, + 0.7, + 0.71, + 0.72, + 0.73, + 0.74, + 0.75, + 0.76, + 0.77, + 0.78, + 0.79, + 0.8, + 0.81, + 0.82, + 0.83, + 0.84, + 0.85, + 0.86, + 0.87, + 0.88, + 0.89, + 0.9, + 0.91, + 0.92, + 0.93, + 0.94, + 0.95, + 0.96, + 0.97, + 0.98, + 0.99, + 0.991, + 0.992, + 0.993, + 0.994, + 0.995, + 0.996, + 0.997, + 0.998, + 0.999, + 0.9995, + 0.9999, + 1.0 + ] + }, + "spike_count": 3 + } + ], + "burst_slot_rank_templates": [ + { + "dwell_length": 1, + "slot_to_rank": { + "0": 0 + } + }, + { + "dwell_length": 1, + "slot_to_rank": { + "0": 0 + } + }, + { + "dwell_length": 1, + "slot_to_rank": { + "0": 0 + } + } + ], + "model_type": "hidden_regime", + "pre_worker_time_ms": { + "distribution_type": "inverse_cdf", + "latency_ms": [ + 0.2871279716491699, + 0.2871279716491699, + 0.3000966912508011, + 0.3125195586681366, + 0.3222240471839905, + 0.3309027588367462, + 0.33220982551574707, + 0.33615071773529054, + 0.34000066518783567, + 0.3445212626457214, + 0.35144575595855715, + 0.3605480372905731, + 0.36516556739807127, + 0.417451810836792, + 0.5077130913734436, + 0.5416164875030518, + 0.5575604438781738, + 0.5679127693176269, + 0.5778009295463562, + 0.5856747627258301, + 0.5923689246177674, + 0.5970578193664551, + 0.6007380723953247, + 0.6050690412521362, + 0.6085355162620545, + 0.6125645637512207, + 0.6163313388824463, + 0.6203382730484008, + 0.6243826389312744, + 0.6274452209472656, + 0.6306731224060058, + 0.633169412612915, + 0.635418975353241, + 0.6376892566680908, + 0.6397212743759155, + 0.6422977685928345, + 0.6441904902458191, + 0.6462940216064453, + 0.6483324766159058, + 0.6513290643692017, + 0.6537760257720947, + 0.657095193862915, + 0.6599553108215332, + 0.6619826078414917, + 0.6641056776046753, + 0.6661466121673584, + 0.6684087514877319, + 0.6695426940917969, + 0.6714422941207886, + 0.6728639125823974, + 0.6742174506187439, + 0.6758885383605957, + 0.6775458693504334, + 0.6791774749755859, + 0.6806173920631409, + 0.6820759773254395, + 0.6841613054275513, + 0.6858915328979492, + 0.6874943017959595, + 0.6891242265701294, + 0.6925053238868714, + 0.6948971748352051, + 0.6978367328643799, + 0.7002923250198364, + 0.7020790815353394, + 0.7040354251861572, + 0.705716609954834, + 0.707164192199707, + 0.7092588067054748, + 0.7109628677368164, + 0.7128337740898132, + 0.7143014669418335, + 0.716428017616272, + 0.718038272857666, + 0.719979465007782, + 0.722154974937439, + 0.7243303060531616, + 0.7261616230010987, + 0.7283614039421081, + 0.7306269884109498, + 0.7336044073104858, + 0.7368490695953369, + 0.7397679209709167, + 0.7427667856216431, + 0.7457111358642579, + 0.7481389045715332, + 0.751095712184906, + 0.7529307126998901, + 0.7558730483055115, + 0.7581016540527343, + 0.7609253048896789, + 0.7642731666564941, + 0.767168390750885, + 0.7698051929473877, + 0.7736105918884277, + 0.7770744323730469, + 0.7799339890480042, + 0.7838371276855468, + 0.7889614105224609, + 0.7920980215072632, + 0.7968101620674133, + 0.8028519153594971, + 0.8096318721771241, + 0.8171645402908325, + 0.8260717749595643, + 0.8360280990600586, + 0.8494804501533508, + 0.8625362396240234, + 0.8821966767311097, + 0.919889736175537, + 0.976528728008272, + 0.9891926014423378, + 1.0017096090316777, + 1.0116808032989506, + 1.0215084385871889, + 1.0286476433277132, + 1.0351726627349853, + 1.0525703740119934, + 1.082165348529816, + 1.0893350529670716, + 1.1060434979200409, + 1.1279549598693848, + 1.1279549598693848 + ], + "quantile_levels": [ + 0.0, + 0.0001, + 0.0005, + 0.001, + 0.002, + 0.003, + 0.004, + 0.005, + 0.006, + 0.007, + 0.008, + 0.009, + 0.01, + 0.02, + 0.03, + 0.04, + 0.05, + 0.06, + 0.07, + 0.08, + 0.09, + 0.1, + 0.11, + 0.12, + 0.13, + 0.14, + 0.15, + 0.16, + 0.17, + 0.18, + 0.19, + 0.2, + 0.21, + 0.22, + 0.23, + 0.24, + 0.25, + 0.26, + 0.27, + 0.28, + 0.29, + 0.3, + 0.31, + 0.32, + 0.33, + 0.34, + 0.35, + 0.36, + 0.37, + 0.38, + 0.39, + 0.4, + 0.41, + 0.42, + 0.43, + 0.44, + 0.45, + 0.46, + 0.47, + 0.48, + 0.49, + 0.5, + 0.51, + 0.52, + 0.53, + 0.54, + 0.55, + 0.56, + 0.57, + 0.58, + 0.59, + 0.6, + 0.61, + 0.62, + 0.63, + 0.64, + 0.65, + 0.66, + 0.67, + 0.68, + 0.69, + 0.7, + 0.71, + 0.72, + 0.73, + 0.74, + 0.75, + 0.76, + 0.77, + 0.78, + 0.79, + 0.8, + 0.81, + 0.82, + 0.83, + 0.84, + 0.85, + 0.86, + 0.87, + 0.88, + 0.89, + 0.9, + 0.91, + 0.92, + 0.93, + 0.94, + 0.95, + 0.96, + 0.97, + 0.98, + 0.99, + 0.991, + 0.992, + 0.993, + 0.994, + 0.995, + 0.996, + 0.997, + 0.998, + 0.999, + 0.9995, + 0.9999, + 1.0 + ] + }, + "regime_step_counts": { + "burst": 3, + "calm": 3892 + }, + "regime_transition_counts": { + "burst": { + "burst": 0, + "calm": 3 + }, + "calm": { + "burst": 3, + "calm": 3884 + } + }, + "reset_scope": "session", + "schema_version": 12, + "spike_median_multiplier": 1.25, + "spike_threshold_ms_by_worker_slot": { + "0": 86.57148361206055 + }, + "worker_count": 1 +} diff --git a/latency-aware/walker2d/small-policy/sample-factory-qwenoft-rtx3090-v1/profile/latency_distribution.json b/latency-aware/walker2d/small-policy/sample-factory-qwenoft-rtx3090-v1/profile/latency_distribution.json new file mode 100644 index 0000000000000000000000000000000000000000..b38e74471cbc78e0c5b711175a994369bd71e08a --- /dev/null +++ b/latency-aware/walker2d/small-policy/sample-factory-qwenoft-rtx3090-v1/profile/latency_distribution.json @@ -0,0 +1,1027 @@ +{ + "schema_version": 3, + "worker_slots": { + "0": { + "all": { + "count": 3895, + "observation_to_action_latency_ms": { + "distribution_type": "inverse_cdf", + "latency_ms": [ + 64.21888089179993, + 64.21888089179993, + 64.26744388699531, + 64.46230773687363, + 64.66379324674607, + 64.89459874749184, + 65.04965202331543, + 65.14420080184937, + 65.27448462486267, + 65.32612436056137, + 65.44193125247955, + 65.54984476804734, + 65.60522656440735, + 66.6580442905426, + 67.43931007385254, + 68.07540552616119, + 68.45212668180466, + 68.74475541114808, + 68.93804426193238, + 69.10382010936738, + 69.19791586399079, + 69.27610993385315, + 69.38696476221085, + 69.47144668102264, + 69.53326894044876, + 69.58634305000305, + 69.65139049291611, + 69.70361912250519, + 69.74212255477906, + 69.77380666732788, + 69.82166377305984, + 69.85122108459473, + 69.88792041540145, + 69.91167421340943, + 69.94627538919448, + 69.97577981948852, + 70.00295335054398, + 70.03254795074463, + 70.0596283197403, + 70.08194353580475, + 70.10447882413864, + 70.1274299621582, + 70.14408979415893, + 70.16187624931335, + 70.18096616268159, + 70.20358176231385, + 70.22562551498413, + 70.24441020488739, + 70.25932915210724, + 70.28569169044495, + 70.31029176712036, + 70.33195090293884, + 70.35108722448349, + 70.37104334831238, + 70.40294820070267, + 70.42530269622803, + 70.44491225481033, + 70.46375651359558, + 70.47658450603485, + 70.49450085163116, + 70.51512625217438, + 70.53450798988342, + 70.55166764259339, + 70.57121002674103, + 70.58782054185868, + 70.60449481010437, + 70.62178468704224, + 70.64538116455078, + 70.6610111117363, + 70.68199644088745, + 70.70090284347535, + 70.71804141998291, + 70.73997579813003, + 70.76166491508484, + 70.78612376451493, + 70.81402132511138, + 70.84088504314423, + 70.87055168151855, + 70.90500563383102, + 70.93476188182831, + 70.96168086528778, + 70.98515176773071, + 71.0100483417511, + 71.03640780448913, + 71.05892330408096, + 71.08110203742982, + 71.10168570280075, + 71.1350690126419, + 71.16858087778091, + 71.19318833351136, + 71.22100743055344, + 71.25229752063751, + 71.28433256149292, + 71.30739164352417, + 71.33299107551575, + 71.3687974691391, + 71.39627569913864, + 71.435498046875, + 71.47914789915085, + 71.52327086925507, + 71.56817677021027, + 71.63192391395569, + 71.68392897844315, + 71.7507515668869, + 71.84099113941193, + 71.91955304145813, + 72.06988722085953, + 72.29740436077118, + 72.64941691160202, + 73.50347318649293, + 74.43839849233628, + 74.50578736782074, + 74.56680265903474, + 74.60346013188362, + 74.66835722446442, + 74.76408451199532, + 74.85193955421448, + 75.35295803785326, + 75.85356700181961, + 84.02688551187528, + 92.33079797923779, + 101.24016690254211, + 101.24016690254211 + ], + "quantile_levels": [ + 0.0, + 0.0001, + 0.0005, + 0.001, + 0.002, + 0.003, + 0.004, + 0.005, + 0.006, + 0.007, + 0.008, + 0.009, + 0.01, + 0.02, + 0.03, + 0.04, + 0.05, + 0.06, + 0.07, + 0.08, + 0.09, + 0.1, + 0.11, + 0.12, + 0.13, + 0.14, + 0.15, + 0.16, + 0.17, + 0.18, + 0.19, + 0.2, + 0.21, + 0.22, + 0.23, + 0.24, + 0.25, + 0.26, + 0.27, + 0.28, + 0.29, + 0.3, + 0.31, + 0.32, + 0.33, + 0.34, + 0.35, + 0.36, + 0.37, + 0.38, + 0.39, + 0.4, + 0.41, + 0.42, + 0.43, + 0.44, + 0.45, + 0.46, + 0.47, + 0.48, + 0.49, + 0.5, + 0.51, + 0.52, + 0.53, + 0.54, + 0.55, + 0.56, + 0.57, + 0.58, + 0.59, + 0.6, + 0.61, + 0.62, + 0.63, + 0.64, + 0.65, + 0.66, + 0.67, + 0.68, + 0.69, + 0.7, + 0.71, + 0.72, + 0.73, + 0.74, + 0.75, + 0.76, + 0.77, + 0.78, + 0.79, + 0.8, + 0.81, + 0.82, + 0.83, + 0.84, + 0.85, + 0.86, + 0.87, + 0.88, + 0.89, + 0.9, + 0.91, + 0.92, + 0.93, + 0.94, + 0.95, + 0.96, + 0.97, + 0.98, + 0.99, + 0.991, + 0.992, + 0.993, + 0.994, + 0.995, + 0.996, + 0.997, + 0.998, + 0.999, + 0.9995, + 0.9999, + 1.0 + ] + }, + "spearman_rho": 0.9962526864986927, + "worker_service_time_ms": { + "distribution_type": "inverse_cdf", + "latency_ms": [ + 63.82746982574463, + 63.82746982574463, + 63.92493312239647, + 64.06386149644851, + 64.26113409757615, + 64.53079591751099, + 64.67386532783509, + 64.76888129115105, + 64.8894479894638, + 64.96433173656463, + 65.04352538585663, + 65.13859939455986, + 65.21461226940156, + 66.16340847015381, + 66.85050959587097, + 67.493039393425, + 67.849141061306, + 68.12691159248352, + 68.3059456706047, + 68.43858327865601, + 68.54641152620316, + 68.63038969039917, + 68.72071390151977, + 68.80180246829987, + 68.86677287817001, + 68.91922354698181, + 68.97205686569214, + 69.01177797317504, + 69.06502404212952, + 69.10255327224732, + 69.13925127983093, + 69.1760745048523, + 69.20698500871659, + 69.23804593086243, + 69.2644897222519, + 69.29399082660674, + 69.32176500558853, + 69.34696702957153, + 69.37006133794785, + 69.39575870037079, + 69.4202886223793, + 69.44288921356201, + 69.46471593379974, + 69.48130025863648, + 69.5000395655632, + 69.51833429336548, + 69.5362491607666, + 69.5574676990509, + 69.57567809820175, + 69.59754242897034, + 69.62273967266083, + 69.63747501373291, + 69.66297578811646, + 69.68167729377747, + 69.70806875228882, + 69.72895283699036, + 69.7458103299141, + 69.76441006660461, + 69.77901536226273, + 69.79512314796447, + 69.81000846624374, + 69.82767796516418, + 69.84250817298889, + 69.86159727573394, + 69.87757424116134, + 69.89953441619873, + 69.919504404068, + 69.9412291765213, + 69.95989866256714, + 69.97936544418334, + 69.99937173128129, + 70.01988303661346, + 70.03974276781082, + 70.06396021842957, + 70.08396409749984, + 70.1135217666626, + 70.13642358779907, + 70.15888624191284, + 70.18657667636872, + 70.21063661575317, + 70.23414570093155, + 70.2652428150177, + 70.28892891407013, + 70.32105104923248, + 70.33835902214051, + 70.36597990989685, + 70.3919090628624, + 70.41169013977051, + 70.4380200624466, + 70.46836476325988, + 70.49924392700196, + 70.51999247074127, + 70.55594639778137, + 70.5849479675293, + 70.6114976644516, + 70.64248011112213, + 70.67629051208496, + 70.72400040626526, + 70.76106050014496, + 70.80091683864593, + 70.84916614294052, + 70.89379096031189, + 70.94596419334411, + 71.01326558589935, + 71.08510189056396, + 71.18451704978942, + 71.31417214870453, + 71.499387383461, + 71.86696890592575, + 72.52735633850098, + 73.49062955379486, + 73.56633122086525, + 73.62575006008149, + 73.70218905806541, + 73.82927883386613, + 73.89994262456894, + 74.03089104652405, + 74.42580151557922, + 74.97561591863632, + 83.3456894540788, + 91.62312696755146, + 100.53900980949402, + 100.53900980949402 + ], + "quantile_levels": [ + 0.0, + 0.0001, + 0.0005, + 0.001, + 0.002, + 0.003, + 0.004, + 0.005, + 0.006, + 0.007, + 0.008, + 0.009, + 0.01, + 0.02, + 0.03, + 0.04, + 0.05, + 0.06, + 0.07, + 0.08, + 0.09, + 0.1, + 0.11, + 0.12, + 0.13, + 0.14, + 0.15, + 0.16, + 0.17, + 0.18, + 0.19, + 0.2, + 0.21, + 0.22, + 0.23, + 0.24, + 0.25, + 0.26, + 0.27, + 0.28, + 0.29, + 0.3, + 0.31, + 0.32, + 0.33, + 0.34, + 0.35, + 0.36, + 0.37, + 0.38, + 0.39, + 0.4, + 0.41, + 0.42, + 0.43, + 0.44, + 0.45, + 0.46, + 0.47, + 0.48, + 0.49, + 0.5, + 0.51, + 0.52, + 0.53, + 0.54, + 0.55, + 0.56, + 0.57, + 0.58, + 0.59, + 0.6, + 0.61, + 0.62, + 0.63, + 0.64, + 0.65, + 0.66, + 0.67, + 0.68, + 0.69, + 0.7, + 0.71, + 0.72, + 0.73, + 0.74, + 0.75, + 0.76, + 0.77, + 0.78, + 0.79, + 0.8, + 0.81, + 0.82, + 0.83, + 0.84, + 0.85, + 0.86, + 0.87, + 0.88, + 0.89, + 0.9, + 0.91, + 0.92, + 0.93, + 0.94, + 0.95, + 0.96, + 0.97, + 0.98, + 0.99, + 0.991, + 0.992, + 0.993, + 0.994, + 0.995, + 0.996, + 0.997, + 0.998, + 0.999, + 0.9995, + 0.9999, + 1.0 + ] + } + }, + "steady": { + "count": 3892, + "observation_to_action_latency_ms": { + "distribution_type": "inverse_cdf", + "latency_ms": [ + 64.21888089179993, + 64.21888089179993, + 64.26729849910735, + 64.46222882843017, + 64.66291616725921, + 64.89447894287109, + 65.04953873825073, + 65.14368352890014, + 65.27354646110534, + 65.32604718589782, + 65.44180738735199, + 65.54967481040954, + 65.60513632774354, + 66.65711528778075, + 67.435560131073, + 68.07530255794525, + 68.4492192029953, + 68.74420424938202, + 68.93682938575745, + 69.10214092254638, + 69.19769524097443, + 69.27595851421356, + 69.38678356647492, + 69.47117605686188, + 69.5328344297409, + 69.58632252216339, + 69.65088012218476, + 69.70318710803986, + 69.74202722549438, + 69.77315920352936, + 69.82142517089844, + 69.85107939243316, + 69.88772933483123, + 69.9112668466568, + 69.94601192474366, + 69.97563450813294, + 70.00201094150543, + 70.0318727684021, + 70.05896360397338, + 70.08169354915618, + 70.10433425903321, + 70.1270259141922, + 70.14405165672302, + 70.16166516304015, + 70.1808714723587, + 70.20343802928925, + 70.22512376308441, + 70.24407590866089, + 70.25929422855377, + 70.28514323234558, + 70.30990863323211, + 70.33184719085693, + 70.3501536655426, + 70.37075836658478, + 70.40146433830262, + 70.42517745494843, + 70.44485967159271, + 70.46351621627808, + 70.47651810169219, + 70.4935804605484, + 70.51407109737396, + 70.53322803974152, + 70.55151409626006, + 70.5701886844635, + 70.58723092556, + 70.60385684967041, + 70.62133185863495, + 70.6435671710968, + 70.660171251297, + 70.68115920066833, + 70.7000481414795, + 70.71734981536865, + 70.73964700698852, + 70.76134051322937, + 70.78443279743195, + 70.81122475147248, + 70.83893203735352, + 70.86912979602813, + 70.90435886859893, + 70.93248003482819, + 70.96037925243378, + 70.9836455821991, + 71.00937434196472, + 71.0327023601532, + 71.05814930438996, + 71.0794553899765, + 71.10040259361267, + 71.13344577789307, + 71.16721335411071, + 71.19253223896027, + 71.22043813228608, + 71.25034034252167, + 71.28195040225982, + 71.3037918806076, + 71.33070304393769, + 71.36692314624786, + 71.39335422515869, + 71.43432743549347, + 71.4773496055603, + 71.52164331912995, + 71.56488280773164, + 71.6233865261078, + 71.67863788127899, + 71.74672186851501, + 71.83145132541657, + 71.9102623796463, + 72.05838644504547, + 72.25901246070862, + 72.62537901878358, + 73.4486136007309, + 74.34915796756744, + 74.44829755020142, + 74.52339792919159, + 74.5730178489685, + 74.61976903152465, + 74.68462880134582, + 74.80154452991485, + 74.93210744476319, + 75.54560989379881, + 76.3117787475587, + 79.51470895004282, + 86.64668917655945, + 86.64668917655945 + ], + "quantile_levels": [ + 0.0, + 0.0001, + 0.0005, + 0.001, + 0.002, + 0.003, + 0.004, + 0.005, + 0.006, + 0.007, + 0.008, + 0.009, + 0.01, + 0.02, + 0.03, + 0.04, + 0.05, + 0.06, + 0.07, + 0.08, + 0.09, + 0.1, + 0.11, + 0.12, + 0.13, + 0.14, + 0.15, + 0.16, + 0.17, + 0.18, + 0.19, + 0.2, + 0.21, + 0.22, + 0.23, + 0.24, + 0.25, + 0.26, + 0.27, + 0.28, + 0.29, + 0.3, + 0.31, + 0.32, + 0.33, + 0.34, + 0.35, + 0.36, + 0.37, + 0.38, + 0.39, + 0.4, + 0.41, + 0.42, + 0.43, + 0.44, + 0.45, + 0.46, + 0.47, + 0.48, + 0.49, + 0.5, + 0.51, + 0.52, + 0.53, + 0.54, + 0.55, + 0.56, + 0.57, + 0.58, + 0.59, + 0.6, + 0.61, + 0.62, + 0.63, + 0.64, + 0.65, + 0.66, + 0.67, + 0.68, + 0.69, + 0.7, + 0.71, + 0.72, + 0.73, + 0.74, + 0.75, + 0.76, + 0.77, + 0.78, + 0.79, + 0.8, + 0.81, + 0.82, + 0.83, + 0.84, + 0.85, + 0.86, + 0.87, + 0.88, + 0.89, + 0.9, + 0.91, + 0.92, + 0.93, + 0.94, + 0.95, + 0.96, + 0.97, + 0.98, + 0.99, + 0.991, + 0.992, + 0.993, + 0.994, + 0.995, + 0.996, + 0.997, + 0.998, + 0.999, + 0.9995, + 0.9999, + 1.0 + ] + }, + "spearman_rho": 0.9962440143954173, + "worker_service_time_ms": { + "distribution_type": "inverse_cdf", + "latency_ms": [ + 63.82746982574463, + 63.82746982574463, + 63.924849551200865, + 64.06365878295898, + 64.25989377117158, + 64.53052592849731, + 64.67336368179322, + 64.76854853630066, + 64.88874032402039, + 64.9639313735962, + 65.04222708225251, + 65.13830837059021, + 65.21415251731872, + 66.16300260543824, + 66.84763409614563, + 67.49238602161408, + 67.84865987300873, + 68.12611276626586, + 68.30496686458588, + 68.43773768901825, + 68.54610390186309, + 68.6302877664566, + 68.7196520614624, + 68.8012242269516, + 68.8665434885025, + 68.9189061164856, + 68.97146592140197, + 69.01117464065551, + 69.06471957206726, + 69.10236324310303, + 69.13914971351623, + 69.17492520809174, + 69.20655922412872, + 69.23778124332428, + 69.26441912174225, + 69.29358471870422, + 69.32019448280334, + 69.34652063846588, + 69.3698172712326, + 69.39486378192902, + 69.41827408790589, + 69.44232745170594, + 69.46447602272033, + 69.48072097301483, + 69.49991744995117, + 69.51753581523896, + 69.53571593761444, + 69.55461455821991, + 69.57534356594086, + 69.59614387512207, + 69.62158482074737, + 69.63672277927398, + 69.66229291915893, + 69.68103442192077, + 69.70800340175629, + 69.72816224575043, + 69.7450707912445, + 69.76414365768433, + 69.77862896442413, + 69.79452700138091, + 69.80988729000092, + 69.8276025056839, + 69.84054414749146, + 69.86028493881226, + 69.87702286720275, + 69.89840402126312, + 69.9181250333786, + 69.94080069541931, + 69.95955774307251, + 69.97868758678436, + 69.99920504570008, + 70.01864914894104, + 70.03822797775268, + 70.06183682441711, + 70.08342313289643, + 70.11323342323303, + 70.1355171918869, + 70.15791974544526, + 70.18509419441223, + 70.20979231834411, + 70.23125278472901, + 70.26408920288085, + 70.28767484664917, + 70.31905544281005, + 70.33731842041016, + 70.36446030139923, + 70.38798296451569, + 70.41008462905884, + 70.43654942035676, + 70.46712791919708, + 70.49677614212037, + 70.51846179962158, + 70.55446595668792, + 70.58218722820281, + 70.60933103084564, + 70.64008875370025, + 70.67384169101715, + 70.72326898097992, + 70.75816891670227, + 70.79773238182068, + 70.84490663528443, + 70.89026074409485, + 70.94181303977966, + 71.01094079017639, + 71.08044451713562, + 71.17663326740265, + 71.31091842651367, + 71.48311145305634, + 71.83137361526488, + 72.4928621673584, + 73.34201078414917, + 73.49197626876831, + 73.5695899362564, + 73.6369322757721, + 73.70509721183777, + 73.8460913658142, + 73.91988869094848, + 74.07588002777099, + 74.60754434394833, + 75.38980871009836, + 78.84191055488596, + 85.96448826789856, + 85.96448826789856 + ], + "quantile_levels": [ + 0.0, + 0.0001, + 0.0005, + 0.001, + 0.002, + 0.003, + 0.004, + 0.005, + 0.006, + 0.007, + 0.008, + 0.009, + 0.01, + 0.02, + 0.03, + 0.04, + 0.05, + 0.06, + 0.07, + 0.08, + 0.09, + 0.1, + 0.11, + 0.12, + 0.13, + 0.14, + 0.15, + 0.16, + 0.17, + 0.18, + 0.19, + 0.2, + 0.21, + 0.22, + 0.23, + 0.24, + 0.25, + 0.26, + 0.27, + 0.28, + 0.29, + 0.3, + 0.31, + 0.32, + 0.33, + 0.34, + 0.35, + 0.36, + 0.37, + 0.38, + 0.39, + 0.4, + 0.41, + 0.42, + 0.43, + 0.44, + 0.45, + 0.46, + 0.47, + 0.48, + 0.49, + 0.5, + 0.51, + 0.52, + 0.53, + 0.54, + 0.55, + 0.56, + 0.57, + 0.58, + 0.59, + 0.6, + 0.61, + 0.62, + 0.63, + 0.64, + 0.65, + 0.66, + 0.67, + 0.68, + 0.69, + 0.7, + 0.71, + 0.72, + 0.73, + 0.74, + 0.75, + 0.76, + 0.77, + 0.78, + 0.79, + 0.8, + 0.81, + 0.82, + 0.83, + 0.84, + 0.85, + 0.86, + 0.87, + 0.88, + 0.89, + 0.9, + 0.91, + 0.92, + 0.93, + 0.94, + 0.95, + 0.96, + 0.97, + 0.98, + 0.99, + 0.991, + 0.992, + 0.993, + 0.994, + 0.995, + 0.996, + 0.997, + 0.998, + 0.999, + 0.9995, + 0.9999, + 1.0 + ] + } + } + } + } +} diff --git a/latency-aware/walker2d/small-policy/sample-factory-qwenoft-rtx3090-v1/profile/latency_profile.png b/latency-aware/walker2d/small-policy/sample-factory-qwenoft-rtx3090-v1/profile/latency_profile.png new file mode 100644 index 0000000000000000000000000000000000000000..9391681923c8583792d0d92cd799fc0464b0b68c Binary files /dev/null and b/latency-aware/walker2d/small-policy/sample-factory-qwenoft-rtx3090-v1/profile/latency_profile.png differ diff --git a/latency-aware/walker2d/small-policy/sample-factory-qwenoft-rtx3090-v1/profile/profile.json b/latency-aware/walker2d/small-policy/sample-factory-qwenoft-rtx3090-v1/profile/profile.json new file mode 100644 index 0000000000000000000000000000000000000000..4877e60530800330ac998c3f9aa83070a936702d --- /dev/null +++ b/latency-aware/walker2d/small-policy/sample-factory-qwenoft-rtx3090-v1/profile/profile.json @@ -0,0 +1,66 @@ +{ + "burst_model_path": "latency_burst_model.json", + "distribution_path": "latency_distribution.json", + "env_fps": 10, + "frame_ms": 100.0, + "gpu_class": "1x-rtx3090", + "instance_id": "instance_859cf1e47bca6046", + "latency_kind": "observation_to_action_latency", + "latency_method": "temporal", + "model_id": "qwenoft", + "n_admitted_observations": 3895, + "n_capacity_drops": 0, + "n_observation_attempts": 3895, + "per_slot_summary": { + "0": { + "admitted_count": 3895, + "mean_observation_to_action_latency_ms": 70.49593999162413, + "mean_worker_service_time_ms": 69.79960327613645, + "p95_observation_to_action_latency_ms": 72.06934230327606, + "p95_worker_service_time_ms": 71.31323595046997, + "p99_worker_service_time_ms": 73.49043211936952 + } + }, + "provenance": { + "base_config": "/workspace/tasks/20260912T092949Z-walker2d/profile.yaml", + "checkpoint_kind": "best", + "model_artifact": { + "checkpoint": "checkpoints/steps_5000_pytorch_model.pt", + "model_config": "config.full.yaml", + "path_in_repo": "OpenVLA/zero-latency/walker2d_rgb_state_l0_return_gt3000_100ep_openvla_native_continuous_projector_sft_5k", + "repo_id": "latency-sensitive-bench/extra-envs-checkpoints", + "source": "local" + }, + "session_ids": [ + 0, + 1, + 2, + 3, + 4 + ] + }, + "sample_model_type": "hidden_regime", + "source_run_id": "20260912T093545326611Z", + "summary": { + "frame_ms": 100.0, + "max_ms": 101.24016690254211, + "mean_effective_frames": 0.7049593999162412, + "mean_ms": 70.49593999162413, + "min_ms": 64.21888089179993, + "n_samples": 3895, + "p50_frames": 0.7053450798988342, + "p50_ms": 70.53450798988342, + "p90_frames": 0.7163049759864808, + "p90_ms": 71.63049759864808, + "p95_frames": 0.7206934230327606, + "p95_ms": 72.06934230327606, + "p99_frames": 0.7443600579738617, + "p99_ms": 74.43600579738617, + "prob_latency_gt_1_frame": 0.00025673940949935817, + "prob_latency_gt_2_frames": 0.0, + "prob_latency_gt_3_frames": 0.0, + "std_ms": 1.483596848515007 + }, + "visualization_path": "latency_profile.png", + "workload_id": "walker2d" +} diff --git a/latency-aware/walker2d/small-policy/sample-factory-qwenoft-rtx3090-v1/provenance.json b/latency-aware/walker2d/small-policy/sample-factory-qwenoft-rtx3090-v1/provenance.json new file mode 100644 index 0000000000000000000000000000000000000000..44d252c938d887887f5e658f8cd721d18b634d97 --- /dev/null +++ b/latency-aware/walker2d/small-policy/sample-factory-qwenoft-rtx3090-v1/provenance.json @@ -0,0 +1,44 @@ +{ + "task": "walker2d", + "model": "sample-factory-appo", + "training_condition": "latency-aware", + "training_run_id": "walker2d_profile_20260912T092949Z", + "source": { + "repo_id": "latency-sensitive-bench/Standard-Pipeline", + "revision": "2208875f92b21e3f8ec2363a30a6705dfe56cc04", + "path": "walker2d/profile_latency/small_model", + "url": "https://huggingface.co/latency-sensitive-bench/Standard-Pipeline/tree/2208875f92b21e3f8ec2363a30a6705dfe56cc04/walker2d/profile_latency/small_model" + }, + "hf_repo_id": "latency-sensitive-bench/benchmark-models", + "hf_path": "latency-aware/walker2d/small-policy/sample-factory-qwenoft-rtx3090-v1", + "checkpoint": { + "source_file": "walker2d/profile_latency/small_model/checkpoint_p0/best_000013520_6922240_reward_4348.151.pth", + "source_sha256": "32607b821cdecb0fcfc15867ea073ea2719706644ab59fd5b135774a0de7dbb4", + "source_bytes": 83973, + "file": "checkpoint.pth", + "selection_rule": "best_reward_after_full_training_budget", + "method": "inference_export", + "sha256": "c90dca9e5eb3ef609daec95bf2334e5d8637ab9d4454afffda26826ec22f71b8", + "bytes": 29877, + "train_step": 13520, + "env_steps": 6922240, + "tensor_count": 15, + "tensor_equality_verified": true, + "removed_fields": [ + "best_performance", + "curr_lr", + "optimizer" + ] + }, + "config_source": "walker2d/profile_latency/small_model/config.json", + "task_contract_file": null, + "missing_source_files": [], + "migration": { + "operation": "copy_selected_inference_bundle", + "source_modified": false, + "evaluation_performed": false, + "experiment_status": "preserved_in_source_evidence" + }, + "algorithm": "APPO", + "latency_model": "qwenoft" +} diff --git a/latency-aware/walker2d/small-policy/sample-factory-qwenoft-rtx3090-v1/rollout_descriptor.json b/latency-aware/walker2d/small-policy/sample-factory-qwenoft-rtx3090-v1/rollout_descriptor.json new file mode 100644 index 0000000000000000000000000000000000000000..f63de27b13073719a463e38812d2fbd1d9ea7464 --- /dev/null +++ b/latency-aware/walker2d/small-policy/sample-factory-qwenoft-rtx3090-v1/rollout_descriptor.json @@ -0,0 +1,310 @@ +{ + "env_name": "walker2d_rgb_state", + "integration_name": "gymnasium", + "action_spec": { + "layout": "gymnasium_continuous_v1", + "labels": [ + "right_thigh_torque", + "right_leg_torque", + "right_foot_torque", + "left_thigh_torque", + "left_leg_torque", + "left_foot_torque" + ], + "row_fields": [ + "action", + "action_text" + ], + "reward_field": "raw_reward", + "episode_return_field": "episode_raw_return", + "reward_semantics": "raw_reward" + }, + "base_prompt": "Move the Walker2d robot forward while keeping its torso upright. Predict six continuous torques in [-1, 1] ordered as right thigh, right leg, right foot, left thigh, left leg, and left foot.", + "checkpoint_experiment_dir": "/mnt/results/latency-sensitive-bench/profile-latency-new/walker2d/20260912T092949Z-walker2d/small_models/walker2d_profile_20260912T092949Z", + "checkpoint_path": "/mnt/results/latency-sensitive-bench/profile-latency-new/walker2d/20260912T092949Z-walker2d/small_models/walker2d_profile_20260912T092949Z/checkpoint_p0/best_000013520_6922240_reward_4348.151.pth", + "flat_cfg": { + "help": false, + "algo": "APPO", + "env": "latency_gymnasium", + "experiment": "walker2d_profile_20260912T092949Z", + "train_dir": "/mnt/results/latency-sensitive-bench/profile-latency-new/walker2d/20260912T092949Z-walker2d/small_models", + "restart_behavior": "overwrite", + "device": "gpu", + "seed": 0, + "initial_model_path": null, + "num_policies": 1, + "async_rl": false, + "serial_mode": false, + "batched_sampling": false, + "num_batches_to_accumulate": 2, + "worker_num_splits": 2, + "policy_workers_per_policy": 1, + "max_policy_lag": 300, + "num_workers": 8, + "num_envs_per_worker": 8, + "batch_size": 1024, + "num_batches_per_epoch": 4, + "num_epochs": 2, + "rollout": 64, + "fasttd3_replay_capacity": 6553600, + "fasttd3_replay_batch_size": 32768, + "fasttd3_transitions_per_update": 64, + "fasttd3_v_min": -250.0, + "fasttd3_v_max": 250.0, + "fasttd3_actor_action_l2": 0.0, + "fasttd3_compile": true, + "fasttd3_sonic_decoder_path": null, + "recurrence": 1, + "shuffle_minibatches": false, + "gamma": 0.99, + "reward_scale": 1.0, + "reward_clip": 1000.0, + "value_bootstrap": true, + "normalize_returns": true, + "exploration_loss_coeff": 0.0, + "value_loss_coeff": 1.3, + "kl_loss_coeff": 0.1, + "exploration_loss": "entropy", + "gae_lambda": 0.95, + "ppo_clip_ratio": 0.2, + "ppo_clip_value": 0.2, + "with_vtrace": false, + "vtrace_rho": 1.0, + "vtrace_c": 1.0, + "optimizer": "adam", + "adam_eps": 1e-06, + "adam_beta1": 0.9, + "adam_beta2": 0.999, + "max_grad_norm": 3.5, + "learning_rate": 0.00295, + "lr_schedule": "linear_decay", + "lr_schedule_kl_threshold": 0.008, + "lr_adaptive_min": 1e-06, + "lr_adaptive_max": 0.01, + "obs_subtract_mean": 0.0, + "obs_scale": 1.0, + "normalize_input": true, + "normalize_input_keys": null, + "decorrelate_experience_max_seconds": 0, + "decorrelate_envs_on_one_worker": true, + "actor_worker_gpus": [], + "set_workers_cpu_affinity": true, + "force_envs_single_thread": false, + "default_niceness": 0, + "log_to_file": true, + "experiment_summaries_interval": 10, + "flush_summaries_interval": 30, + "stats_avg": 100, + "summaries_use_frameskip": true, + "heartbeat_interval": 20, + "heartbeat_reporting_interval": 180, + "train_for_env_steps": 10000000, + "train_for_seconds": 10000000000, + "save_every_sec": 600, + "keep_checkpoints": 5, + "load_checkpoint_kind": "latest", + "save_milestones_sec": -1, + "save_best_every_sec": 5, + "save_best_metric": "reward", + "save_best_after": 100000, + "benchmark": false, + "encoder_mlp_layers": [ + 64, + 64 + ], + "encoder_conv_architecture": "convnet_simple", + "encoder_conv_mlp_layers": [ + 512 + ], + "use_rnn": false, + "rnn_size": 512, + "rnn_type": "gru", + "rnn_num_layers": 1, + "decoder_mlp_layers": [], + "nonlinearity": "tanh", + "policy_initialization": "torch_default", + "policy_init_gain": 1.0, + "actor_critic_share_weights": true, + "adaptive_stddev": false, + "continuous_tanh_scale": 0.0, + "initial_stddev": 1.0, + "use_env_info_cache": false, + "env_gpu_actions": false, + "env_gpu_observations": true, + "env_frameskip": 1, + "env_framestack": 1, + "pixel_format": "CHW", + "use_record_episode_statistics": false, + "with_wandb": false, + "wandb_user": null, + "wandb_project": "sample_factory", + "wandb_group": null, + "wandb_job_type": "SF", + "wandb_tags": [], + "with_pbt": false, + "pbt_mix_policies_in_one_env": true, + "pbt_period_env_steps": 5000000, + "pbt_start_mutation": 20000000, + "pbt_replace_fraction": 0.3, + "pbt_mutation_rate": 0.15, + "pbt_replace_reward_gap": 0.1, + "pbt_replace_reward_gap_absolute": 1e-06, + "pbt_optimize_gamma": false, + "pbt_target_objective": "true_objective", + "pbt_perturb_min": 1.1, + "pbt_perturb_max": 1.5, + "gym_task_name": "walker2d_rgb_state", + "gym_env_id": "LatencyBench/Walker2dContinuous-v0", + "gym_make_kwargs_json": "{\"base_env_id\": \"Walker2d-v4\", \"base_make_kwargs\": {\"ctrl_cost_weight\": 0.001, \"exclude_current_positions_from_observation\": true, \"forward_reward_weight\": 1.0, \"healthy_reward\": 1.0, \"reset_noise_scale\": 0.005, \"terminate_when_unhealthy\": true}, \"render_mode\": \"rgb_array\"}", + "gym_registration_imports_json": "[\"latency_bench.envs.gymnasium_walker2d\"]", + "gym_action_space_json": "{\"dtype\": \"float32\", \"high\": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0], \"labels\": [\"right_thigh_torque\", \"right_leg_torque\", \"right_foot_torque\", \"left_thigh_torque\", \"left_leg_torque\", \"left_foot_torque\"], \"low\": [-1.0, -1.0, -1.0, -1.0, -1.0, -1.0], \"type\": \"box\"}", + "gym_action_labels_json": null, + "gym_action_values_json": null, + "gym_noop_action_id": null, + "gym_noop_action_json": "[0.0, 0.0, 0.0, 0.0, 0.0, 0.0]", + "gym_base_prompt": "Move the Walker2d robot forward while keeping its torso upright. Predict six continuous torques in [-1, 1] ordered as right thigh, right leg, right foot, left thigh, left leg, and left foot.", + "gym_state_labels_json": "[\"torso_height\", \"torso_angle\", \"right_thigh_angle\", \"right_leg_angle\", \"right_foot_angle\", \"left_thigh_angle\", \"left_leg_angle\", \"left_foot_angle\", \"torso_x_velocity\", \"torso_z_velocity\", \"torso_angular_velocity\", \"right_thigh_angular_velocity\", \"right_leg_angular_velocity\", \"right_foot_angular_velocity\", \"left_thigh_angular_velocity\", \"left_leg_angular_velocity\", \"left_foot_angular_velocity\"]", + "env_fps": 10.0, + "obs_fps": 10.0, + "frame_stack": 1, + "export_env_raw_rgb_frames": false, + "mode": "train", + "latency_type": "iid", + "fixed_latency_ms": null, + "mean_latency_ms": null, + "std_latency_ms": null, + "min_latency_ms": null, + "max_latency_ms": null, + "latency_seed": 0, + "latency_profile_path": "/mnt/results/latency-sensitive-bench/profile-latency-new/walker2d/20260912T092949Z-walker2d/profile/profile.json", + "latency_profile_worker_slot": 0, + "simulated_worker_capacity": null, + "add_latency_info": false, + "max_pending_actions": null, + "hold_policy": "hold", + "ordering_policy": "issue_order_fifo", + "eval_episodes": 5, + "eval_parallel_envs": 1, + "eval_latency_raw_frame_values": null, + "eval_max_steps": 1000, + "eval_deterministic": true, + "eval_raw_reward": false, + "episode_metrics_path": "/mnt/results/latency-sensitive-bench/profile-latency-new/walker2d/20260912T092949Z-walker2d/small_models/walker2d_profile_20260912T092949Z/episode_metrics.jsonl", + "command_line": "--mode train --algo APPO --env latency_gymnasium --experiment walker2d_profile_20260912T092949Z --train_dir /mnt/results/latency-sensitive-bench/profile-latency-new/walker2d/20260912T092949Z-walker2d/small_models --restart_behavior overwrite --device gpu --seed 0 --episode_metrics_path /mnt/results/latency-sensitive-bench/profile-latency-new/walker2d/20260912T092949Z-walker2d/small_models/walker2d_profile_20260912T092949Z/episode_metrics.jsonl --train_for_env_steps 10000000 --num_workers 8 --num_envs_per_worker 8 --num_policies 1 --batch_size 1024 --rollout 64 --recurrence 1 --num_epochs 2 --num_batches_per_epoch 4 --worker_num_splits 2 --max_policy_lag 300 --num_batches_to_accumulate 2 --policy_workers_per_policy 1 --learning_rate 0.00295 --kl_loss_coeff 0.1 --lr_schedule_kl_threshold 0.008 --nonlinearity tanh --policy_initialization torch_default --continuous_tanh_scale 0.0 --initial_stddev 1.0 --env_framestack 1 --gamma 0.99 --gae_lambda 0.95 --ppo_clip_ratio 0.2 --ppo_clip_value 0.2 --exploration_loss entropy --exploration_loss_coeff 0.0 --reward_scale 1.0 --reward_clip 1000.0 --value_loss_coeff 1.3 --max_grad_norm 3.5 --optimizer adam --adam_eps 1e-06 --adam_beta1 0.9 --adam_beta2 0.999 --obs_subtract_mean 0.0 --obs_scale 1.0 --decorrelate_experience_max_seconds 0 --default_niceness 0 --rnn_type gru --rnn_size 512 --save_every_sec 600 --keep_checkpoints 5 --save_milestones_sec -1 --save_best_every_sec 5 --save_best_after 100000 --stats_avg 100 --experiment_summaries_interval 10 --async_rl False --batched_sampling False --serial_mode False --use_rnn False --normalize_returns True --normalize_input True --adaptive_stddev False --shuffle_minibatches False --value_bootstrap True --with_vtrace False --decorrelate_envs_on_one_worker True --set_workers_cpu_affinity True --force_envs_single_thread False --actor_critic_share_weights True --with_wandb False --encoder_mlp_layers 64 64 --latency-type iid --latency-seed 0 --latency-profile-path /mnt/results/latency-sensitive-bench/profile-latency-new/walker2d/20260912T092949Z-walker2d/profile/profile.json --latency-profile-worker-slot 0 --add-latency-info False --eval-episodes 5 --eval-parallel-envs 1 --eval-max-steps 1000 --eval-deterministic True --gym-task-name walker2d_rgb_state --gym-env-id LatencyBench/Walker2dContinuous-v0 --gym-make-kwargs-json {\"base_env_id\": \"Walker2d-v4\", \"base_make_kwargs\": {\"ctrl_cost_weight\": 0.001, \"exclude_current_positions_from_observation\": true, \"forward_reward_weight\": 1.0, \"healthy_reward\": 1.0, \"reset_noise_scale\": 0.005, \"terminate_when_unhealthy\": true}, \"render_mode\": \"rgb_array\"} --gym-registration-imports-json [\"latency_bench.envs.gymnasium_walker2d\"] --gym-action-space-json {\"dtype\": \"float32\", \"high\": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0], \"labels\": [\"right_thigh_torque\", \"right_leg_torque\", \"right_foot_torque\", \"left_thigh_torque\", \"left_leg_torque\", \"left_foot_torque\"], \"low\": [-1.0, -1.0, -1.0, -1.0, -1.0, -1.0], \"type\": \"box\"} --gym-noop-action-json [0.0, 0.0, 0.0, 0.0, 0.0, 0.0] --gym-base-prompt Move the Walker2d robot forward while keeping its torso upright. Predict six continuous torques in [-1, 1] ordered as right thigh, right leg, right foot, left thigh, left leg, and left foot. --gym-state-labels-json [\"torso_height\", \"torso_angle\", \"right_thigh_angle\", \"right_leg_angle\", \"right_foot_angle\", \"left_thigh_angle\", \"left_leg_angle\", \"left_foot_angle\", \"torso_x_velocity\", \"torso_z_velocity\", \"torso_angular_velocity\", \"right_thigh_angular_velocity\", \"right_leg_angular_velocity\", \"right_foot_angular_velocity\", \"left_thigh_angular_velocity\", \"left_leg_angular_velocity\", \"left_foot_angular_velocity\"] --env-fps 10.0 --obs-fps 10.0 --frame-stack 1 --hold-policy hold --ordering-policy issue_order_fifo --use_env_info_cache False", + "cli_args": { + "algo": "APPO", + "env": "latency_gymnasium", + "experiment": "walker2d_profile_20260912T092949Z", + "train_dir": "/mnt/results/latency-sensitive-bench/profile-latency-new/walker2d/20260912T092949Z-walker2d/small_models", + "restart_behavior": "overwrite", + "device": "gpu", + "seed": 0, + "num_policies": 1, + "async_rl": false, + "serial_mode": false, + "batched_sampling": false, + "num_batches_to_accumulate": 2, + "worker_num_splits": 2, + "policy_workers_per_policy": 1, + "max_policy_lag": 300, + "num_workers": 8, + "num_envs_per_worker": 8, + "batch_size": 1024, + "num_batches_per_epoch": 4, + "num_epochs": 2, + "rollout": 64, + "recurrence": 1, + "shuffle_minibatches": false, + "gamma": 0.99, + "reward_scale": 1.0, + "reward_clip": 1000.0, + "value_bootstrap": true, + "normalize_returns": true, + "exploration_loss_coeff": 0.0, + "value_loss_coeff": 1.3, + "kl_loss_coeff": 0.1, + "exploration_loss": "entropy", + "gae_lambda": 0.95, + "ppo_clip_ratio": 0.2, + "ppo_clip_value": 0.2, + "with_vtrace": false, + "optimizer": "adam", + "adam_eps": 1e-06, + "adam_beta1": 0.9, + "adam_beta2": 0.999, + "max_grad_norm": 3.5, + "learning_rate": 0.00295, + "lr_schedule_kl_threshold": 0.008, + "obs_subtract_mean": 0.0, + "obs_scale": 1.0, + "normalize_input": true, + "decorrelate_experience_max_seconds": 0, + "decorrelate_envs_on_one_worker": true, + "set_workers_cpu_affinity": true, + "force_envs_single_thread": false, + "default_niceness": 0, + "experiment_summaries_interval": 10, + "stats_avg": 100, + "train_for_env_steps": 10000000, + "save_every_sec": 600, + "keep_checkpoints": 5, + "save_milestones_sec": -1, + "save_best_every_sec": 5, + "save_best_after": 100000, + "encoder_mlp_layers": [ + 64, + 64 + ], + "use_rnn": false, + "rnn_size": 512, + "rnn_type": "gru", + "nonlinearity": "tanh", + "policy_initialization": "torch_default", + "actor_critic_share_weights": true, + "adaptive_stddev": false, + "continuous_tanh_scale": 0.0, + "initial_stddev": 1.0, + "use_env_info_cache": false, + "env_framestack": 1, + "with_wandb": false, + "gym_task_name": "walker2d_rgb_state", + "gym_env_id": "LatencyBench/Walker2dContinuous-v0", + "gym_make_kwargs_json": "{\"base_env_id\": \"Walker2d-v4\", \"base_make_kwargs\": {\"ctrl_cost_weight\": 0.001, \"exclude_current_positions_from_observation\": true, \"forward_reward_weight\": 1.0, \"healthy_reward\": 1.0, \"reset_noise_scale\": 0.005, \"terminate_when_unhealthy\": true}, \"render_mode\": \"rgb_array\"}", + "gym_registration_imports_json": "[\"latency_bench.envs.gymnasium_walker2d\"]", + "gym_action_space_json": "{\"dtype\": \"float32\", \"high\": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0], \"labels\": [\"right_thigh_torque\", \"right_leg_torque\", \"right_foot_torque\", \"left_thigh_torque\", \"left_leg_torque\", \"left_foot_torque\"], \"low\": [-1.0, -1.0, -1.0, -1.0, -1.0, -1.0], \"type\": \"box\"}", + "gym_noop_action_json": "[0.0, 0.0, 0.0, 0.0, 0.0, 0.0]", + "gym_base_prompt": "Move the Walker2d robot forward while keeping its torso upright. Predict six continuous torques in [-1, 1] ordered as right thigh, right leg, right foot, left thigh, left leg, and left foot.", + "gym_state_labels_json": "[\"torso_height\", \"torso_angle\", \"right_thigh_angle\", \"right_leg_angle\", \"right_foot_angle\", \"left_thigh_angle\", \"left_leg_angle\", \"left_foot_angle\", \"torso_x_velocity\", \"torso_z_velocity\", \"torso_angular_velocity\", \"right_thigh_angular_velocity\", \"right_leg_angular_velocity\", \"right_foot_angular_velocity\", \"left_thigh_angular_velocity\", \"left_leg_angular_velocity\", \"left_foot_angular_velocity\"]", + "env_fps": 10.0, + "obs_fps": 10.0, + "frame_stack": 1, + "mode": "train", + "latency_type": "iid", + "latency_seed": 0, + "latency_profile_path": "/mnt/results/latency-sensitive-bench/profile-latency-new/walker2d/20260912T092949Z-walker2d/profile/profile.json", + "latency_profile_worker_slot": 0, + "add_latency_info": false, + "hold_policy": "hold", + "ordering_policy": "issue_order_fifo", + "eval_episodes": 5, + "eval_parallel_envs": 1, + "eval_max_steps": 1000, + "eval_deterministic": true, + "episode_metrics_path": "/mnt/results/latency-sensitive-bench/profile-latency-new/walker2d/20260912T092949Z-walker2d/small_models/walker2d_profile_20260912T092949Z/episode_metrics.jsonl", + "output_dir": "/mnt/results/latency-sensitive-bench/profile-latency-new/walker2d/20260912T092949Z-walker2d/small_train" + }, + "git_hash": "unknown", + "git_repo_name": "not a git repository", + "eval_env_frameskip": 1, + "output_dir": "/mnt/results/latency-sensitive-bench/profile-latency-new/walker2d/20260912T092949Z-walker2d/small_train" + }, + "device_override": "gpu", + "latency_override": { + "method": "iid", + "fixed_latency_ms": null, + "profile_path": "/mnt/results/latency-sensitive-bench/profile-latency-new/walker2d/20260912T092949Z-walker2d/profile/profile.json", + "profile_worker_slot": 0, + "seed": 0, + "add_latency_info": false + } +} diff --git a/latency-aware/walker2d/small-policy/sample-factory-qwenoft-rtx3090-v1/rollout_metadata.json b/latency-aware/walker2d/small-policy/sample-factory-qwenoft-rtx3090-v1/rollout_metadata.json new file mode 100644 index 0000000000000000000000000000000000000000..f21541ae2aca5b278b1b494330590baaa2607182 --- /dev/null +++ b/latency-aware/walker2d/small-policy/sample-factory-qwenoft-rtx3090-v1/rollout_metadata.json @@ -0,0 +1,370 @@ +{ + "checkpoint_experiment_dir": "/mnt/results/latency-sensitive-bench/profile-latency-new/walker2d/20260912T092949Z-walker2d/small_models/walker2d_profile_20260912T092949Z", + "checkpoint_path": "/mnt/results/latency-sensitive-bench/profile-latency-new/walker2d/20260912T092949Z-walker2d/small_models/walker2d_profile_20260912T092949Z/checkpoint_p0/best_000013520_6922240_reward_4348.151.pth", + "config_source": "/mnt/results/latency-sensitive-bench/profile-latency-new/walker2d/20260912T092949Z-walker2d/small_models/walker2d_profile_20260912T092949Z/config.json", + "latency_source": "/mnt/results/latency-sensitive-bench/profile-latency-new/walker2d/20260912T092949Z-walker2d/h100/small_train.yaml", + "checkpoint_config": { + "help": false, + "algo": "APPO", + "env": "latency_gymnasium", + "experiment": "walker2d_profile_20260912T092949Z", + "train_dir": "/mnt/results/latency-sensitive-bench/profile-latency-new/walker2d/20260912T092949Z-walker2d/small_models", + "restart_behavior": "overwrite", + "device": "gpu", + "seed": 0, + "initial_model_path": null, + "num_policies": 1, + "async_rl": false, + "serial_mode": false, + "batched_sampling": false, + "num_batches_to_accumulate": 2, + "worker_num_splits": 2, + "policy_workers_per_policy": 1, + "max_policy_lag": 300, + "num_workers": 8, + "num_envs_per_worker": 8, + "batch_size": 1024, + "num_batches_per_epoch": 4, + "num_epochs": 2, + "rollout": 64, + "fasttd3_replay_capacity": 6553600, + "fasttd3_replay_batch_size": 32768, + "fasttd3_transitions_per_update": 64, + "fasttd3_v_min": -250.0, + "fasttd3_v_max": 250.0, + "fasttd3_actor_action_l2": 0.0, + "fasttd3_compile": true, + "fasttd3_sonic_decoder_path": null, + "recurrence": 1, + "shuffle_minibatches": false, + "gamma": 0.99, + "reward_scale": 1.0, + "reward_clip": 1000.0, + "value_bootstrap": true, + "normalize_returns": true, + "exploration_loss_coeff": 0.0, + "value_loss_coeff": 1.3, + "kl_loss_coeff": 0.1, + "exploration_loss": "entropy", + "gae_lambda": 0.95, + "ppo_clip_ratio": 0.2, + "ppo_clip_value": 0.2, + "with_vtrace": false, + "vtrace_rho": 1.0, + "vtrace_c": 1.0, + "optimizer": "adam", + "adam_eps": 1e-06, + "adam_beta1": 0.9, + "adam_beta2": 0.999, + "max_grad_norm": 3.5, + "learning_rate": 0.00295, + "lr_schedule": "linear_decay", + "lr_schedule_kl_threshold": 0.008, + "lr_adaptive_min": 1e-06, + "lr_adaptive_max": 0.01, + "obs_subtract_mean": 0.0, + "obs_scale": 1.0, + "normalize_input": true, + "normalize_input_keys": null, + "decorrelate_experience_max_seconds": 0, + "decorrelate_envs_on_one_worker": true, + "actor_worker_gpus": [], + "set_workers_cpu_affinity": true, + "force_envs_single_thread": false, + "default_niceness": 0, + "log_to_file": true, + "experiment_summaries_interval": 10, + "flush_summaries_interval": 30, + "stats_avg": 100, + "summaries_use_frameskip": true, + "heartbeat_interval": 20, + "heartbeat_reporting_interval": 180, + "train_for_env_steps": 10000000, + "train_for_seconds": 10000000000, + "save_every_sec": 600, + "keep_checkpoints": 5, + "load_checkpoint_kind": "latest", + "save_milestones_sec": -1, + "save_best_every_sec": 5, + "save_best_metric": "reward", + "save_best_after": 100000, + "benchmark": false, + "encoder_mlp_layers": [ + 64, + 64 + ], + "encoder_conv_architecture": "convnet_simple", + "encoder_conv_mlp_layers": [ + 512 + ], + "use_rnn": false, + "rnn_size": 512, + "rnn_type": "gru", + "rnn_num_layers": 1, + "decoder_mlp_layers": [], + "nonlinearity": "tanh", + "policy_initialization": "torch_default", + "policy_init_gain": 1.0, + "actor_critic_share_weights": true, + "adaptive_stddev": false, + "continuous_tanh_scale": 0.0, + "initial_stddev": 1.0, + "use_env_info_cache": false, + "env_gpu_actions": false, + "env_gpu_observations": true, + "env_frameskip": 1, + "env_framestack": 1, + "pixel_format": "CHW", + "use_record_episode_statistics": false, + "with_wandb": false, + "wandb_user": null, + "wandb_project": "sample_factory", + "wandb_group": null, + "wandb_job_type": "SF", + "wandb_tags": [], + "with_pbt": false, + "pbt_mix_policies_in_one_env": true, + "pbt_period_env_steps": 5000000, + "pbt_start_mutation": 20000000, + "pbt_replace_fraction": 0.3, + "pbt_mutation_rate": 0.15, + "pbt_replace_reward_gap": 0.1, + "pbt_replace_reward_gap_absolute": 1e-06, + "pbt_optimize_gamma": false, + "pbt_target_objective": "true_objective", + "pbt_perturb_min": 1.1, + "pbt_perturb_max": 1.5, + "gym_task_name": "walker2d_rgb_state", + "gym_env_id": "LatencyBench/Walker2dContinuous-v0", + "gym_make_kwargs_json": "{\"base_env_id\": \"Walker2d-v4\", \"base_make_kwargs\": {\"ctrl_cost_weight\": 0.001, \"exclude_current_positions_from_observation\": true, \"forward_reward_weight\": 1.0, \"healthy_reward\": 1.0, \"reset_noise_scale\": 0.005, \"terminate_when_unhealthy\": true}, \"render_mode\": \"rgb_array\"}", + "gym_registration_imports_json": "[\"latency_bench.envs.gymnasium_walker2d\"]", + "gym_action_space_json": "{\"dtype\": \"float32\", \"high\": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0], \"labels\": [\"right_thigh_torque\", \"right_leg_torque\", \"right_foot_torque\", \"left_thigh_torque\", \"left_leg_torque\", \"left_foot_torque\"], \"low\": [-1.0, -1.0, -1.0, -1.0, -1.0, -1.0], \"type\": \"box\"}", + "gym_action_labels_json": null, + "gym_action_values_json": null, + "gym_noop_action_id": null, + "gym_noop_action_json": "[0.0, 0.0, 0.0, 0.0, 0.0, 0.0]", + "gym_base_prompt": "Move the Walker2d robot forward while keeping its torso upright. Predict six continuous torques in [-1, 1] ordered as right thigh, right leg, right foot, left thigh, left leg, and left foot.", + "gym_state_labels_json": "[\"torso_height\", \"torso_angle\", \"right_thigh_angle\", \"right_leg_angle\", \"right_foot_angle\", \"left_thigh_angle\", \"left_leg_angle\", \"left_foot_angle\", \"torso_x_velocity\", \"torso_z_velocity\", \"torso_angular_velocity\", \"right_thigh_angular_velocity\", \"right_leg_angular_velocity\", \"right_foot_angular_velocity\", \"left_thigh_angular_velocity\", \"left_leg_angular_velocity\", \"left_foot_angular_velocity\"]", + "env_fps": 10.0, + "obs_fps": 10.0, + "frame_stack": 1, + "export_env_raw_rgb_frames": false, + "mode": "train", + "latency_type": "iid", + "fixed_latency_ms": null, + "mean_latency_ms": null, + "std_latency_ms": null, + "min_latency_ms": null, + "max_latency_ms": null, + "latency_seed": 0, + "latency_profile_path": "/mnt/results/latency-sensitive-bench/profile-latency-new/walker2d/20260912T092949Z-walker2d/profile/profile.json", + "latency_profile_worker_slot": 0, + "simulated_worker_capacity": null, + "add_latency_info": false, + "max_pending_actions": null, + "hold_policy": "hold", + "ordering_policy": "issue_order_fifo", + "eval_episodes": 5, + "eval_parallel_envs": 1, + "eval_latency_raw_frame_values": null, + "eval_max_steps": 1000, + "eval_deterministic": true, + "eval_raw_reward": false, + "episode_metrics_path": "/mnt/results/latency-sensitive-bench/profile-latency-new/walker2d/20260912T092949Z-walker2d/small_models/walker2d_profile_20260912T092949Z/episode_metrics.jsonl", + "command_line": "--mode train --algo APPO --env latency_gymnasium --experiment walker2d_profile_20260912T092949Z --train_dir /mnt/results/latency-sensitive-bench/profile-latency-new/walker2d/20260912T092949Z-walker2d/small_models --restart_behavior overwrite --device gpu --seed 0 --episode_metrics_path /mnt/results/latency-sensitive-bench/profile-latency-new/walker2d/20260912T092949Z-walker2d/small_models/walker2d_profile_20260912T092949Z/episode_metrics.jsonl --train_for_env_steps 10000000 --num_workers 8 --num_envs_per_worker 8 --num_policies 1 --batch_size 1024 --rollout 64 --recurrence 1 --num_epochs 2 --num_batches_per_epoch 4 --worker_num_splits 2 --max_policy_lag 300 --num_batches_to_accumulate 2 --policy_workers_per_policy 1 --learning_rate 0.00295 --kl_loss_coeff 0.1 --lr_schedule_kl_threshold 0.008 --nonlinearity tanh --policy_initialization torch_default --continuous_tanh_scale 0.0 --initial_stddev 1.0 --env_framestack 1 --gamma 0.99 --gae_lambda 0.95 --ppo_clip_ratio 0.2 --ppo_clip_value 0.2 --exploration_loss entropy --exploration_loss_coeff 0.0 --reward_scale 1.0 --reward_clip 1000.0 --value_loss_coeff 1.3 --max_grad_norm 3.5 --optimizer adam --adam_eps 1e-06 --adam_beta1 0.9 --adam_beta2 0.999 --obs_subtract_mean 0.0 --obs_scale 1.0 --decorrelate_experience_max_seconds 0 --default_niceness 0 --rnn_type gru --rnn_size 512 --save_every_sec 600 --keep_checkpoints 5 --save_milestones_sec -1 --save_best_every_sec 5 --save_best_after 100000 --stats_avg 100 --experiment_summaries_interval 10 --async_rl False --batched_sampling False --serial_mode False --use_rnn False --normalize_returns True --normalize_input True --adaptive_stddev False --shuffle_minibatches False --value_bootstrap True --with_vtrace False --decorrelate_envs_on_one_worker True --set_workers_cpu_affinity True --force_envs_single_thread False --actor_critic_share_weights True --with_wandb False --encoder_mlp_layers 64 64 --latency-type iid --latency-seed 0 --latency-profile-path /mnt/results/latency-sensitive-bench/profile-latency-new/walker2d/20260912T092949Z-walker2d/profile/profile.json --latency-profile-worker-slot 0 --add-latency-info False --eval-episodes 5 --eval-parallel-envs 1 --eval-max-steps 1000 --eval-deterministic True --gym-task-name walker2d_rgb_state --gym-env-id LatencyBench/Walker2dContinuous-v0 --gym-make-kwargs-json {\"base_env_id\": \"Walker2d-v4\", \"base_make_kwargs\": {\"ctrl_cost_weight\": 0.001, \"exclude_current_positions_from_observation\": true, \"forward_reward_weight\": 1.0, \"healthy_reward\": 1.0, \"reset_noise_scale\": 0.005, \"terminate_when_unhealthy\": true}, \"render_mode\": \"rgb_array\"} --gym-registration-imports-json [\"latency_bench.envs.gymnasium_walker2d\"] --gym-action-space-json {\"dtype\": \"float32\", \"high\": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0], \"labels\": [\"right_thigh_torque\", \"right_leg_torque\", \"right_foot_torque\", \"left_thigh_torque\", \"left_leg_torque\", \"left_foot_torque\"], \"low\": [-1.0, -1.0, -1.0, -1.0, -1.0, -1.0], \"type\": \"box\"} --gym-noop-action-json [0.0, 0.0, 0.0, 0.0, 0.0, 0.0] --gym-base-prompt Move the Walker2d robot forward while keeping its torso upright. Predict six continuous torques in [-1, 1] ordered as right thigh, right leg, right foot, left thigh, left leg, and left foot. --gym-state-labels-json [\"torso_height\", \"torso_angle\", \"right_thigh_angle\", \"right_leg_angle\", \"right_foot_angle\", \"left_thigh_angle\", \"left_leg_angle\", \"left_foot_angle\", \"torso_x_velocity\", \"torso_z_velocity\", \"torso_angular_velocity\", \"right_thigh_angular_velocity\", \"right_leg_angular_velocity\", \"right_foot_angular_velocity\", \"left_thigh_angular_velocity\", \"left_leg_angular_velocity\", \"left_foot_angular_velocity\"] --env-fps 10.0 --obs-fps 10.0 --frame-stack 1 --hold-policy hold --ordering-policy issue_order_fifo --use_env_info_cache False", + "cli_args": { + "algo": "APPO", + "env": "latency_gymnasium", + "experiment": "walker2d_profile_20260912T092949Z", + "train_dir": "/mnt/results/latency-sensitive-bench/profile-latency-new/walker2d/20260912T092949Z-walker2d/small_models", + "restart_behavior": "overwrite", + "device": "gpu", + "seed": 0, + "num_policies": 1, + "async_rl": false, + "serial_mode": false, + "batched_sampling": false, + "num_batches_to_accumulate": 2, + "worker_num_splits": 2, + "policy_workers_per_policy": 1, + "max_policy_lag": 300, + "num_workers": 8, + "num_envs_per_worker": 8, + "batch_size": 1024, + "num_batches_per_epoch": 4, + "num_epochs": 2, + "rollout": 64, + "recurrence": 1, + "shuffle_minibatches": false, + "gamma": 0.99, + "reward_scale": 1.0, + "reward_clip": 1000.0, + "value_bootstrap": true, + "normalize_returns": true, + "exploration_loss_coeff": 0.0, + "value_loss_coeff": 1.3, + "kl_loss_coeff": 0.1, + "exploration_loss": "entropy", + "gae_lambda": 0.95, + "ppo_clip_ratio": 0.2, + "ppo_clip_value": 0.2, + "with_vtrace": false, + "optimizer": "adam", + "adam_eps": 1e-06, + "adam_beta1": 0.9, + "adam_beta2": 0.999, + "max_grad_norm": 3.5, + "learning_rate": 0.00295, + "lr_schedule_kl_threshold": 0.008, + "obs_subtract_mean": 0.0, + "obs_scale": 1.0, + "normalize_input": true, + "decorrelate_experience_max_seconds": 0, + "decorrelate_envs_on_one_worker": true, + "set_workers_cpu_affinity": true, + "force_envs_single_thread": false, + "default_niceness": 0, + "experiment_summaries_interval": 10, + "stats_avg": 100, + "train_for_env_steps": 10000000, + "save_every_sec": 600, + "keep_checkpoints": 5, + "save_milestones_sec": -1, + "save_best_every_sec": 5, + "save_best_after": 100000, + "encoder_mlp_layers": [ + 64, + 64 + ], + "use_rnn": false, + "rnn_size": 512, + "rnn_type": "gru", + "nonlinearity": "tanh", + "policy_initialization": "torch_default", + "actor_critic_share_weights": true, + "adaptive_stddev": false, + "continuous_tanh_scale": 0.0, + "initial_stddev": 1.0, + "use_env_info_cache": false, + "env_framestack": 1, + "with_wandb": false, + "gym_task_name": "walker2d_rgb_state", + "gym_env_id": "LatencyBench/Walker2dContinuous-v0", + "gym_make_kwargs_json": "{\"base_env_id\": \"Walker2d-v4\", \"base_make_kwargs\": {\"ctrl_cost_weight\": 0.001, \"exclude_current_positions_from_observation\": true, \"forward_reward_weight\": 1.0, \"healthy_reward\": 1.0, \"reset_noise_scale\": 0.005, \"terminate_when_unhealthy\": true}, \"render_mode\": \"rgb_array\"}", + "gym_registration_imports_json": "[\"latency_bench.envs.gymnasium_walker2d\"]", + "gym_action_space_json": "{\"dtype\": \"float32\", \"high\": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0], \"labels\": [\"right_thigh_torque\", \"right_leg_torque\", \"right_foot_torque\", \"left_thigh_torque\", \"left_leg_torque\", \"left_foot_torque\"], \"low\": [-1.0, -1.0, -1.0, -1.0, -1.0, -1.0], \"type\": \"box\"}", + "gym_noop_action_json": "[0.0, 0.0, 0.0, 0.0, 0.0, 0.0]", + "gym_base_prompt": "Move the Walker2d robot forward while keeping its torso upright. Predict six continuous torques in [-1, 1] ordered as right thigh, right leg, right foot, left thigh, left leg, and left foot.", + "gym_state_labels_json": "[\"torso_height\", \"torso_angle\", \"right_thigh_angle\", \"right_leg_angle\", \"right_foot_angle\", \"left_thigh_angle\", \"left_leg_angle\", \"left_foot_angle\", \"torso_x_velocity\", \"torso_z_velocity\", \"torso_angular_velocity\", \"right_thigh_angular_velocity\", \"right_leg_angular_velocity\", \"right_foot_angular_velocity\", \"left_thigh_angular_velocity\", \"left_leg_angular_velocity\", \"left_foot_angular_velocity\"]", + "env_fps": 10.0, + "obs_fps": 10.0, + "frame_stack": 1, + "mode": "train", + "latency_type": "iid", + "latency_seed": 0, + "latency_profile_path": "/mnt/results/latency-sensitive-bench/profile-latency-new/walker2d/20260912T092949Z-walker2d/profile/profile.json", + "latency_profile_worker_slot": 0, + "add_latency_info": false, + "hold_policy": "hold", + "ordering_policy": "issue_order_fifo", + "eval_episodes": 5, + "eval_parallel_envs": 1, + "eval_max_steps": 1000, + "eval_deterministic": true, + "episode_metrics_path": "/mnt/results/latency-sensitive-bench/profile-latency-new/walker2d/20260912T092949Z-walker2d/small_models/walker2d_profile_20260912T092949Z/episode_metrics.jsonl", + "output_dir": "/mnt/results/latency-sensitive-bench/profile-latency-new/walker2d/20260912T092949Z-walker2d/small_train" + }, + "git_hash": "unknown", + "git_repo_name": "not a git repository", + "eval_env_frameskip": 1, + "output_dir": "/mnt/results/latency-sensitive-bench/profile-latency-new/walker2d/20260912T092949Z-walker2d/small_train" + }, + "latency_override": { + "method": "iid", + "fixed_latency_ms": null, + "profile_path": "/mnt/results/latency-sensitive-bench/profile-latency-new/walker2d/20260912T092949Z-walker2d/profile/profile.json", + "profile_worker_slot": 0, + "seed": 0, + "add_latency_info": false + }, + "gymnasium_task": { + "task_name": "walker2d_rgb_state", + "env_id": "LatencyBench/Walker2dContinuous-v0", + "make_kwargs": { + "base_env_id": "Walker2d-v4", + "base_make_kwargs": { + "ctrl_cost_weight": 0.001, + "exclude_current_positions_from_observation": true, + "forward_reward_weight": 1.0, + "healthy_reward": 1.0, + "reset_noise_scale": 0.005, + "terminate_when_unhealthy": true + }, + "render_mode": "rgb_array" + }, + "registration_imports": [ + "latency_bench.envs.gymnasium_walker2d" + ], + "base_prompt": "Move the Walker2d robot forward while keeping its torso upright. Predict six continuous torques in [-1, 1] ordered as right thigh, right leg, right foot, left thigh, left leg, and left foot.", + "env_fps": 10.0, + "obs_fps": 10.0, + "frame_stack": 1, + "state_space": { + "labels": [ + "torso_height", + "torso_angle", + "right_thigh_angle", + "right_leg_angle", + "right_foot_angle", + "left_thigh_angle", + "left_leg_angle", + "left_foot_angle", + "torso_x_velocity", + "torso_z_velocity", + "torso_angular_velocity", + "right_thigh_angular_velocity", + "right_leg_angular_velocity", + "right_foot_angular_velocity", + "left_thigh_angular_velocity", + "left_leg_angular_velocity", + "left_foot_angular_velocity" + ] + }, + "action_space": { + "dtype": "float32", + "high": [ + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0 + ], + "labels": [ + "right_thigh_torque", + "right_leg_torque", + "right_foot_torque", + "left_thigh_torque", + "left_leg_torque", + "left_foot_torque" + ], + "low": [ + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0 + ], + "type": "box" + }, + "noop_action": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + } +} diff --git a/latency-aware/walker2d/small-policy/sample-factory-qwenoft-rtx3090-v1/selection.json b/latency-aware/walker2d/small-policy/sample-factory-qwenoft-rtx3090-v1/selection.json new file mode 100644 index 0000000000000000000000000000000000000000..7f1ce0fe0fd69b072651f0ab9a459f4ce1cb6f42 --- /dev/null +++ b/latency-aware/walker2d/small-policy/sample-factory-qwenoft-rtx3090-v1/selection.json @@ -0,0 +1,14 @@ +{ + "run_id": "20260912T092949Z-walker2d", + "source_profile_run_id": "20260912T093545326611Z", + "selection_rule": "best_reward_after_full_training_budget", + "best_checkpoint": "/mnt/results/latency-sensitive-bench/profile-latency-new/walker2d/20260912T092949Z-walker2d/small_models/walker2d_profile_20260912T092949Z/checkpoint_p0/best_000013520_6922240_reward_4348.151.pth", + "best_bundle_path": "checkpoint_p0/best_000013520_6922240_reward_4348.151.pth", + "final_checkpoint": "/mnt/results/latency-sensitive-bench/profile-latency-new/walker2d/20260912T092949Z-walker2d/small_models/walker2d_profile_20260912T092949Z/checkpoint_p0/checkpoint_000019544_10006528.pth", + "final_env_steps": 10006528, + "final_train_step": 19544, + "training_exit_code": 0, + "profile_sha256": "60fc378a94af03b24f71a879e2374647b90f3bf0e74be83e9a27544cdde2c12f", + "code_root_sha": "86df5ecef6735d5d824944e7acdee5bf204e1545", + "note": "config and rollout descriptor preserve exact H100 training paths; rebind latency profile to bundled profile/profile.json when moving hosts" +} diff --git a/latency-aware/walker2d/small-policy/sample-factory-qwenoft-rtx3090-v1/small_train.yaml b/latency-aware/walker2d/small-policy/sample-factory-qwenoft-rtx3090-v1/small_train.yaml new file mode 100644 index 0000000000000000000000000000000000000000..0dc751ce9e2d360dd67849a274939ea51c3a86a8 --- /dev/null +++ b/latency-aware/walker2d/small-policy/sample-factory-qwenoft-rtx3090-v1/small_train.yaml @@ -0,0 +1,178 @@ +experiment: + name: walker2d_profile_20260912T092949Z + seed: 0 +backend: + type: sample_factory + algo: APPO + device: cuda + train_dir: /mnt/results/latency-sensitive-bench/profile-latency-new/walker2d/20260912T092949Z-walker2d/small_models + restart_behavior: overwrite + run_mode: train +executor: + mode: simulated +env: + name: gymnasium + action_space: + dtype: float32 + high: + - 1.0 + - 1.0 + - 1.0 + - 1.0 + - 1.0 + - 1.0 + labels: + - right_thigh_torque + - right_leg_torque + - right_foot_torque + - left_thigh_torque + - left_leg_torque + - left_foot_torque + low: + - -1.0 + - -1.0 + - -1.0 + - -1.0 + - -1.0 + - -1.0 + type: box + base_prompt: Move the Walker2d robot forward while keeping its torso upright. Predict + six continuous torques in [-1, 1] ordered as right thigh, right leg, right foot, + left thigh, left leg, and left foot. + env_fps: 10.0 + env_id: LatencyBench/Walker2dContinuous-v0 + frame_stack: 1 + make_kwargs: + base_env_id: Walker2d-v4 + base_make_kwargs: + ctrl_cost_weight: 0.001 + exclude_current_positions_from_observation: true + forward_reward_weight: 1.0 + healthy_reward: 1.0 + reset_noise_scale: 0.005 + terminate_when_unhealthy: true + render_mode: rgb_array + noop_action: + - 0.0 + - 0.0 + - 0.0 + - 0.0 + - 0.0 + - 0.0 + obs_fps: 10.0 + registration_imports: + - latency_bench.envs.gymnasium_walker2d + state_space: + labels: + - torso_height + - torso_angle + - right_thigh_angle + - right_leg_angle + - right_foot_angle + - left_thigh_angle + - left_leg_angle + - left_foot_angle + - torso_x_velocity + - torso_z_velocity + - torso_angular_velocity + - right_thigh_angular_velocity + - right_leg_angular_velocity + - right_foot_angular_velocity + - left_thigh_angular_velocity + - left_leg_angular_velocity + - left_foot_angular_velocity + task_name: walker2d_rgb_state +latency: + method: iid + fixed_latency_ms: null + profile_path: /mnt/results/latency-sensitive-bench/profile-latency-new/walker2d/20260912T092949Z-walker2d/profile/profile.json + profile_worker_slot: 0 + seed: 0 + add_latency_info: false +scheduler: + hold_policy: hold + ordering_policy: issue_order_fifo +policy: + type: random +training: + train_for_env_steps: 10000000 + num_workers: 8 + num_envs_per_worker: 8 + worker_num_splits: 2 + num_policies: 1 + batch_size: 1024 + rollout: 64 + recurrence: 1 + num_epochs: 2 + num_batches_per_epoch: 4 + max_policy_lag: 300 + learning_rate: 0.00295 + gamma: 0.99 + gae_lambda: 0.95 + ppo_clip_ratio: 0.2 + ppo_clip_value: 0.2 + value_loss_coeff: 1.3 + max_grad_norm: 3.5 + exploration_loss_coeff: 0.0 + async_rl: false + batched_sampling: false + use_rnn: false + encoder_mlp_layers: + - 64 + - 64 + normalize_input: true + normalize_returns: true + save_every_sec: 600 + keep_checkpoints: 5 + num_batches_to_accumulate: 2 + policy_workers_per_policy: 1 + kl_loss_coeff: 0.1 + lr_schedule_kl_threshold: 0.008 + nonlinearity: tanh + policy_initialization: torch_default + continuous_tanh_scale: 0.0 + initial_stddev: 1.0 + env_framestack: 1 + exploration_loss: entropy + reward_scale: 1.0 + reward_clip: 1000.0 + optimizer: adam + adam_eps: 1.0e-06 + adam_beta1: 0.9 + adam_beta2: 0.999 + obs_subtract_mean: 0.0 + obs_scale: 1.0 + decorrelate_experience_max_seconds: 0 + default_niceness: 0 + rnn_type: gru + rnn_size: 512 + save_milestones_sec: -1 + save_best_every_sec: 5 + save_best_after: 100000 + stats_avg: 100 + experiment_summaries_interval: 10 + serial_mode: false + adaptive_stddev: false + shuffle_minibatches: false + value_bootstrap: true + with_vtrace: false + decorrelate_envs_on_one_worker: true + set_workers_cpu_affinity: true + force_envs_single_thread: false + actor_critic_share_weights: true + with_wandb: false + lr_schedule: linear_decay +evaluation: + eval_interval_steps: null + eval_episodes: 5 + eval_parallel_envs: 1 + eval_max_steps: 1000 + eval_deterministic: true +logging: + output_dir: /mnt/results/latency-sensitive-bench/profile-latency-new/walker2d/20260912T092949Z-walker2d/small_train + video: + enabled: false + num_bins: 1 + save_step_records: false + save_action_records: false + save_latency_records: false diff --git a/latency-aware/walker2d/small-policy/sample-factory-qwenpi_v3-rtx3090-v1/README.md b/latency-aware/walker2d/small-policy/sample-factory-qwenpi_v3-rtx3090-v1/README.md new file mode 100644 index 0000000000000000000000000000000000000000..6050d70dd2ad4bf424af3c8641364681b31c3b22 --- /dev/null +++ b/latency-aware/walker2d/small-policy/sample-factory-qwenpi_v3-rtx3090-v1/README.md @@ -0,0 +1,30 @@ +# walker2d / sample-factory-appo + +Training condition: `latency-aware`. Run: `pi05_walker2d_profile_appo_h1_5fps_g128_20260921`. + +Copied from [Standard-Pipeline](https://huggingface.co/latency-sensitive-bench/Standard-Pipeline/tree/2208875f92b21e3f8ec2363a30a6705dfe56cc04/walker2d/profile_latency/small_model/Pi05) at `2208875f92b21e3f8ec2363a30a6705dfe56cc04`. + +- Checkpoint: `checkpoint.pth` +- Selection: training_best +- Checkpoint SHA256: `c9164f8e440bd6ac8b41e47f493dc2eea6945bb7520bbc08e851870fda2ac21f` +- Configuration, FPS, actions, normalization and latency semantics retain their source values. +- This import does not run evaluations or change training/evaluation/acceptance status. + +Sample Factory APPO inference bundle: `config.json` and the explicit `checkpoint.pth`. +The checkpoint contains `model`, `train_step`, and `env_steps`; every exported model +tensor was reloaded and compared bit-for-bit with the selected source checkpoint. + +See [the original model notes](source/README.md) for experiment results and limitations. + +[Original provenance](source/provenance.json) retains dataset references and recorded experiment evidence. + +## Download + +```python +from huggingface_hub import snapshot_download +prefix = "latency-aware/walker2d/small-policy/sample-factory-qwenpi_v3-rtx3090-v1" +snapshot_download("latency-sensitive-bench/benchmark-models", revision="", + allow_patterns=[prefix + "/**"]) +``` + +Use the verified import commit listed in the repository's import report. diff --git a/latency-aware/walker2d/small-policy/sample-factory-qwenpi_v3-rtx3090-v1/checkpoint.pth b/latency-aware/walker2d/small-policy/sample-factory-qwenpi_v3-rtx3090-v1/checkpoint.pth new file mode 100644 index 0000000000000000000000000000000000000000..34e2c8a550422bacd8324ac11e226fc199db42d5 --- /dev/null +++ b/latency-aware/walker2d/small-policy/sample-factory-qwenpi_v3-rtx3090-v1/checkpoint.pth @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c9164f8e440bd6ac8b41e47f493dc2eea6945bb7520bbc08e851870fda2ac21f +size 29835 diff --git a/latency-aware/walker2d/small-policy/sample-factory-qwenpi_v3-rtx3090-v1/config.json b/latency-aware/walker2d/small-policy/sample-factory-qwenpi_v3-rtx3090-v1/config.json new file mode 100644 index 0000000000000000000000000000000000000000..882143de9f47103982de8193d365f65a09ce3223 --- /dev/null +++ b/latency-aware/walker2d/small-policy/sample-factory-qwenpi_v3-rtx3090-v1/config.json @@ -0,0 +1,272 @@ +{ + "help": false, + "algo": "APPO", + "env": "latency_gymnasium", + "experiment": "pi05_walker2d_profile_appo_h1_5fps_g128_20260921", + "train_dir": "${PI05_RUN_DIR}/profile_latency/teacher/checkpoints", + "restart_behavior": "overwrite", + "device": "gpu", + "seed": 0, + "initial_model_path": null, + "num_policies": 1, + "async_rl": false, + "serial_mode": false, + "batched_sampling": false, + "num_batches_to_accumulate": 2, + "worker_num_splits": 2, + "policy_workers_per_policy": 1, + "max_policy_lag": 300, + "num_workers": 8, + "num_envs_per_worker": 8, + "batch_size": 1024, + "num_batches_per_epoch": 4, + "num_epochs": 2, + "rollout": 64, + "fasttd3_replay_capacity": 6553600, + "fasttd3_replay_batch_size": 32768, + "fasttd3_action_chunk_horizon": 1, + "fasttd3_transitions_per_update": 64, + "fasttd3_train_for_optimizer_steps": 10000000000, + "fasttd3_v_min": -250.0, + "fasttd3_v_max": 250.0, + "fasttd3_actor_action_l2": 0.0, + "fasttd3_compile": true, + "recurrence": 1, + "shuffle_minibatches": false, + "gamma": 0.99, + "reward_scale": 1.0, + "reward_clip": 1000.0, + "value_bootstrap": true, + "normalize_returns": true, + "exploration_loss_coeff": 0.0, + "value_loss_coeff": 1.3, + "kl_loss_coeff": 0.1, + "exploration_loss": "entropy", + "gae_lambda": 0.95, + "ppo_clip_ratio": 0.2, + "ppo_clip_value": 0.2, + "with_vtrace": false, + "vtrace_rho": 1.0, + "vtrace_c": 1.0, + "optimizer": "adam", + "adam_eps": 1e-06, + "adam_beta1": 0.9, + "adam_beta2": 0.999, + "max_grad_norm": 3.5, + "learning_rate": 0.00295, + "lr_schedule": "linear_decay", + "lr_schedule_kl_threshold": 0.008, + "lr_adaptive_min": 1e-06, + "lr_adaptive_max": 0.01, + "obs_subtract_mean": 0.0, + "obs_scale": 1.0, + "normalize_input": true, + "normalize_input_keys": null, + "decorrelate_experience_max_seconds": 0, + "decorrelate_envs_on_one_worker": true, + "actor_worker_gpus": [], + "set_workers_cpu_affinity": true, + "force_envs_single_thread": false, + "default_niceness": 0, + "log_to_file": true, + "experiment_summaries_interval": 10, + "flush_summaries_interval": 30, + "stats_avg": 100, + "summaries_use_frameskip": true, + "heartbeat_interval": 20, + "heartbeat_reporting_interval": 180, + "train_for_env_steps": 10000000, + "train_for_seconds": 10000000000, + "save_every_sec": 600, + "keep_checkpoints": 5, + "load_checkpoint_kind": "latest", + "save_milestones_sec": -1, + "save_best_every_sec": 5, + "save_best_metric": "reward", + "save_best_after": 100000, + "benchmark": false, + "encoder_mlp_layers": [ + 64, + 64 + ], + "encoder_conv_architecture": "convnet_simple", + "encoder_conv_mlp_layers": [ + 512 + ], + "use_rnn": false, + "rnn_size": 512, + "rnn_type": "gru", + "rnn_num_layers": 1, + "decoder_mlp_layers": [], + "nonlinearity": "tanh", + "policy_initialization": "torch_default", + "policy_init_gain": 1.0, + "actor_critic_share_weights": true, + "adaptive_stddev": false, + "continuous_tanh_scale": 0.0, + "initial_stddev": 1.0, + "use_env_info_cache": false, + "env_gpu_actions": false, + "env_gpu_observations": true, + "env_frameskip": 1, + "env_framestack": 1, + "pixel_format": "CHW", + "use_record_episode_statistics": false, + "with_wandb": true, + "wandb_user": "dongqianyu99-zhejiang-university", + "wandb_project": "latency-sensitive-bench", + "wandb_group": "pi05-profile-teachers-h1-5fps-20260921", + "wandb_job_type": "profile_teacher", + "wandb_tags": [ + "walker2d", + "APPO", + "Pi05", + "measured-IID-profile", + "H1", + "5FPS" + ], + "with_pbt": false, + "pbt_mix_policies_in_one_env": true, + "pbt_period_env_steps": 5000000, + "pbt_start_mutation": 20000000, + "pbt_replace_fraction": 0.3, + "pbt_mutation_rate": 0.15, + "pbt_replace_reward_gap": 0.1, + "pbt_replace_reward_gap_absolute": 1e-06, + "pbt_optimize_gamma": false, + "pbt_target_objective": "true_objective", + "pbt_perturb_min": 1.1, + "pbt_perturb_max": 1.5, + "gym_task_name": "walker2d", + "gym_env_id": "LatencyBench/Walker2dContinuous-v0", + "gym_make_kwargs_json": "{\"base_env_id\": \"Walker2d-v4\", \"render_mode\": \"rgb_array\", \"base_make_kwargs\": {\"forward_reward_weight\": 1.0, \"ctrl_cost_weight\": 0.001, \"healthy_reward\": 1.0, \"terminate_when_unhealthy\": true, \"reset_noise_scale\": 0.005, \"exclude_current_positions_from_observation\": true}}", + "gym_registration_imports_json": "[\"latency_bench.envs.gymnasium_walker2d\"]", + "gym_action_space_json": "{\"type\": \"box\", \"labels\": [\"right_thigh_torque\", \"right_leg_torque\", \"right_foot_torque\", \"left_thigh_torque\", \"left_leg_torque\", \"left_foot_torque\"], \"low\": [-1.0, -1.0, -1.0, -1.0, -1.0, -1.0], \"high\": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0], \"dtype\": \"float32\"}", + "gym_action_labels_json": null, + "gym_action_values_json": null, + "gym_noop_action_id": null, + "gym_noop_action_json": "[0.0, 0.0, 0.0, 0.0, 0.0, 0.0]", + "gym_base_prompt": "Move the Walker2d robot forward while keeping its torso upright. Predict six continuous torques in [-1, 1] ordered as right thigh, right leg, right foot, left thigh, left leg, and left foot.", + "gym_state_labels_json": "[\"torso_height\", \"torso_angle\", \"right_thigh_angle\", \"right_leg_angle\", \"right_foot_angle\", \"left_thigh_angle\", \"left_leg_angle\", \"left_foot_angle\", \"torso_x_velocity\", \"torso_z_velocity\", \"torso_angular_velocity\", \"right_thigh_angular_velocity\", \"right_leg_angular_velocity\", \"right_foot_angular_velocity\", \"left_thigh_angular_velocity\", \"left_leg_angular_velocity\", \"left_foot_angular_velocity\"]", + "env_fps": 5.0, + "obs_fps": 5.0, + "frame_stack": 1, + "export_env_raw_rgb_frames": false, + "mode": "train", + "latency_type": "iid", + "fixed_latency_ms": null, + "mean_latency_ms": null, + "std_latency_ms": null, + "min_latency_ms": null, + "max_latency_ms": null, + "latency_seed": 0, + "latency_profile_path": "${PI05_RUN_DIR}/profile_latency/profile/profile.json", + "latency_profile_worker_slot": 0, + "simulated_worker_capacity": null, + "add_latency_info": false, + "max_pending_actions": null, + "hold_last_chunk_action": false, + "hold_policy": "hold", + "ordering_policy": "issue_order_fifo", + "eval_episodes": 5, + "eval_parallel_envs": 1, + "eval_latency_raw_frame_values": null, + "eval_max_steps": 1000, + "eval_deterministic": true, + "eval_raw_reward": false, + "episode_metrics_path": "${PI05_RUN_DIR}/profile_latency/teacher/checkpoints/pi05_walker2d_profile_appo_h1_5fps_g128_20260921/episode_metrics.jsonl", + "ppo": null, + "command_line": "--mode train --algo APPO --env latency_gymnasium --experiment pi05_walker2d_profile_appo_h1_5fps_g128_20260921 --train_dir ${PI05_RUN_DIR}/profile_latency/teacher/checkpoints --restart_behavior overwrite --device gpu --seed 0 --episode_metrics_path ${PI05_RUN_DIR}/profile_latency/teacher/checkpoints/pi05_walker2d_profile_appo_h1_5fps_g128_20260921/episode_metrics.jsonl --train_for_env_steps 10000000 --num_workers 8 --num_envs_per_worker 8 --num_policies 1 --batch_size 1024 --rollout 64 --recurrence 1 --num_epochs 2 --num_batches_per_epoch 4 --worker_num_splits 2 --max_policy_lag 300 --learning_rate 0.00295 --kl_loss_coeff 0.1 --nonlinearity tanh --policy_initialization torch_default --initial_stddev 1.0 --gamma 0.99 --gae_lambda 0.95 --ppo_clip_ratio 0.2 --ppo_clip_value 0.2 --exploration_loss_coeff 0.0 --value_loss_coeff 1.3 --max_grad_norm 3.5 --save_every_sec 600 --keep_checkpoints 5 --async_rl False --batched_sampling False --serial_mode False --use_rnn False --normalize_returns True --normalize_input True --adaptive_stddev False --shuffle_minibatches False --value_bootstrap True --encoder_mlp_layers 64 64 --latency-type iid --latency-seed 0 --latency-profile-path ${PI05_RUN_DIR}/profile_latency/profile/profile.json --latency-profile-worker-slot 0 --add-latency-info False --eval-episodes 5 --eval-parallel-envs 1 --eval-max-steps 1000 --eval-deterministic True --with_wandb True --wandb_project latency-sensitive-bench --wandb_group pi05-profile-teachers-h1-5fps-20260921 --wandb_job_type profile_teacher --wandb_tags walker2d APPO Pi05 measured-IID-profile H1 5FPS --wandb_user dongqianyu99-zhejiang-university --gym-task-name walker2d --gym-env-id LatencyBench/Walker2dContinuous-v0 --gym-make-kwargs-json {\"base_env_id\": \"Walker2d-v4\", \"render_mode\": \"rgb_array\", \"base_make_kwargs\": {\"forward_reward_weight\": 1.0, \"ctrl_cost_weight\": 0.001, \"healthy_reward\": 1.0, \"terminate_when_unhealthy\": true, \"reset_noise_scale\": 0.005, \"exclude_current_positions_from_observation\": true}} --gym-registration-imports-json [\"latency_bench.envs.gymnasium_walker2d\"] --gym-action-space-json {\"type\": \"box\", \"labels\": [\"right_thigh_torque\", \"right_leg_torque\", \"right_foot_torque\", \"left_thigh_torque\", \"left_leg_torque\", \"left_foot_torque\"], \"low\": [-1.0, -1.0, -1.0, -1.0, -1.0, -1.0], \"high\": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0], \"dtype\": \"float32\"} --gym-noop-action-json [0.0, 0.0, 0.0, 0.0, 0.0, 0.0] --gym-base-prompt Move the Walker2d robot forward while keeping its torso upright. Predict six continuous torques in [-1, 1] ordered as right thigh, right leg, right foot, left thigh, left leg, and left foot. --gym-state-labels-json [\"torso_height\", \"torso_angle\", \"right_thigh_angle\", \"right_leg_angle\", \"right_foot_angle\", \"left_thigh_angle\", \"left_leg_angle\", \"left_foot_angle\", \"torso_x_velocity\", \"torso_z_velocity\", \"torso_angular_velocity\", \"right_thigh_angular_velocity\", \"right_leg_angular_velocity\", \"right_foot_angular_velocity\", \"left_thigh_angular_velocity\", \"left_leg_angular_velocity\", \"left_foot_angular_velocity\"] --env-fps 5 --obs-fps 5 --frame-stack 1 --hold-policy hold --ordering-policy issue_order_fifo --use_env_info_cache False", + "cli_args": { + "algo": "APPO", + "env": "latency_gymnasium", + "experiment": "pi05_walker2d_profile_appo_h1_5fps_g128_20260921", + "train_dir": "${PI05_RUN_DIR}/profile_latency/teacher/checkpoints", + "restart_behavior": "overwrite", + "device": "gpu", + "seed": 0, + "num_policies": 1, + "async_rl": false, + "serial_mode": false, + "batched_sampling": false, + "worker_num_splits": 2, + "max_policy_lag": 300, + "num_workers": 8, + "num_envs_per_worker": 8, + "batch_size": 1024, + "num_batches_per_epoch": 4, + "num_epochs": 2, + "rollout": 64, + "recurrence": 1, + "shuffle_minibatches": false, + "gamma": 0.99, + "value_bootstrap": true, + "normalize_returns": true, + "exploration_loss_coeff": 0.0, + "value_loss_coeff": 1.3, + "kl_loss_coeff": 0.1, + "gae_lambda": 0.95, + "ppo_clip_ratio": 0.2, + "ppo_clip_value": 0.2, + "max_grad_norm": 3.5, + "learning_rate": 0.00295, + "normalize_input": true, + "train_for_env_steps": 10000000, + "save_every_sec": 600, + "keep_checkpoints": 5, + "encoder_mlp_layers": [ + 64, + 64 + ], + "use_rnn": false, + "nonlinearity": "tanh", + "policy_initialization": "torch_default", + "adaptive_stddev": false, + "initial_stddev": 1.0, + "use_env_info_cache": false, + "with_wandb": true, + "wandb_user": "dongqianyu99-zhejiang-university", + "wandb_project": "latency-sensitive-bench", + "wandb_group": "pi05-profile-teachers-h1-5fps-20260921", + "wandb_job_type": "profile_teacher", + "wandb_tags": [ + "walker2d", + "APPO", + "Pi05", + "measured-IID-profile", + "H1", + "5FPS" + ], + "gym_task_name": "walker2d", + "gym_env_id": "LatencyBench/Walker2dContinuous-v0", + "gym_make_kwargs_json": "{\"base_env_id\": \"Walker2d-v4\", \"render_mode\": \"rgb_array\", \"base_make_kwargs\": {\"forward_reward_weight\": 1.0, \"ctrl_cost_weight\": 0.001, \"healthy_reward\": 1.0, \"terminate_when_unhealthy\": true, \"reset_noise_scale\": 0.005, \"exclude_current_positions_from_observation\": true}}", + "gym_registration_imports_json": "[\"latency_bench.envs.gymnasium_walker2d\"]", + "gym_action_space_json": "{\"type\": \"box\", \"labels\": [\"right_thigh_torque\", \"right_leg_torque\", \"right_foot_torque\", \"left_thigh_torque\", \"left_leg_torque\", \"left_foot_torque\"], \"low\": [-1.0, -1.0, -1.0, -1.0, -1.0, -1.0], \"high\": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0], \"dtype\": \"float32\"}", + "gym_noop_action_json": "[0.0, 0.0, 0.0, 0.0, 0.0, 0.0]", + "gym_base_prompt": "Move the Walker2d robot forward while keeping its torso upright. Predict six continuous torques in [-1, 1] ordered as right thigh, right leg, right foot, left thigh, left leg, and left foot.", + "gym_state_labels_json": "[\"torso_height\", \"torso_angle\", \"right_thigh_angle\", \"right_leg_angle\", \"right_foot_angle\", \"left_thigh_angle\", \"left_leg_angle\", \"left_foot_angle\", \"torso_x_velocity\", \"torso_z_velocity\", \"torso_angular_velocity\", \"right_thigh_angular_velocity\", \"right_leg_angular_velocity\", \"right_foot_angular_velocity\", \"left_thigh_angular_velocity\", \"left_leg_angular_velocity\", \"left_foot_angular_velocity\"]", + "env_fps": 5.0, + "obs_fps": 5.0, + "frame_stack": 1, + "mode": "train", + "latency_type": "iid", + "latency_seed": 0, + "latency_profile_path": "${PI05_RUN_DIR}/profile_latency/profile/profile.json", + "latency_profile_worker_slot": 0, + "add_latency_info": false, + "hold_policy": "hold", + "ordering_policy": "issue_order_fifo", + "eval_episodes": 5, + "eval_parallel_envs": 1, + "eval_max_steps": 1000, + "eval_deterministic": true, + "episode_metrics_path": "${PI05_RUN_DIR}/profile_latency/teacher/checkpoints/pi05_walker2d_profile_appo_h1_5fps_g128_20260921/episode_metrics.jsonl", + "output_dir": "${PI05_RUN_DIR}/profile_latency/teacher/training", + "wandb_unique_id": "pi05_walker2d_profile_appo_h1_5fps_g128_20260921" + }, + "git_hash": "unknown", + "git_repo_name": "not a git repository", + "eval_env_frameskip": 1, + "output_dir": "${PI05_RUN_DIR}/profile_latency/teacher/training", + "wandb_unique_id": "pi05_walker2d_profile_appo_h1_5fps_g128_20260921" +} \ No newline at end of file diff --git a/latency-aware/walker2d/small-policy/sample-factory-qwenpi_v3-rtx3090-v1/provenance.json b/latency-aware/walker2d/small-policy/sample-factory-qwenpi_v3-rtx3090-v1/provenance.json new file mode 100644 index 0000000000000000000000000000000000000000..103c549350bc08d0bb371806d9cb6c701d455aef --- /dev/null +++ b/latency-aware/walker2d/small-policy/sample-factory-qwenpi_v3-rtx3090-v1/provenance.json @@ -0,0 +1,40 @@ +{ + "task": "walker2d", + "model": "sample-factory-appo", + "training_condition": "latency-aware", + "training_run_id": "pi05_walker2d_profile_appo_h1_5fps_g128_20260921", + "source": { + "repo_id": "latency-sensitive-bench/Standard-Pipeline", + "revision": "2208875f92b21e3f8ec2363a30a6705dfe56cc04", + "path": "walker2d/profile_latency/small_model/Pi05", + "url": "https://huggingface.co/latency-sensitive-bench/Standard-Pipeline/tree/2208875f92b21e3f8ec2363a30a6705dfe56cc04/walker2d/profile_latency/small_model/Pi05" + }, + "hf_repo_id": "latency-sensitive-bench/benchmark-models", + "hf_path": "latency-aware/walker2d/small-policy/sample-factory-qwenpi_v3-rtx3090-v1", + "checkpoint": { + "source_file": "walker2d/profile_latency/small_model/Pi05/checkpoint_p0/selected.pth", + "source_sha256": "c9164f8e440bd6ac8b41e47f493dc2eea6945bb7520bbc08e851870fda2ac21f", + "source_bytes": 29835, + "file": "checkpoint.pth", + "selection_rule": "training_best", + "method": "copy", + "sha256": "c9164f8e440bd6ac8b41e47f493dc2eea6945bb7520bbc08e851870fda2ac21f", + "bytes": 29835, + "train_step": 18928, + "env_steps": 9691136, + "tensor_count": 15, + "tensor_equality_verified": true, + "removed_fields": [] + }, + "config_source": "walker2d/profile_latency/small_model/Pi05/config.json", + "task_contract_file": null, + "missing_source_files": [], + "migration": { + "operation": "copy_selected_inference_bundle", + "source_modified": false, + "evaluation_performed": false, + "experiment_status": "preserved_in_source_evidence" + }, + "algorithm": "APPO", + "latency_model": "qwenpi_v3" +} diff --git a/latency-aware/walker2d/small-policy/sample-factory-qwenpi_v3-rtx3090-v1/selection.json b/latency-aware/walker2d/small-policy/sample-factory-qwenpi_v3-rtx3090-v1/selection.json new file mode 100644 index 0000000000000000000000000000000000000000..11e2057a3fb90ab3ce26a5b976d1a5f2a98c6775 --- /dev/null +++ b/latency-aware/walker2d/small-policy/sample-factory-qwenpi_v3-rtx3090-v1/selection.json @@ -0,0 +1,116 @@ +{ + "selected": "training_best", + "selection_policy": "higher_mean_over_paired_20; ties_select_final", + "candidates": { + "final": { + "checkpoint": "${PI05_RUN_DIR}/profile_latency/teacher/checkpoints/pi05_walker2d_profile_appo_h1_5fps_g128_20260921/checkpoint_p0/checkpoint_000019536_10002432.pth", + "sha256": "e157a7be891b0b498028293fc7be14eda97cd41b485220f558ea3d5a88c77171", + "train_step": 19536, + "env_steps": 10002432, + "episodes": 20, + "seeds": [ + 104771, + 104772, + 104773, + 104774, + 104775, + 104776, + 104777, + 104778, + 104779, + 104780, + 104781, + 104782, + 104783, + 104784, + 104785, + 104786, + 104787, + 104788, + 104789, + 104790 + ], + "returns": [ + 4337.574338552904, + 2124.168087172507, + 3803.271661035685, + 4247.230688017356, + 1180.6139208871832, + 1196.3837779316377, + 3778.6057355245894, + 3562.7070253642705, + 1156.377364775773, + 4000.2209267455473, + 1196.9960923429844, + 4196.1107606718415, + 1161.7798657840565, + 1163.344082665548, + 4193.403531977741, + 2077.0328874033744, + 2716.3877118621144, + 1142.383303495549, + 1197.1537112200667, + 2535.48704209563 + ], + "mean": 2548.361625776318, + "population_std": 1289.7687169980788, + "strict_gt3000": 8 + }, + "training_best": { + "checkpoint": "${PI05_RUN_DIR}/profile_latency/teacher/checkpoints/pi05_walker2d_profile_appo_h1_5fps_g128_20260921/checkpoint_p0/best_000018928_9691136_reward_4148.139.pth", + "sha256": "3d51959c7d65d8e9d1ec34cfc4a8dc7741abf9726439512bc7a0442fd265044e", + "train_step": 18928, + "env_steps": 9691136, + "episodes": 20, + "seeds": [ + 104771, + 104772, + 104773, + 104774, + 104775, + 104776, + 104777, + 104778, + 104779, + 104780, + 104781, + 104782, + 104783, + 104784, + 104785, + 104786, + 104787, + 104788, + 104789, + 104790 + ], + "returns": [ + 2454.017948041446, + 3517.5093910555765, + 4443.205483476128, + 4155.968346173417, + 3904.8882905894393, + 1210.1385452181764, + 3550.0328741089725, + 1841.8260301793923, + 4000.010779902804, + 3974.0655123428087, + 2544.925200849052, + 3898.4951632085545, + 3853.089408732446, + 4233.270564039822, + 3649.040265741407, + 3954.793190236864, + 3944.9064060851774, + 3956.211685978782, + 3755.029851073333, + 3969.2680678441934 + ], + "mean": 3540.5346502438897, + "population_std": 826.7207979036164, + "strict_gt3000": 16 + } + }, + "selected_checkpoint_sha256": "3d51959c7d65d8e9d1ec34cfc4a8dc7741abf9726439512bc7a0442fd265044e", + "completed_at_utc": "2026-09-21T14:08:56.270953+00:00" +} diff --git a/latency-aware/walker2d/small-policy/sample-factory-qwenpi_v3-rtx3090-v1/source/README.md b/latency-aware/walker2d/small-policy/sample-factory-qwenpi_v3-rtx3090-v1/source/README.md new file mode 100644 index 0000000000000000000000000000000000000000..de5b46055d79d97d357b754e8e4907818bc08c9e --- /dev/null +++ b/latency-aware/walker2d/small-policy/sample-factory-qwenpi_v3-rtx3090-v1/source/README.md @@ -0,0 +1,7 @@ +# Walker2D Pi0.5 profile-latency APPO checkpoint + +Selected `training_best` at train step 18928 / 9691136 Sample Factory environment steps, seed 0, native state/action 17/6, and 5/5 FPS. The selected paired evaluation mean was 3540.534650 (population SD 826.720798); 16 of 20 returns were strictly above 3000. P profile SHA256 `b5409f8578bec62e2c5232c3916b3ddd2e0e9494031259265540ba6aef882e5e` (publication revision `e1a9bdd566b96d5da4703f769d40eab656af491e`). + +Selected training-best scored 3540.535 with 16/20 returns >3000. Final reached 10002432 environment steps and satisfies the 10M budget; the selected training-best is below that budget. + +Only the selected inference model, `train_step`, and `env_steps` are retained; CPU reload confirmed every model tensor bit-identical. Optimizer, RNG, W&B cache, credentials, VLA artifacts, and demonstration data are excluded. See `verification.json` for the original audit and `provenance.json` for source/export hashes and any separate data publication. diff --git a/latency-aware/walker2d/small-policy/sample-factory-qwenpi_v3-rtx3090-v1/source/provenance.json b/latency-aware/walker2d/small-policy/sample-factory-qwenpi_v3-rtx3090-v1/source/provenance.json new file mode 100644 index 0000000000000000000000000000000000000000..ee081053585ebfb35f74ed5f34c6686f628eb0d4 --- /dev/null +++ b/latency-aware/walker2d/small-policy/sample-factory-qwenpi_v3-rtx3090-v1/source/provenance.json @@ -0,0 +1,50 @@ +{ + "state": "SELECTED_TEACHER_INFERENCE_EXPORT_CPU_RELOAD_VERIFIED", + "task": "walker2d", + "selected": "training_best", + "source_checkpoint": "best_000018928_9691136_reward_4148.139.pth", + "source_checkpoint_sha256": "3d51959c7d65d8e9d1ec34cfc4a8dc7741abf9726439512bc7a0442fd265044e", + "export_checkpoint": "checkpoint_p0/selected.pth", + "export_checkpoint_sha256": "c9164f8e440bd6ac8b41e47f493dc2eea6945bb7520bbc08e851870fda2ac21f", + "retained_keys": [ + "model", + "train_step", + "env_steps" + ], + "removed_keys": [ + "best_performance", + "curr_lr", + "optimizer" + ], + "model_tensors_bit_identical": true, + "model_tensor_count": 15, + "selected_train_step": 18928, + "selected_env_steps": 9691136, + "actual_final_train_step": 19536, + "actual_final_env_steps": 10002432, + "budget_sample_factory_env_steps": 10000000, + "seed": 0, + "native_state_dim": 17, + "native_action_dim": 6, + "profile_sha256": "b5409f8578bec62e2c5232c3916b3ddd2e0e9494031259265540ba6aef882e5e", + "profile_publication_revision": "e1a9bdd566b96d5da4703f769d40eab656af491e", + "audit_state": "TEACHER_FULL_TRACE_AUDIT_PASSED", + "audit_gate": "PASSED", + "selected_mean": 3540.5346502438897, + "selected_population_std": 826.7207979036164, + "selected_strict_gt3000_episodes": 16, + "data": { + "state": "PUBLISHED_SEPARATELY", + "revision": "42d03372359f14ab830b1dcfebaf15986aa0e984", + "accepted": 100, + "attempts": 118, + "rows": 98708, + "gate": "episode_raw_return >3000" + }, + "source_sha256": { + "config.json": "38016732c3acc95545e55085a9c27b9c9594c235efcec0785d385def301b9036", + "teacher.yaml": "ce72f655f96f9bf53bb26bc748f9646d71d5732fec76bd201bb2155f5c1a8be5", + "selection.json": "8c66d50b25b1a79d8dfe3841604a3ca9a5b57816fdfa7be0760a0de936f10f05", + "verification.json": "20a4dde06a7ab677d8cb59116b3cb69d75204aa0fc463e8c48511a69288920e2" + } +} diff --git a/latency-aware/walker2d/small-policy/sample-factory-qwenpi_v3-rtx3090-v1/teacher.yaml b/latency-aware/walker2d/small-policy/sample-factory-qwenpi_v3-rtx3090-v1/teacher.yaml new file mode 100644 index 0000000000000000000000000000000000000000..e31e2aaeec9b4c5938d97757dd0f8680f39566b2 --- /dev/null +++ b/latency-aware/walker2d/small-policy/sample-factory-qwenpi_v3-rtx3090-v1/teacher.yaml @@ -0,0 +1,162 @@ +experiment: + name: pi05_walker2d_profile_appo_h1_5fps_g128_20260921 + seed: 0 +backend: + type: sample_factory + algo: APPO + device: cuda + train_dir: ${PI05_RUN_DIR}/profile_latency/teacher/checkpoints + restart_behavior: overwrite + run_mode: train + extra_args: + - --wandb_user + - dongqianyu99-zhejiang-university +executor: + mode: simulated +env: + name: gymnasium + task_name: walker2d + env_id: LatencyBench/Walker2dContinuous-v0 + registration_imports: + - latency_bench.envs.gymnasium_walker2d + make_kwargs: + base_env_id: Walker2d-v4 + render_mode: rgb_array + base_make_kwargs: + forward_reward_weight: 1.0 + ctrl_cost_weight: 0.001 + healthy_reward: 1.0 + terminate_when_unhealthy: true + reset_noise_scale: 0.005 + exclude_current_positions_from_observation: true + env_fps: 5 + obs_fps: 5 + frame_stack: 1 + action_space: + type: box + labels: + - right_thigh_torque + - right_leg_torque + - right_foot_torque + - left_thigh_torque + - left_leg_torque + - left_foot_torque + low: + - -1.0 + - -1.0 + - -1.0 + - -1.0 + - -1.0 + - -1.0 + high: + - 1.0 + - 1.0 + - 1.0 + - 1.0 + - 1.0 + - 1.0 + dtype: float32 + noop_action: + - 0.0 + - 0.0 + - 0.0 + - 0.0 + - 0.0 + - 0.0 + base_prompt: Move the Walker2d robot forward while keeping its torso upright. Predict + six continuous torques in [-1, 1] ordered as right thigh, right leg, right foot, + left thigh, left leg, and left foot. + state_space: + labels: + - torso_height + - torso_angle + - right_thigh_angle + - right_leg_angle + - right_foot_angle + - left_thigh_angle + - left_leg_angle + - left_foot_angle + - torso_x_velocity + - torso_z_velocity + - torso_angular_velocity + - right_thigh_angular_velocity + - right_leg_angular_velocity + - right_foot_angular_velocity + - left_thigh_angular_velocity + - left_leg_angular_velocity + - left_foot_angular_velocity +latency: + method: iid + profile_path: ${PI05_RUN_DIR}/profile_latency/profile/profile.json + profile_worker_slot: 0 + seed: 0 + add_latency_info: false +scheduler: + hold_policy: hold + ordering_policy: issue_order_fifo +policy: + type: random +training: + train_for_env_steps: 10000000 + num_workers: 8 + num_envs_per_worker: 8 + worker_num_splits: 2 + num_policies: 1 + batch_size: 1024 + rollout: 64 + recurrence: 1 + num_epochs: 2 + num_batches_per_epoch: 4 + max_policy_lag: 300 + learning_rate: 0.00295 + gamma: 0.99 + gae_lambda: 0.95 + ppo_clip_ratio: 0.2 + ppo_clip_value: 0.2 + value_loss_coeff: 1.3 + max_grad_norm: 3.5 + exploration_loss_coeff: 0.0 + async_rl: false + batched_sampling: false + use_rnn: false + encoder_mlp_layers: + - 64 + - 64 + normalize_input: true + normalize_returns: true + save_every_sec: 600 + keep_checkpoints: 5 + lr_schedule: linear_decay + kl_loss_coeff: 0.1 + serial_mode: false + nonlinearity: tanh + adaptive_stddev: false + policy_initialization: torch_default + initial_stddev: 1.0 + shuffle_minibatches: false + value_bootstrap: true +evaluation: + eval_interval_steps: null + eval_episodes: 5 + eval_parallel_envs: 1 + eval_max_steps: 1000 + eval_deterministic: true +logging: + output_dir: ${PI05_RUN_DIR}/profile_latency/teacher/training + video: + enabled: false + num_bins: 1 + save_step_records: false + save_action_records: false + save_latency_records: false + wandb_project: latency-sensitive-bench + wandb_group: pi05-profile-teachers-h1-5fps-20260921 + wandb_name: pi05_walker2d_profile_appo_h1_5fps_g128_20260921 + wandb_job_type: profile_teacher + wandb_tags: + - walker2d + - APPO + - Pi05 + - measured-IID-profile + - H1 + - 5FPS diff --git a/latency-aware/walker2d/small-policy/sample-factory-qwenpi_v3-rtx3090-v1/verification.json b/latency-aware/walker2d/small-policy/sample-factory-qwenpi_v3-rtx3090-v1/verification.json new file mode 100644 index 0000000000000000000000000000000000000000..cb5eab8ece807108af0090b09c1e02532fd181d1 --- /dev/null +++ b/latency-aware/walker2d/small-policy/sample-factory-qwenpi_v3-rtx3090-v1/verification.json @@ -0,0 +1,473 @@ +{ + "state": "TEACHER_FULL_TRACE_AUDIT_PASSED", + "verified_at": "2026-09-21T14:35:27.633103+00:00", + "task": "walker2d", + "P_profile_sha256": "b5409f8578bec62e2c5232c3916b3ddd2e0e9494031259265540ba6aef882e5e", + "P_publication_revision": "e1a9bdd566b96d5da4703f769d40eab656af491e", + "actual_final_env_steps": 10002432, + "actual_final_train_step": 19536, + "audited_evaluation_episodes": 50, + "selection": { + "selected": "training_best", + "selection_policy": "higher_mean_over_paired_20; ties_select_final", + "candidates": { + "final": { + "checkpoint": "${PI05_RUN_DIR}/profile_latency/teacher/checkpoints/pi05_walker2d_profile_appo_h1_5fps_g128_20260921/checkpoint_p0/checkpoint_000019536_10002432.pth", + "sha256": "e157a7be891b0b498028293fc7be14eda97cd41b485220f558ea3d5a88c77171", + "train_step": 19536, + "env_steps": 10002432, + "episodes": 20, + "seeds": [ + 104771, + 104772, + 104773, + 104774, + 104775, + 104776, + 104777, + 104778, + 104779, + 104780, + 104781, + 104782, + 104783, + 104784, + 104785, + 104786, + 104787, + 104788, + 104789, + 104790 + ], + "returns": [ + 4337.574338552904, + 2124.168087172507, + 3803.271661035685, + 4247.230688017356, + 1180.6139208871832, + 1196.3837779316377, + 3778.6057355245894, + 3562.7070253642705, + 1156.377364775773, + 4000.2209267455473, + 1196.9960923429844, + 4196.1107606718415, + 1161.7798657840565, + 1163.344082665548, + 4193.403531977741, + 2077.0328874033744, + 2716.3877118621144, + 1142.383303495549, + 1197.1537112200667, + 2535.48704209563 + ], + "mean": 2548.361625776318, + "population_std": 1289.7687169980788, + "strict_gt3000": 8 + }, + "training_best": { + "checkpoint": "${PI05_RUN_DIR}/profile_latency/teacher/checkpoints/pi05_walker2d_profile_appo_h1_5fps_g128_20260921/checkpoint_p0/best_000018928_9691136_reward_4148.139.pth", + "sha256": "3d51959c7d65d8e9d1ec34cfc4a8dc7741abf9726439512bc7a0442fd265044e", + "train_step": 18928, + "env_steps": 9691136, + "episodes": 20, + "seeds": [ + 104771, + 104772, + 104773, + 104774, + 104775, + 104776, + 104777, + 104778, + 104779, + 104780, + 104781, + 104782, + 104783, + 104784, + 104785, + 104786, + 104787, + 104788, + 104789, + 104790 + ], + "returns": [ + 2454.017948041446, + 3517.5093910555765, + 4443.205483476128, + 4155.968346173417, + 3904.8882905894393, + 1210.1385452181764, + 3550.0328741089725, + 1841.8260301793923, + 4000.010779902804, + 3974.0655123428087, + 2544.925200849052, + 3898.4951632085545, + 3853.089408732446, + 4233.270564039822, + 3649.040265741407, + 3954.793190236864, + 3944.9064060851774, + 3956.211685978782, + 3755.029851073333, + 3969.2680678441934 + ], + "mean": 3540.5346502438897, + "population_std": 826.7207979036164, + "strict_gt3000": 16 + } + }, + "selected_checkpoint_sha256": "3d51959c7d65d8e9d1ec34cfc4a8dc7741abf9726439512bc7a0442fd265044e", + "completed_at_utc": "2026-09-21T14:08:56.270953+00:00" + }, + "evaluations": { + "final": { + "episodes": 20, + "seeds": [ + 104771, + 104772, + 104773, + 104774, + 104775, + 104776, + 104777, + 104778, + 104779, + 104780, + 104781, + 104782, + 104783, + 104784, + 104785, + 104786, + 104787, + 104788, + 104789, + 104790 + ], + "returns": [ + 4337.574338552904, + 2124.168087172507, + 3803.271661035685, + 4247.230688017356, + 1180.6139208871832, + 1196.3837779316377, + 3778.6057355245894, + 3562.7070253642705, + 1156.377364775773, + 4000.2209267455473, + 1196.9960923429844, + 4196.1107606718415, + 1161.7798657840565, + 1163.344082665548, + 4193.403531977741, + 2077.0328874033744, + 2716.3877118621144, + 1142.383303495549, + 1197.1537112200667, + 2535.48704209563 + ], + "lengths": [ + 1000, + 601, + 1000, + 1000, + 397, + 387, + 1000, + 1000, + 389, + 1000, + 414, + 1000, + 374, + 392, + 1000, + 620, + 808, + 373, + 403, + 785 + ], + "mean_return": 2548.361625776318, + "population_std_return": 1289.768716998079, + "mean_length": 697.15, + "raw_steps": 13943, + "strict_gt3000": 8, + "actions": 12457, + "action_latency_mean_ms": 188.65755549282437, + "action_latency_p95_ms": 209.4810075740261, + "commands_requiring_native_clip": 0, + "invalid_actions": 0, + "dropped_actions": 0, + "submitted_observations": 12457, + "dropped_observations": 1486, + "observation_drop_fraction": 0.10657677687728609, + "config_sha256": "318b6a23b6ad09082fa52378a5527fe11bffa50d6b82958a23217055cf09dcac", + "state_contract": "Native Walker2d state17 and action6 are bound to the exact teacher/evaluation configuration; no per-step state vector is recorded." + }, + "training_best": { + "episodes": 20, + "seeds": [ + 104771, + 104772, + 104773, + 104774, + 104775, + 104776, + 104777, + 104778, + 104779, + 104780, + 104781, + 104782, + 104783, + 104784, + 104785, + 104786, + 104787, + 104788, + 104789, + 104790 + ], + "returns": [ + 2454.017948041446, + 3517.5093910555765, + 4443.205483476128, + 4155.968346173417, + 3904.8882905894393, + 1210.1385452181764, + 3550.0328741089725, + 1841.8260301793923, + 4000.010779902804, + 3974.0655123428087, + 2544.925200849052, + 3898.4951632085545, + 3853.089408732446, + 4233.270564039822, + 3649.040265741407, + 3954.793190236864, + 3944.9064060851774, + 3956.211685978782, + 3755.029851073333, + 3969.2680678441934 + ], + "lengths": [ + 588, + 1000, + 1000, + 1000, + 1000, + 392, + 1000, + 480, + 1000, + 1000, + 773, + 1000, + 879, + 1000, + 1000, + 1000, + 1000, + 1000, + 1000, + 1000 + ], + "mean_return": 3540.5346502438892, + "population_std_return": 826.7207979036165, + "mean_length": 905.6, + "raw_steps": 18112, + "strict_gt3000": 16, + "actions": 16208, + "action_latency_mean_ms": 188.587423517079, + "action_latency_p95_ms": 209.24617030890124, + "commands_requiring_native_clip": 0, + "invalid_actions": 0, + "dropped_actions": 0, + "submitted_observations": 16208, + "dropped_observations": 1904, + "observation_drop_fraction": 0.10512367491166077, + "config_sha256": "cbf847662b86166c3543091ed3de91eb6bb36afe35fedc326a4a2ffdb757f64f", + "state_contract": "Native Walker2d state17 and action6 are bound to the exact teacher/evaluation configuration; no per-step state vector is recorded." + } + }, + "E10": { + "episodes": 10, + "seeds": [ + 104771, + 104772, + 104773, + 104774, + 104775, + 104776, + 104777, + 104778, + 104779, + 104780 + ], + "returns": [ + 2454.017948041446, + 3517.5093910555765, + 4443.205483476128, + 4155.968346173417, + 3904.8882905894393, + 1210.1385452181764, + 3550.0328741089725, + 1841.8260301793923, + 4000.010779902804, + 3974.0655123428087 + ], + "lengths": [ + 588, + 1000, + 1000, + 1000, + 1000, + 392, + 1000, + 480, + 1000, + 1000 + ], + "mean_return": 3305.166320108816, + "population_std_return": 1032.90404014517, + "mean_length": 846.0, + "raw_steps": 8460, + "strict_gt3000": 7, + "actions": 7551, + "action_latency_mean_ms": 188.58387721197627, + "action_latency_p95_ms": 209.47112560796126, + "commands_requiring_native_clip": 0, + "invalid_actions": 0, + "dropped_actions": 0, + "submitted_observations": 7551, + "dropped_observations": 909, + "observation_drop_fraction": 0.1074468085106383, + "config_sha256": "58e7857f1995a34924ed156b9718f1c462d5c0a733c2ae0f30e211f58051aa4c", + "state_contract": "Native Walker2d state17 and action6 are bound to the exact teacher/evaluation configuration; no per-step state vector is recorded." + }, + "probe": { + "attempts": 12, + "accepted": 12, + "rejected": 0, + "source_state_sha256": "266f2ec22ca31030bcc2e2161543314b579d178a1f45ff428b3278aa0bdead58", + "accepted_specs": [ + { + "attempt_idx": 0, + "episode_idx": 0, + "episode_length": 1000, + "episode_raw_frames": 1000, + "episode_raw_return": 4652.138299047947, + "seed": 0, + "split": "train" + }, + { + "attempt_idx": 1, + "episode_idx": 1, + "episode_length": 1000, + "episode_raw_frames": 1000, + "episode_raw_return": 4344.816579580307, + "seed": 1, + "split": "val" + }, + { + "attempt_idx": 2, + "episode_idx": 2, + "episode_length": 1000, + "episode_raw_frames": 1000, + "episode_raw_return": 4667.5336265563965, + "seed": 2, + "split": "train" + }, + { + "attempt_idx": 3, + "episode_idx": 3, + "episode_length": 1000, + "episode_raw_frames": 1000, + "episode_raw_return": 4701.752601087093, + "seed": 3, + "split": "val" + }, + { + "attempt_idx": 4, + "episode_idx": 4, + "episode_length": 1000, + "episode_raw_frames": 1000, + "episode_raw_return": 4646.38482773304, + "seed": 4, + "split": "train" + }, + { + "attempt_idx": 5, + "episode_idx": 5, + "episode_length": 1000, + "episode_raw_frames": 1000, + "episode_raw_return": 4742.2653895020485, + "seed": 5, + "split": "train" + }, + { + "attempt_idx": 6, + "episode_idx": 6, + "episode_length": 1000, + "episode_raw_frames": 1000, + "episode_raw_return": 4722.492495238781, + "seed": 6, + "split": "train" + }, + { + "attempt_idx": 7, + "episode_idx": 7, + "episode_length": 1000, + "episode_raw_frames": 1000, + "episode_raw_return": 4588.488040447235, + "seed": 7, + "split": "val" + }, + { + "attempt_idx": 8, + "episode_idx": 8, + "episode_length": 939, + "episode_raw_frames": 939, + "episode_raw_return": 4499.97767752409, + "seed": 8, + "split": "train" + }, + { + "attempt_idx": 9, + "episode_idx": 9, + "episode_length": 1000, + "episode_raw_frames": 1000, + "episode_raw_return": 4676.904938817024, + "seed": 9, + "split": "train" + }, + { + "attempt_idx": 10, + "episode_idx": 10, + "episode_length": 1000, + "episode_raw_frames": 1000, + "episode_raw_return": 4704.461377084255, + "seed": 10, + "split": "train" + }, + { + "attempt_idx": 11, + "episode_idx": 11, + "episode_length": 1000, + "episode_raw_frames": 1000, + "episode_raw_return": 4679.474701523781, + "seed": 11, + "split": "train" + } + ] + }, + "strict_gate": "episode_raw_return > 3000", + "E10_caveat": "Repeats the first ten selection seeds; not an independent gate.", + "training_unit_exit_claim": false, + "gate": "PASSED" +} diff --git a/latency-aware/walker2d/vla/starvla-qwengr00t-h1/walker2d_profile_gr00t_h1_5fps_2h200_20260917/F-summary.json b/latency-aware/walker2d/vla/starvla-qwengr00t-h1/walker2d_profile_gr00t_h1_5fps_2h200_20260917/F-summary.json new file mode 100644 index 0000000000000000000000000000000000000000..18dfecf3d1185deaf8b3b9fb1eb07236a70490b4 --- /dev/null +++ b/latency-aware/walker2d/vla/starvla-qwengr00t-h1/walker2d_profile_gr00t_h1_5fps_2h200_20260917/F-summary.json @@ -0,0 +1,163 @@ +{ + "verified_at": "2026-09-17T21:54:00.894032+00:00", + "task": "walker2d", + "B": { + "episodes": 20, + "mean_return": 63.24147456221736, + "population_std": 6.584119944585635, + "mean_length": 61.15, + "mean_episode_latency_ms": 107.04820905731547, + "observation_drops": 0, + "submitted_observations": 1223, + "action_drops": 0, + "invalid_actions": 0, + "metrics_sha256": "0c585c07d028987d60f03f5cad990f7b424e97917d139fa66f44e6573f75ee34" + }, + "F": { + "episodes": 20, + "mean_return": 852.8724033441786, + "population_std": 357.102322157348, + "mean_length": 250.6, + "mean_episode_latency_ms": 107.06145728669419, + "observation_drops": 0, + "submitted_observations": 5012, + "action_drops": 0, + "invalid_actions": 0, + "metrics_sha256": "456db545a2fb07edfe7a339761b34782dd5a2987f3f28eb0e241e550c7aa6ced" + }, + "paired": [ + { + "seed": 42, + "B": 66.16242032595262, + "F": 991.3639332213364, + "delta": 925.2015128953838 + }, + { + "seed": 43, + "B": 60.03168286501273, + "F": 369.8818098433932, + "delta": 309.8501269783805 + }, + { + "seed": 44, + "B": 64.47357571737881, + "F": 720.6103530557076, + "delta": 656.1367773383288 + }, + { + "seed": 45, + "B": 59.60303463811989, + "F": 1150.3193138698, + "delta": 1090.7162792316801 + }, + { + "seed": 46, + "B": 61.890443606636744, + "F": 310.6481592092358, + "delta": 248.75771560259906 + }, + { + "seed": 47, + "B": 61.404429063445264, + "F": 1068.212039574912, + "delta": 1006.8076105114668 + }, + { + "seed": 48, + "B": 71.25021337481104, + "F": 509.5536777031986, + "delta": 438.3034643283876 + }, + { + "seed": 49, + "B": 64.32829050611694, + "F": 1316.4247834069251, + "delta": 1252.0964929008082 + }, + { + "seed": 50, + "B": 59.65748458065071, + "F": 718.9444266379952, + "delta": 659.2869420573445 + }, + { + "seed": 51, + "B": 78.19331421262474, + "F": 1182.669513035368, + "delta": 1104.4761988227433 + }, + { + "seed": 52, + "B": 69.53756289500063, + "F": 1241.9130556220862, + "delta": 1172.3754927270857 + }, + { + "seed": 53, + "B": 52.28806362435964, + "F": 711.9996561729512, + "delta": 659.7115925485915 + }, + { + "seed": 54, + "B": 61.4441390040874, + "F": 306.83449127304596, + "delta": 245.39035226895857 + }, + { + "seed": 55, + "B": 78.21931462223985, + "F": 1222.080977245407, + "delta": 1143.8616626231671 + }, + { + "seed": 56, + "B": 61.4247094298816, + "F": 800.6216178219371, + "delta": 739.1969083920555 + }, + { + "seed": 57, + "B": 63.617009505124926, + "F": 373.15968866798323, + "delta": 309.5426791628583 + }, + { + "seed": 58, + "B": 58.18109265657494, + "F": 954.8027340055005, + "delta": 896.6216413489256 + }, + { + "seed": 59, + "B": 54.21753227541805, + "F": 787.7657032501974, + "delta": 733.5481709747793 + }, + { + "seed": 60, + "B": 59.20549653493986, + "F": 741.1351754045119, + "delta": 681.9296788695721 + }, + { + "seed": 61, + "B": 59.69968180597084, + "F": 1578.506957862078, + "delta": 1518.8072760561072 + } + ], + "wins": 20, + "ties": 0, + "losses": 0, + "mean_difference": 789.6309287819612, + "F_step_interval_ms": { + "mean": 200.00296332158868, + "min": 198.95315496250987, + "max": 201.21660106815398 + }, + "F_revision": "af4f38060b0374c5684c43f22cd504257cf2bffd", + "F_checkpoint_sha256": "1859cec5319db37b97a62d8190f081771f38978fa2ade39cfb82b155ac56b83b", + "protocol": "20 matched seeds42-61;5FPS;cap1000;measured-only;same B/F runtime with model and prompt0\u21921 changes", + "interpretation": "Compare measured outcomes and realized latencies; no equal-latency causal claim and no teacher gate applied to VLA." +} diff --git a/latency-aware/walker2d/vla/starvla-qwengr00t-h1/walker2d_profile_gr00t_h1_5fps_2h200_20260917/F-verification.json b/latency-aware/walker2d/vla/starvla-qwengr00t-h1/walker2d_profile_gr00t_h1_5fps_2h200_20260917/F-verification.json new file mode 100644 index 0000000000000000000000000000000000000000..186633358011ab5373a4185f0c67777886b4d3ee --- /dev/null +++ b/latency-aware/walker2d/vla/starvla-qwengr00t-h1/walker2d_profile_gr00t_h1_5fps_2h200_20260917/F-verification.json @@ -0,0 +1,6 @@ +{ + "state": "F20_AND_MATCHED_COMPARISON_VERIFIED", + "summary": "/workspace/lzj/latency-sensitive-bench/runs/walker2d/gr00t_h1_5fps_20260917/F-summary.json", + "steps": 5012, + "verified_at": "2026-09-17T21:54:00.894032+00:00" +} diff --git a/latency-aware/walker2d/vla/starvla-qwengr00t-h1/walker2d_profile_gr00t_h1_5fps_2h200_20260917/README.md b/latency-aware/walker2d/vla/starvla-qwengr00t-h1/walker2d_profile_gr00t_h1_5fps_2h200_20260917/README.md new file mode 100644 index 0000000000000000000000000000000000000000..712962129486507a29c83c4aa6d82a3f8e9d365c --- /dev/null +++ b/latency-aware/walker2d/vla/starvla-qwengr00t-h1/walker2d_profile_gr00t_h1_5fps_2h200_20260917/README.md @@ -0,0 +1,31 @@ +# walker2d / qwengr00t + +Training condition: `latency-aware`. Run: `walker2d_profile_gr00t_h1_5fps_2h200_20260917`. + +Copied from [Standard-Pipeline](https://huggingface.co/latency-sensitive-bench/Standard-Pipeline/tree/2208875f92b21e3f8ec2363a30a6705dfe56cc04/walker2d/profile_latency/GR00T) at `2208875f92b21e3f8ec2363a30a6705dfe56cc04`. + +- Checkpoint: `checkpoints/model.pt` +- Selection: only published VLA checkpoint +- Checkpoint SHA256: `1859cec5319db37b97a62d8190f081771f38978fa2ade39cfb82b155ac56b83b` +- Configuration, FPS, actions, normalization and latency semantics retain their source values. +- This import does not run evaluations or change training/evaluation/acceptance status. + +Action horizon is 1. Use `config.yaml` and `dataset_statistics.json` with the matching +prepared backbone. `task_contract.json` is extracted from the original training configuration. +Original machine paths in historical configurations are provenance: bind checkpoint, +backbone, manifest and profile paths to downloaded local files when running inference. + +See [the original model notes](source/README.md) for experiment results and limitations. + +[Original provenance](source/provenance.json) retains dataset references and recorded experiment evidence. + +## Download + +```python +from huggingface_hub import snapshot_download +prefix = "latency-aware/walker2d/vla/starvla-qwengr00t-h1/walker2d_profile_gr00t_h1_5fps_2h200_20260917" +snapshot_download("latency-sensitive-bench/benchmark-models", revision="", + allow_patterns=[prefix + "/**"]) +``` + +Use the verified import commit listed in the repository's import report. diff --git a/latency-aware/walker2d/vla/starvla-qwengr00t-h1/walker2d_profile_gr00t_h1_5fps_2h200_20260917/checkpoints/model.pt b/latency-aware/walker2d/vla/starvla-qwengr00t-h1/walker2d_profile_gr00t_h1_5fps_2h200_20260917/checkpoints/model.pt new file mode 100644 index 0000000000000000000000000000000000000000..b5d18bb10eee83e9ffedbb1eee7c28a5f406218e --- /dev/null +++ b/latency-aware/walker2d/vla/starvla-qwengr00t-h1/walker2d_profile_gr00t_h1_5fps_2h200_20260917/checkpoints/model.pt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1859cec5319db37b97a62d8190f081771f38978fa2ade39cfb82b155ac56b83b +size 9976860963 diff --git a/latency-aware/walker2d/vla/starvla-qwengr00t-h1/walker2d_profile_gr00t_h1_5fps_2h200_20260917/config.full.yaml b/latency-aware/walker2d/vla/starvla-qwengr00t-h1/walker2d_profile_gr00t_h1_5fps_2h200_20260917/config.full.yaml new file mode 100644 index 0000000000000000000000000000000000000000..4d446ebb593be02ef5333155ef02f538ef3efe03 --- /dev/null +++ b/latency-aware/walker2d/vla/starvla-qwengr00t-h1/walker2d_profile_gr00t_h1_5fps_2h200_20260917/config.full.yaml @@ -0,0 +1,261 @@ +framework: + name: QwenGR00T + qwenvl: + base_vlm: ${oc.env:GR00T_BACKBONE_DIR} + attn_implementation: flash_attention_2 + vl_hidden_dim: 2048 + enable_gradient_checkpointing: true + action_model: + action_model_type: DiT-B + action_hidden_dim: 1024 + hidden_size: 1024 + add_pos_embed: true + max_seq_len: 1024 + action_dim: 6 + state_dim: 17 + task_objective: null + action_horizon: 1 + repeated_diffusion_steps: 8 + noise_beta_alpha: 1.5 + noise_beta_beta: 1.0 + noise_s: 0.999 + num_timestep_buckets: 1000 + num_inference_timesteps: 4 + num_target_vision_tokens: 32 + diffusion_model_cfg: + cross_attention_dim: 2560 + dropout: 0.2 + final_dropout: true + interleave_self_attention: true + norm_type: ada_norm + num_layers: 16 + output_dim: 1024 + positional_embeddings: null + action_env_dim: 6 + future_action_window_size: 0 + past_action_window_size: 0 +datasets: + vla_data: + dataset_py: lerobot_datasets + include_state: true + data_root_dir: / + data_mix: train + eval_data_mix: validation + custom_mixtures_path: /mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/vla/mixture.json + action_type: continuous + sequential_step_sampling: true + eval_sequential_step_sampling: null + num_workers: 4 + eval_num_workers: 0 + prefetch_factor: 2 + persistent_workers: true + pin_memory: true + shuffle: true + action_balance: + enabled: false + strategy: balanced_epoch + action_key: action_id + target_flap_fraction: 0.3 + noop_id: 0 + flap_id: 1 + latency_curriculum: + enabled: false + strategy: exclusive + latencies: null + phase_steps: null + phase_distributions: null + new_latency_passes: 1.0 + replay_passes: 0.25 + target_total_passes: 2.0 + final_equalization: true + step_budget_mode: auto + eval_at_phase_end: false + save_at_phase_end: false + computed_plan: null + per_device_batch_size: 32 + load_all_data_for_training: true + num_obs_frames: 1 + image_mode: single + prompt_mode: raw + stitch_grid: + - 2 + - 2 + obs_image_size: + - 224 + - 224 + video_backend: pyav + lerobot_version: v2.0 + gymnasium_task_contract: + action_space: + dtype: float32 + high: + - 1.0 + - 1.0 + - 1.0 + - 1.0 + - 1.0 + - 1.0 + labels: + - right_thigh_torque + - right_leg_torque + - right_foot_torque + - left_thigh_torque + - left_leg_torque + - left_foot_torque + low: + - -1.0 + - -1.0 + - -1.0 + - -1.0 + - -1.0 + - -1.0 + type: box + base_prompt: Move the Walker2d robot forward while keeping its torso upright. + Predict six continuous torques in [-1, 1] ordered as right thigh, right leg, + right foot, left thigh, left leg, and left foot. + env_fps: 5.0 + env_id: LatencyBench/Walker2dContinuous-v0 + frame_stack: 1 + make_kwargs: + base_env_id: Walker2d-v4 + base_make_kwargs: + ctrl_cost_weight: 0.001 + exclude_current_positions_from_observation: true + forward_reward_weight: 1.0 + healthy_reward: 1.0 + reset_noise_scale: 0.005 + terminate_when_unhealthy: true + render_mode: rgb_array + noop_action: + - 0.0 + - 0.0 + - 0.0 + - 0.0 + - 0.0 + - 0.0 + obs_fps: 5.0 + registration_imports: + - latency_bench.envs.gymnasium_walker2d + state_space: + labels: + - torso_height + - torso_angle + - right_thigh_angle + - right_leg_angle + - right_foot_angle + - left_thigh_angle + - left_leg_angle + - left_foot_angle + - torso_x_velocity + - torso_z_velocity + - torso_angular_velocity + - right_thigh_angular_velocity + - right_leg_angular_velocity + - right_foot_angular_velocity + - left_thigh_angular_velocity + - left_leg_angular_velocity + - left_foot_angular_velocity + task_name: walker2d + active_action_dim: 6 +dataset: + source_hf: '' + config_name: null + source_subdir: null + converted_name: flappy_train + single_source_hf: '' + mixed_source_hf: '' + single_converted_name: flappy_train + mixed_converted_name: flappy_mixed_latency_train + single_latency_filter: null + mixed_latency_filter: null + force_download: false + setup_force: false + skip_verification: false + target_latency_unit: observation_steps + verify_rows: 200 + max_episodes: null + episodes_per_latency: null + latency_filter: null + debug_subset: + enabled: false + max_episodes: 5 + suffix: debug +base_model: + repo_id: Qwen/Qwen3-VL-4B-Instruct +initialization: + checkpoint_local_dir: null + checkpoint_hf_repo_id: null + checkpoint_filename: null +trainer: + max_train_steps: 5000 + num_warmup_steps: 100 + save_interval: 500 + eval_interval: 500 + eval_num_batches: 200 + per_latency_eval_num_batches: null + eval_action_classification: false + eval_action_classification_interval: null + cc_f1_tolerance: 1 + learning_rate: + base: 2.0e-05 + qwen_vl_interface: 1.0e-05 + action_model: 0.0001 + lr_scheduler_type: cosine_with_min_lr + scheduler_specific_kwargs: + min_lr: 1.0e-06 + freeze_modules: '' + freeze_vit: false + freeze_tied_embedding: false + freeze_llm_layers: [] + loss_scale: + vla: 1.0 + vlm: 0.1 + max_grad_norm: 1.0 + weight_decay: 0.0 + logging_frequency: 1 + profile_timing: + enabled: false + log_interval: 10 + gradient_clipping: 1.0 + gradient_accumulation_steps: 1 + distributed_backend: deepspeed + is_resume: false + pretrained_checkpoint: null + resume_step: 0 + reload_modules: null + optimizer: + name: AdamW + betas: + - 0.9 + - 0.95 + eps: 1.0e-08 + weight_decay: 1.0e-08 + fused: true + save_format: pt +run_id: walker2d_profile_gr00t_h1_5fps_2h200_20260917 +run_root_dir: /mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/vla/training +seed: 42 +is_debug: false +version_id: '0.21' +wandb_project: latency-sensitive-bench +wandb_entity: dongqianyu99-zhejiang-university +wandb_name: walker2d_profile_gr00t_h1_5fps_2h200_20260917 +wandb_group: gr00t-six-env-h1-5fps +wandb_tags: +- walker2d +- profile_latency +- GR00T +- h1 +training_latency_condition: profile_latency +backbone: + repo_id: Qwen/Qwen3-VL-4B-Instruct + revision: ebb281ec70b05090aa6165b016eac8ec08e71b17 +checkpoint: + save_best_model: false + save_final_model: true + save_pt_file: true + save_training_state: true + local: + keep_last_n: 1 +config_yaml: /mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/train.yaml +output_dir: /mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/vla/training/walker2d_profile_gr00t_h1_5fps_2h200_20260917 diff --git a/latency-aware/walker2d/vla/starvla-qwengr00t-h1/walker2d_profile_gr00t_h1_5fps_2h200_20260917/config.yaml b/latency-aware/walker2d/vla/starvla-qwengr00t-h1/walker2d_profile_gr00t_h1_5fps_2h200_20260917/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..f386e6c63d723d3c59581f2f4fa22bd28ba691e6 --- /dev/null +++ b/latency-aware/walker2d/vla/starvla-qwengr00t-h1/walker2d_profile_gr00t_h1_5fps_2h200_20260917/config.yaml @@ -0,0 +1,138 @@ +framework: + name: QwenGR00T + qwenvl: + base_vlm: ${oc.env:GR00T_BACKBONE_DIR} + attn_implementation: flash_attention_2 + vl_hidden_dim: 2048 + enable_gradient_checkpointing: true + action_model: + action_model_type: DiT-B + action_hidden_dim: 1024 + hidden_size: 1024 + add_pos_embed: true + max_seq_len: 1024 + action_dim: 6 + state_dim: 17 + task_objective: null + action_horizon: 1 + repeated_diffusion_steps: 8 + noise_beta_alpha: 1.5 + noise_beta_beta: 1.0 + noise_s: 0.999 + num_timestep_buckets: 1000 + num_inference_timesteps: 4 + num_target_vision_tokens: 32 + diffusion_model_cfg: + cross_attention_dim: 2560 + dropout: 0.2 + final_dropout: true + interleave_self_attention: true + norm_type: ada_norm + num_layers: 16 + output_dim: 1024 + positional_embeddings: null + action_env_dim: 6 + future_action_window_size: 0 + past_action_window_size: 0 +version_id: '0.21' +datasets: + vla_data: + include_state: true + obs_image_size: + - 224 + - 224 + image_mode: single + stitch_grid: + - 2 + - 2 + num_obs_frames: 1 + gymnasium_task_contract: &id001 + action_space: + dtype: float32 + high: + - 1.0 + - 1.0 + - 1.0 + - 1.0 + - 1.0 + - 1.0 + labels: + - right_thigh_torque + - right_leg_torque + - right_foot_torque + - left_thigh_torque + - left_leg_torque + - left_foot_torque + low: + - -1.0 + - -1.0 + - -1.0 + - -1.0 + - -1.0 + - -1.0 + type: box + base_prompt: Move the Walker2d robot forward while keeping its torso upright. + Predict six continuous torques in [-1, 1] ordered as right thigh, right leg, + right foot, left thigh, left leg, and left foot. + env_fps: 5.0 + env_id: LatencyBench/Walker2dContinuous-v0 + frame_stack: 1 + make_kwargs: + base_env_id: Walker2d-v4 + base_make_kwargs: + ctrl_cost_weight: 0.001 + exclude_current_positions_from_observation: true + forward_reward_weight: 1.0 + healthy_reward: 1.0 + reset_noise_scale: 0.005 + terminate_when_unhealthy: true + render_mode: rgb_array + noop_action: + - 0.0 + - 0.0 + - 0.0 + - 0.0 + - 0.0 + - 0.0 + obs_fps: 5.0 + registration_imports: + - latency_bench.envs.gymnasium_walker2d + state_space: + labels: + - torso_height + - torso_angle + - right_thigh_angle + - right_leg_angle + - right_foot_angle + - left_thigh_angle + - left_leg_angle + - left_foot_angle + - torso_x_velocity + - torso_z_velocity + - torso_angular_velocity + - right_thigh_angular_velocity + - right_leg_angular_velocity + - right_foot_angular_velocity + - left_thigh_angular_velocity + - left_leg_angular_velocity + - left_foot_angular_velocity + task_name: walker2d +trainer: + pretrained_checkpoint: null + profile_timing: + enabled: false + log_interval: 10 +backbone: + repo_id: Qwen/Qwen3-VL-4B-Instruct + revision: ebb281ec70b05090aa6165b016eac8ec08e71b17 +rl_games: + model_alias: gr00t + task: gymnasium + action_carrier: native + gymnasium: + task_contract: *id001 + env_eval: + image_size: 224 + frameskip: 1 + image_transform: raw_rgb + prompt_mode: raw diff --git a/latency-aware/walker2d/vla/starvla-qwengr00t-h1/walker2d_profile_gr00t_h1_5fps_2h200_20260917/dataset_statistics.json b/latency-aware/walker2d/vla/starvla-qwengr00t-h1/walker2d_profile_gr00t_h1_5fps_2h200_20260917/dataset_statistics.json new file mode 100644 index 0000000000000000000000000000000000000000..fac195ad9be9609a516dd161c849524c31c22dfd --- /dev/null +++ b/latency-aware/walker2d/vla/starvla-qwengr00t-h1/walker2d_profile_gr00t_h1_5fps_2h200_20260917/dataset_statistics.json @@ -0,0 +1,180 @@ +{ + "new_embodiment": { + "action": { + "mean": [ + 0.9479884505271912, + 0.8967278599739075, + 0.22252006828784943, + 0.42712029814720154, + 0.6054255366325378, + 0.39046671986579895 + ], + "std": [ + 0.22479695081710815, + 0.28594842553138733, + 0.8105338215827942, + 0.6386038661003113, + 0.6382305026054382, + 0.7736172080039978 + ], + "max": [ + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0 + ], + "min": [ + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0 + ], + "q01": [ + -0.12693553328514098, + -0.4622062122821808, + -1.0, + -1.0, + -1.0, + -1.0 + ], + "q99": [ + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0 + ], + "mask": [ + true, + true, + true, + true, + true, + true + ] + }, + "state": { + "mean": [ + 0.080454520881176, + 0.3073728084564209, + 0.5504761934280396, + 0.4364703595638275, + 0.17060819268226624, + 0.4959196150302887, + 0.60721755027771, + 0.274997740983963, + 0.21275795996189117, + -0.10817987471818924, + -0.06906983256340027, + -0.001813000999391079, + -0.0005867783329449594, + -0.051718927919864655, + -0.0353797972202301, + -0.017273033037781715, + -0.006954238750040531 + ], + "std": [ + 0.3848876655101776, + 0.386168897151947, + 0.25338760018348694, + 0.17777718603610992, + 0.5525822639465332, + 0.4858602285385132, + 0.26794859766960144, + 0.5490415692329407, + 0.2725364565849304, + 0.3741426467895508, + 0.26948603987693787, + 0.23391664028167725, + 0.18887001276016235, + 0.6265960931777954, + 0.3519318103790283, + 0.4073410630226135, + 0.6110035181045532 + ], + "max": [ + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0 + ], + "min": [ + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0 + ], + "q01": [ + -0.8464448672533035, + -0.6947578257322311, + -0.28605210840702056, + -0.46256462395191195, + -0.8968109232187271, + -0.7328857761621476, + -0.48083224475383757, + -0.9292185813188553, + -0.5419902771711349, + -0.7907932072877883, + -0.8048153072595596, + -0.686601927280426, + -1.0, + -1.0, + -0.6300148075819015, + -1.0, + -1.0 + ], + "q99": [ + 0.9185441315174102, + 0.9097266328334804, + 0.7716626930236811, + 0.5981208789348602, + 0.8097793257236477, + 0.9137463271617888, + 0.9044068646430968, + 0.9431438720226286, + 0.7647204828262327, + 0.7835678279399867, + 0.5781928598880763, + 0.7853474819660186, + 0.6514471054077148, + 1.0, + 1.0, + 1.0, + 1.0 + ] + }, + "num_transitions": 90000, + "num_trajectories": 90 + } +} \ No newline at end of file diff --git a/latency-aware/walker2d/vla/starvla-qwengr00t-h1/walker2d_profile_gr00t_h1_5fps_2h200_20260917/latency_prompt_map.json b/latency-aware/walker2d/vla/starvla-qwengr00t-h1/walker2d_profile_gr00t_h1_5fps_2h200_20260917/latency_prompt_map.json new file mode 100644 index 0000000000000000000000000000000000000000..ab7cb73c4a6870635a012917737f130eac0545b7 --- /dev/null +++ b/latency-aware/walker2d/vla/starvla-qwengr00t-h1/walker2d_profile_gr00t_h1_5fps_2h200_20260917/latency_prompt_map.json @@ -0,0 +1,7 @@ +{ + "1": { + "prompt": "Move the Walker2d robot forward while keeping its torso upright. Predict six continuous torques in [-1, 1] ordered as right thigh, right leg, right foot, left thigh, left leg, and left foot. Current action latency is 1 raw frames (200.00 ms). The environment runs at 5 FPS and observations are emitted at 5 FPS. Choose the best next action.", + "latency_raw_frames": 1, + "latency_ms": 200.0 + } +} \ No newline at end of file diff --git a/latency-aware/walker2d/vla/starvla-qwengr00t-h1/walker2d_profile_gr00t_h1_5fps_2h200_20260917/manifest.json b/latency-aware/walker2d/vla/starvla-qwengr00t-h1/walker2d_profile_gr00t_h1_5fps_2h200_20260917/manifest.json new file mode 100644 index 0000000000000000000000000000000000000000..8f921aebb55d06a96594b2459279dfca7165e0eb --- /dev/null +++ b/latency-aware/walker2d/vla/starvla-qwengr00t-h1/walker2d_profile_gr00t_h1_5fps_2h200_20260917/manifest.json @@ -0,0 +1,184 @@ +{ + "dataset_name": "walker2d_h1_5fps_profile", + "env_name": "walker2d", + "episodes": 90, + "frames": 90000, + "task_prompts": [ + "Move the Walker2d robot forward while keeping its torso upright. Predict six continuous torques in [-1, 1] ordered as right thigh, right leg, right foot, left thigh, left leg, and left foot. Current action latency is 1 raw frames (200.00 ms). The environment runs at 5 FPS and observations are emitted at 5 FPS. Choose the best next action." + ], + "format": "lerobot_v2_image_parquet", + "source": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/data/raw_profile", + "integration_name": "gymnasium", + "task_name": "walker2d", + "action_layout": "gymnasium_continuous_v1", + "action_labels": [ + "right_thigh_torque", + "right_leg_torque", + "right_foot_torque", + "left_thigh_torque", + "left_leg_torque", + "left_foot_torque" + ], + "carrier_action_labels": [ + "right_thigh_torque", + "right_leg_torque", + "right_foot_torque", + "left_thigh_torque", + "left_leg_torque", + "left_foot_torque" + ], + "action_dim": 6, + "active_action_dim": 6, + "action_carrier": "native", + "rows_unit": "decision_step", + "obs_fps": 5.0, + "obs_stride_raw_frames": 1, + "uses_state": true, + "state_dim": 17, + "state_labels": [ + "torso_height", + "torso_angle", + "right_thigh_angle", + "right_leg_angle", + "right_foot_angle", + "left_thigh_angle", + "left_leg_angle", + "left_foot_angle", + "torso_x_velocity", + "torso_z_velocity", + "torso_angular_velocity", + "right_thigh_angular_velocity", + "right_leg_angular_velocity", + "right_foot_angular_velocity", + "left_thigh_angular_velocity", + "left_leg_angular_velocity", + "left_foot_angular_velocity" + ], + "state_normalization": { + "type": "min_max", + "min": [ + 0.9851876497268677, + -0.1729077845811844, + -0.5815715193748474, + -0.41291648149490356, + -1.1557613611221313, + -1.565748929977417, + -1.5937821865081787, + -1.4450324773788452, + -1.448801040649414, + -2.645611524581909, + -8.647480010986328, + -10.0, + -10.0, + -10.0, + -10.0, + -10.0, + -10.0 + ], + "max": [ + 1.5055257081985474, + 0.6807975769042969, + 0.14722545444965363, + 0.17162981629371643, + 1.3325903415679932, + 0.18669182062149048, + 0.3009708821773529, + 1.3312314748764038, + 6.26270866394043, + 3.290994644165039, + 10.0, + 10.0, + 10.0, + 10.0, + 10.0, + 10.0, + 10.0 + ] + }, + "robot_type": "rl_games_gymnasium", + "latency_prompt_map_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/data/lerobot/walker2d_h1_5fps_profile/latency_prompt_map.json", + "custom_mixtures_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/data/lerobot/_generated_mixtures/walker2d_h1_5fps_profile.json", + "gymnasium_task": { + "action_space": { + "dtype": "float32", + "high": [ + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0 + ], + "labels": [ + "right_thigh_torque", + "right_leg_torque", + "right_foot_torque", + "left_thigh_torque", + "left_leg_torque", + "left_foot_torque" + ], + "low": [ + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0 + ], + "type": "box" + }, + "base_prompt": "Move the Walker2d robot forward while keeping its torso upright. Predict six continuous torques in [-1, 1] ordered as right thigh, right leg, right foot, left thigh, left leg, and left foot.", + "env_fps": 5.0, + "env_id": "LatencyBench/Walker2dContinuous-v0", + "frame_stack": 1, + "make_kwargs": { + "base_env_id": "Walker2d-v4", + "base_make_kwargs": { + "ctrl_cost_weight": 0.001, + "exclude_current_positions_from_observation": true, + "forward_reward_weight": 1.0, + "healthy_reward": 1.0, + "reset_noise_scale": 0.005, + "terminate_when_unhealthy": true + }, + "render_mode": "rgb_array" + }, + "noop_action": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ], + "obs_fps": 5.0, + "registration_imports": [ + "latency_bench.envs.gymnasium_walker2d" + ], + "state_space": { + "labels": [ + "torso_height", + "torso_angle", + "right_thigh_angle", + "right_leg_angle", + "right_foot_angle", + "left_thigh_angle", + "left_leg_angle", + "left_foot_angle", + "torso_x_velocity", + "torso_z_velocity", + "torso_angular_velocity", + "right_thigh_angular_velocity", + "right_leg_angular_velocity", + "right_foot_angular_velocity", + "left_thigh_angular_velocity", + "left_leg_angular_velocity", + "left_foot_angular_velocity" + ] + }, + "task_name": "walker2d" + }, + "validation_dataset_name": "walker2d_h1_5fps_profile__val", + "validation_episodes": 10, + "validation_frames": 10000 +} \ No newline at end of file diff --git a/latency-aware/walker2d/vla/starvla-qwengr00t-h1/walker2d_profile_gr00t_h1_5fps_2h200_20260917/provenance.json b/latency-aware/walker2d/vla/starvla-qwengr00t-h1/walker2d_profile_gr00t_h1_5fps_2h200_20260917/provenance.json new file mode 100644 index 0000000000000000000000000000000000000000..491f5ee7779cf45801abe9bc5141ac42cf4ed9bf --- /dev/null +++ b/latency-aware/walker2d/vla/starvla-qwengr00t-h1/walker2d_profile_gr00t_h1_5fps_2h200_20260917/provenance.json @@ -0,0 +1,34 @@ +{ + "task": "walker2d", + "model": "qwengr00t", + "training_condition": "latency-aware", + "training_run_id": "walker2d_profile_gr00t_h1_5fps_2h200_20260917", + "source": { + "repo_id": "latency-sensitive-bench/Standard-Pipeline", + "revision": "2208875f92b21e3f8ec2363a30a6705dfe56cc04", + "path": "walker2d/profile_latency/GR00T", + "url": "https://huggingface.co/latency-sensitive-bench/Standard-Pipeline/tree/2208875f92b21e3f8ec2363a30a6705dfe56cc04/walker2d/profile_latency/GR00T" + }, + "hf_repo_id": "latency-sensitive-bench/benchmark-models", + "hf_path": "latency-aware/walker2d/vla/starvla-qwengr00t-h1/walker2d_profile_gr00t_h1_5fps_2h200_20260917", + "checkpoint": { + "source_file": "walker2d/profile_latency/GR00T/checkpoints/model.pt", + "source_sha256": "1859cec5319db37b97a62d8190f081771f38978fa2ade39cfb82b155ac56b83b", + "source_bytes": 9976860963, + "file": "checkpoints/model.pt", + "selection_rule": "only published VLA checkpoint", + "method": "copy", + "sha256": "1859cec5319db37b97a62d8190f081771f38978fa2ade39cfb82b155ac56b83b", + "bytes": 9976860963 + }, + "config_source": "walker2d/profile_latency/GR00T/config.yaml", + "task_contract_file": "task_contract.json", + "missing_source_files": [], + "migration": { + "operation": "copy_selected_inference_bundle", + "source_modified": false, + "evaluation_performed": false, + "experiment_status": "preserved_in_source_evidence" + }, + "action_horizon": 1 +} diff --git a/latency-aware/walker2d/vla/starvla-qwengr00t-h1/walker2d_profile_gr00t_h1_5fps_2h200_20260917/reload-validation.json b/latency-aware/walker2d/vla/starvla-qwengr00t-h1/walker2d_profile_gr00t_h1_5fps_2h200_20260917/reload-validation.json new file mode 100644 index 0000000000000000000000000000000000000000..c8d10350ecd9c9d074cce8b37bc8027ed87e53b4 --- /dev/null +++ b/latency-aware/walker2d/vla/starvla-qwengr00t-h1/walker2d_profile_gr00t_h1_5fps_2h200_20260917/reload-validation.json @@ -0,0 +1,23 @@ +{ + "verified_at": "2026-09-17T20:36:51.671176+00:00", + "state": "TRAINING_AND_SAVED_BUNDLE_FORWARD_VERIFIED", + "training_updates": 5000, + "checkpoint_sha256": "1859cec5319db37b97a62d8190f081771f38978fa2ade39cfb82b155ac56b83b", + "action_horizon": 1, + "state_dim": 17, + "action_dim": 6, + "env_fps": 5, + "obs_fps": 5, + "global_batch": 64, + "seed": 42, + "sample_source": "actual held-out profile raw episode; manifest train-only minmax applied once", + "reload_contract": "existing saved-model loader checks missing/unexpected keys, allowing its documented tied-Qwen lm_head equivalence", + "forward_shape": [ + 1, + 1, + 6 + ], + "forward_finite": true, + "F20_status": "not yet run", + "acceptance": "not established by this engineering check" +} diff --git a/latency-aware/walker2d/vla/starvla-qwengr00t-h1/walker2d_profile_gr00t_h1_5fps_2h200_20260917/source/README.md b/latency-aware/walker2d/vla/starvla-qwengr00t-h1/walker2d_profile_gr00t_h1_5fps_2h200_20260917/source/README.md new file mode 100644 index 0000000000000000000000000000000000000000..372c982f7f630f23dc5ef5b69b2be04852cd794b --- /dev/null +++ b/latency-aware/walker2d/vla/starvla-qwengr00t-h1/walker2d_profile_gr00t_h1_5fps_2h200_20260917/source/README.md @@ -0,0 +1,5 @@ +# Walker2D GR00T profile H1 /5FPS + +Fresh5000update training, savedbundle verification and matchedRTX3090 measuredF20 are complete. B=63.241475 ± 6.584120; F=852.872403 ± 357.102322; populationSD,20matchedseeds42–61,20pairedgains. This improvement is not a high-performance robotic acceptance claim. The >3000 gate applies to teacher demonstrations only. + +B/F observedlatencies differ; no equal-latency causal claim. See F-summary.json and F-verification.json. Fullrawstep/action/latencytraces remain on3090; compactevidencepublished in dataset revision b59922566cb475e61455342b1111b9e9771c09de. Original reload-validation remains a historical snapshot. Evaluatedweights unchanged at af4f38060b0374c5684c43f22cd504257cf2bffd. diff --git a/latency-aware/walker2d/vla/starvla-qwengr00t-h1/walker2d_profile_gr00t_h1_5fps_2h200_20260917/source/provenance.json b/latency-aware/walker2d/vla/starvla-qwengr00t-h1/walker2d_profile_gr00t_h1_5fps_2h200_20260917/source/provenance.json new file mode 100644 index 0000000000000000000000000000000000000000..776da948f48c1501dd59c4cd5f5b6238606eb98c --- /dev/null +++ b/latency-aware/walker2d/vla/starvla-qwengr00t-h1/walker2d_profile_gr00t_h1_5fps_2h200_20260917/source/provenance.json @@ -0,0 +1,172 @@ +{ + "task": "walker2d", + "backbone": { + "repo_id": "Qwen/Qwen3-VL-4B-Instruct", + "revision": "ebb281ec70b05090aa6165b016eac8ec08e71b17" + }, + "checkpoint": "steps_5000_pytorch_model.pt", + "training_steps": 5000, + "seed": 42, + "global_batch": 64, + "training_run_id": "walker2d_profile_gr00t_h1_5fps_2h200_20260917", + "condition": "profile_latency", + "source": { + "code_source": { + "repo": "latency-sensitive-bench/Standard-Pipeline", + "revision": "571cb5801720488ed6458b9cfe3f4c595b069cea", + "prefix": "walker2d/zero_latency/shared/source_100ep_v1/demonstrations/raw", + "sha256": { + "metadata.json": "c8ea2669493beabd97d9e7ecd06cc443adcb1760d8e71dd03588ea4adea20ae9", + "train.parquet": "6a105846e7def09f10fd2a985413cc9a04b9433f57519d330ce33a1033ba7e65", + "val.parquet": "687c1a5d08b3641b5ae76d218dd6afe483f8873b057aa3ab7a81108a6772f424" + }, + "source_tar_sha256": "9847e8924ba7a6446aaa3ce735f4dc2849046c37e21aeb8867f3154ab163ede0" + }, + "teacher_selection": { + "selected": "training_best", + "candidates": { + "final": { + "checkpoint": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/teacher/checkpoints/walker2d_profile_appo_h1_5fps_20260917/checkpoint_p0/checkpoint_000019536_10002432.pth", + "sha256": "dcfe291db668f89db2fead0028a76fb5a005965b9daaab7a29bd19eb13c43ca0", + "episodes": 20, + "returns": [ + 4299.208001651971, + 4253.381546607346, + 4274.26267070416, + 4281.764485481581, + 4316.5965023405115, + 4185.72463542741, + 4268.629567678589, + 4265.304288469715, + 4331.873574180083, + 3269.8377559505593, + 3583.941712492686, + 4264.461568188047, + 4268.998666838904, + 4293.114110604841, + 4274.87183914279, + 4358.281606522924, + 4247.960230740107, + 4363.387419962692, + 4258.273723341316, + 4286.0125421364155 + ], + "mean": 4197.294322423132, + "std": 264.3552584808844, + "strict_gt3000": 20 + }, + "training_best": { + "checkpoint": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/teacher/checkpoints/walker2d_profile_appo_h1_5fps_20260917/checkpoint_p0/best_000014536_7442432_reward_4186.861.pth", + "sha256": "a9f0e63f8094fe10f82d2395fc7b59958dc22ac1abc59a6da52772eb1fd944ab", + "episodes": 20, + "returns": [ + 4232.421844289257, + 4211.834128554937, + 4235.009681429353, + 4216.192126194387, + 4225.968774742952, + 4246.130242600228, + 4259.369933944325, + 4211.69597436189, + 4227.7128439529915, + 4219.107550141977, + 4176.009289189152, + 4216.315331519501, + 4233.440001863116, + 4210.597842410642, + 4216.913958972969, + 4236.856116118128, + 4241.158612744575, + 4224.4414913018245, + 4191.601768644205, + 4233.529106106846 + ], + "mean": 4223.315330954163, + "std": 18.175464884559204, + "strict_gt3000": 20 + } + }, + "completed_at": "2026-09-17T17:24:56.075104+00:00" + }, + "profile": { + "state": "PROFILE_TEACHER_CONFIG_READY", + "profile": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/profile_3090_20260917T164637291306Z/profile.json", + "profile_sha256": "acf0a997fa7866b5a101714ec1479b8cc19b36ae1262f11081ab828de3e79661", + "source_recipe": "/mnt/local/lzj/latency-sensitive-bench/code/h1-5fps-20260917-v2/configs/examples/gymnasium/walker2d/small_model_train_sf_official_10m.yaml", + "source_recipe_sha256": "ef0abc795b6c33d152dbdeaa4a5a817df2546f400c17ce27ac9159e01bb92901", + "config": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/teacher.yaml", + "algo": "APPO", + "budget_env_steps": 10000000, + "seed": 0, + "fps": 5, + "latency": "iid", + "initialization": "fresh", + "metadata_derivation": "current task_name walker2d and state17 labels from wrapper; original official10m recipe seed0/hyperparameters/native physics preserved", + "next": "Run only after immutable new P is verified; Final/best20 tieFinal, E10, bounded12probe, strictreturn>3000." + }, + "data": { + "train": { + "episodes": 90, + "frames": 90000, + "min_replay_return": 4128.808384731412, + "max_replay_return": 4261.9987871050835, + "sha256": "a161eb4de7d2539c456f5c21ae109c1561ee7c60a8cc273e3a32a9abae862fd1" + }, + "val": { + "episodes": 10, + "frames": 10000, + "min_replay_return": 4201.307677522302, + "max_replay_return": 4247.824013382196, + "sha256": "5f17b3e18e7c79e94eefb773f52b7abd82873c8629e7a950eb763c849ca5bb4c" + } + }, + "dataset": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/profile_latency/data/lerobot/walker2d_h1_5fps_profile", + "condition": "profile_latency", + "action_horizon": 1, + "state_normalization": { + "type": "min_max", + "min": [ + 0.9851876497268677, + -0.1729077845811844, + -0.5815715193748474, + -0.41291648149490356, + -1.1557613611221313, + -1.565748929977417, + -1.5937821865081787, + -1.4450324773788452, + -1.448801040649414, + -2.645611524581909, + -8.647480010986328, + -10.0, + -10.0, + -10.0, + -10.0, + -10.0, + -10.0 + ], + "max": [ + 1.5055257081985474, + 0.6807975769042969, + 0.14722545444965363, + 0.17162981629371643, + 1.3325903415679932, + 0.18669182062149048, + 0.3009708821773529, + 1.3312314748764038, + 6.26270866394043, + 3.290994644165039, + 10.0, + 10.0, + 10.0, + 10.0, + 10.0, + 10.0, + 10.0 + ] + }, + "prompt_key": 1, + "fresh_vla_initialization": true + }, + "training_config_sha256": "54eef4fa88502b723bdc47150eb134cd4adf7b81296c4a57e0936ffad3016f86", + "dataset_manifest_sha256": "44e468feb3bd491709b90a800bba465aacd566b83092c44aac557e9acbd479ee" +} diff --git a/latency-aware/walker2d/vla/starvla-qwengr00t-h1/walker2d_profile_gr00t_h1_5fps_2h200_20260917/task_contract.json b/latency-aware/walker2d/vla/starvla-qwengr00t-h1/walker2d_profile_gr00t_h1_5fps_2h200_20260917/task_contract.json new file mode 100644 index 0000000000000000000000000000000000000000..83e3865ecb5e5aad835cb8de453f17e41e7eb216 --- /dev/null +++ b/latency-aware/walker2d/vla/starvla-qwengr00t-h1/walker2d_profile_gr00t_h1_5fps_2h200_20260917/task_contract.json @@ -0,0 +1,80 @@ +{ + "action_space": { + "dtype": "float32", + "high": [ + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0 + ], + "labels": [ + "right_thigh_torque", + "right_leg_torque", + "right_foot_torque", + "left_thigh_torque", + "left_leg_torque", + "left_foot_torque" + ], + "low": [ + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0 + ], + "type": "box" + }, + "base_prompt": "Move the Walker2d robot forward while keeping its torso upright. Predict six continuous torques in [-1, 1] ordered as right thigh, right leg, right foot, left thigh, left leg, and left foot.", + "env_fps": 5.0, + "env_id": "LatencyBench/Walker2dContinuous-v0", + "frame_stack": 1, + "make_kwargs": { + "base_env_id": "Walker2d-v4", + "base_make_kwargs": { + "ctrl_cost_weight": 0.001, + "exclude_current_positions_from_observation": true, + "forward_reward_weight": 1.0, + "healthy_reward": 1.0, + "reset_noise_scale": 0.005, + "terminate_when_unhealthy": true + }, + "render_mode": "rgb_array" + }, + "noop_action": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ], + "obs_fps": 5.0, + "registration_imports": [ + "latency_bench.envs.gymnasium_walker2d" + ], + "state_space": { + "labels": [ + "torso_height", + "torso_angle", + "right_thigh_angle", + "right_leg_angle", + "right_foot_angle", + "left_thigh_angle", + "left_leg_angle", + "left_foot_angle", + "torso_x_velocity", + "torso_z_velocity", + "torso_angular_velocity", + "right_thigh_angular_velocity", + "right_leg_angular_velocity", + "right_foot_angular_velocity", + "left_thigh_angular_velocity", + "left_leg_angular_velocity", + "left_foot_angular_velocity" + ] + }, + "task_name": "walker2d" +} diff --git a/latency-aware/walker2d/vla/starvla-qwenoft-h1/walker2d_rgb_state_profile_20260912T092949Z_openvla_native_continuous_projector_sft_5k/README.md b/latency-aware/walker2d/vla/starvla-qwenoft-h1/walker2d_rgb_state_profile_20260912T092949Z_openvla_native_continuous_projector_sft_5k/README.md new file mode 100644 index 0000000000000000000000000000000000000000..e3969e51781f9174ea394a1f9d9d725fcc468adc --- /dev/null +++ b/latency-aware/walker2d/vla/starvla-qwenoft-h1/walker2d_rgb_state_profile_20260912T092949Z_openvla_native_continuous_projector_sft_5k/README.md @@ -0,0 +1,29 @@ +# walker2d / qwenoft + +Training condition: `latency-aware`. Run: `walker2d_rgb_state_profile_20260912T092949Z_openvla_native_continuous_projector_sft_5k`. + +Copied from [Standard-Pipeline](https://huggingface.co/latency-sensitive-bench/Standard-Pipeline/tree/2208875f92b21e3f8ec2363a30a6705dfe56cc04/walker2d/profile_latency/OpenVLA) at `2208875f92b21e3f8ec2363a30a6705dfe56cc04`. + +- Checkpoint: `checkpoints/model.pt` +- Selection: only published VLA checkpoint +- Checkpoint SHA256: `cbd8a14cfa3fe1a97b6d131294e0264e43e0267943189f776b4191922ee4559d` +- Configuration, FPS, actions, normalization and latency semantics retain their source values. +- This import does not run evaluations or change training/evaluation/acceptance status. + +Action horizon is 1. Use `config.yaml` and `dataset_statistics.json` with the matching +prepared backbone. `task_contract.json` is extracted from the original training configuration. +Original machine paths in historical configurations are provenance: bind checkpoint, +backbone, manifest and profile paths to downloaded local files when running inference. + +[Original provenance](source/provenance.json) retains dataset references and recorded experiment evidence. + +## Download + +```python +from huggingface_hub import snapshot_download +prefix = "latency-aware/walker2d/vla/starvla-qwenoft-h1/walker2d_rgb_state_profile_20260912T092949Z_openvla_native_continuous_projector_sft_5k" +snapshot_download("latency-sensitive-bench/benchmark-models", revision="", + allow_patterns=[prefix + "/**"]) +``` + +Use the verified import commit listed in the repository's import report. diff --git a/latency-aware/walker2d/vla/starvla-qwenoft-h1/walker2d_rgb_state_profile_20260912T092949Z_openvla_native_continuous_projector_sft_5k/checkpoints/model.pt b/latency-aware/walker2d/vla/starvla-qwenoft-h1/walker2d_rgb_state_profile_20260912T092949Z_openvla_native_continuous_projector_sft_5k/checkpoints/model.pt new file mode 100644 index 0000000000000000000000000000000000000000..14e3c194bac41a6d2dab88f70687ad2fe9b0cfd3 --- /dev/null +++ b/latency-aware/walker2d/vla/starvla-qwenoft-h1/walker2d_rgb_state_profile_20260912T092949Z_openvla_native_continuous_projector_sft_5k/checkpoints/model.pt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:cbd8a14cfa3fe1a97b6d131294e0264e43e0267943189f776b4191922ee4559d +size 9785142689 diff --git a/latency-aware/walker2d/vla/starvla-qwenoft-h1/walker2d_rgb_state_profile_20260912T092949Z_openvla_native_continuous_projector_sft_5k/config.full.yaml b/latency-aware/walker2d/vla/starvla-qwenoft-h1/walker2d_rgb_state_profile_20260912T092949Z_openvla_native_continuous_projector_sft_5k/config.full.yaml new file mode 100644 index 0000000000000000000000000000000000000000..3bdcc2d2be445b159a5f3300081f7a9f588b8529 --- /dev/null +++ b/latency-aware/walker2d/vla/starvla-qwenoft-h1/walker2d_rgb_state_profile_20260912T092949Z_openvla_native_continuous_projector_sft_5k/config.full.yaml @@ -0,0 +1,396 @@ +framework: + name: QwenOFT + qwenvl: + base_vlm: /mnt/checkpoints/latency-sensitive-bench/openvla/Qwen3-VL-4B-Instruct + attn_implementation: flash_attention_2 + enable_gradient_checkpointing: true + action_model: + action_model_type: MLP + action_dim: 6 + state_dim: 17 + action_hidden_dim: 2560 + future_action_window_size: 0 + past_action_window_size: 0 + loss_type: l1 + state_encoding: continuous_projector + task_objective: null + action_horizon: 1 + action_env_dim: 6 + kv_memory: + enabled: false + window: 4 + rollout_len: 8 + packed_train: false + rebased_sink: true +datasets: + vla_data: + dataset_py: lerobot_datasets + include_state: true + data_root_dir: /mnt/results/latency-sensitive-bench/profile-latency-new/walker2d/20260912T092949Z-walker2d/converted + data_mix: walker2d_rgb_state_profile_20260912T092949Z + eval_data_mix: walker2d_rgb_state_profile_20260912T092949Z__val + custom_mixtures_path: /mnt/results/latency-sensitive-bench/profile-latency-new/walker2d/20260912T092949Z-walker2d/converted/_generated_mixtures/walker2d_rgb_state_profile_20260912T092949Z.json + action_type: discrete + sequential_step_sampling: false + eval_sequential_step_sampling: null + num_workers: 8 + eval_num_workers: 8 + prefetch_factor: 4 + persistent_workers: true + pin_memory: true + shuffle: true + action_balance: + enabled: false + strategy: balanced_epoch + action_key: action_id + target_flap_fraction: 0.3 + noop_id: 0 + flap_id: 1 + latency_curriculum: + enabled: false + strategy: exclusive + latencies: null + phase_steps: null + phase_distributions: null + new_latency_passes: 1.0 + replay_passes: 0.25 + target_total_passes: 2.0 + final_equalization: true + step_budget_mode: auto + eval_at_phase_end: false + save_at_phase_end: false + computed_plan: null + per_device_batch_size: 16 + load_all_data_for_training: true + num_obs_frames: 1 + image_mode: single + prompt_mode: raw + stitch_grid: + - 2 + - 2 + obs_image_size: null + video_backend: torchvision_av + gymnasium_task_contract: + action_space: + dtype: float32 + high: + - 1.0 + - 1.0 + - 1.0 + - 1.0 + - 1.0 + - 1.0 + labels: + - right_thigh_torque + - right_leg_torque + - right_foot_torque + - left_thigh_torque + - left_leg_torque + - left_foot_torque + low: + - -1.0 + - -1.0 + - -1.0 + - -1.0 + - -1.0 + - -1.0 + type: box + base_prompt: Move the Walker2d robot forward while keeping its torso upright. + Predict six continuous torques in [-1, 1] ordered as right thigh, right leg, + right foot, left thigh, left leg, and left foot. + env_fps: 10.0 + env_id: LatencyBench/Walker2dContinuous-v0 + frame_stack: 1 + make_kwargs: + base_env_id: Walker2d-v4 + base_make_kwargs: + ctrl_cost_weight: 0.001 + exclude_current_positions_from_observation: true + forward_reward_weight: 1.0 + healthy_reward: 1.0 + reset_noise_scale: 0.005 + terminate_when_unhealthy: true + render_mode: rgb_array + noop_action: + - 0.0 + - 0.0 + - 0.0 + - 0.0 + - 0.0 + - 0.0 + obs_fps: 10.0 + registration_imports: + - latency_bench.envs.gymnasium_walker2d + state_space: + labels: + - torso_height + - torso_angle + - right_thigh_angle + - right_leg_angle + - right_foot_angle + - left_thigh_angle + - left_leg_angle + - left_foot_angle + - torso_x_velocity + - torso_z_velocity + - torso_angular_velocity + - right_thigh_angular_velocity + - right_leg_angular_velocity + - right_foot_angular_velocity + - left_thigh_angular_velocity + - left_leg_angular_velocity + - left_foot_angular_velocity + task_name: walker2d_rgb_state + active_action_dim: 6 +dataset: + source_hf: '' + config_name: null + source_subdir: null + converted_name: walker2d_rgb_state_profile_20260912T092949Z + single_source_hf: '' + mixed_source_hf: '' + single_converted_name: walker2d_rgb_state_profile_20260912T092949Z + mixed_converted_name: walker2d_rgb_state_profile_20260912T092949Z + single_latency_filter: null + mixed_latency_filter: null + force_download: false + setup_force: false + skip_verification: false + target_latency_unit: observation_steps + verify_rows: 200 + max_episodes: null + episodes_per_latency: null + latency_filter: null + debug_subset: + enabled: false + max_episodes: 5 + suffix: debug +base_model: + repo_id: Qwen/Qwen3-VL-4B-Instruct +initialization: + checkpoint_local_dir: null + checkpoint_hf_repo_id: null + checkpoint_filename: null +trainer: + max_train_steps: 5000 + num_warmup_steps: 100 + save_interval: 500 + eval_interval: 500 + eval_num_batches: 200 + per_latency_eval_num_batches: null + eval_action_classification: false + eval_action_classification_interval: null + cc_f1_tolerance: 1 + learning_rate: + base: 2.0e-05 + qwen_vl_interface: 1.0e-05 + action_model: 0.0001 + lr_scheduler_type: cosine_with_min_lr + scheduler_specific_kwargs: + min_lr: 1.0e-06 + freeze_modules: '' + freeze_vit: false + freeze_tied_embedding: false + freeze_llm_layers: [] + loss_scale: + vla: 1.0 + vlm: 0.1 + max_grad_norm: 1.0 + weight_decay: 0.0 + logging_frequency: 1 + profile_timing: + enabled: true + log_interval: 10 + gradient_clipping: 1.0 + gradient_accumulation_steps: 1 + distributed_backend: deepspeed + is_resume: false + pretrained_checkpoint: null + resume_step: 0 + reload_modules: null + optimizer: + name: AdamW + betas: + - 0.9 + - 0.95 + eps: 1.0e-08 + weight_decay: 1.0e-08 + fused: true + save_format: pt +workspace_dir: /home/lzj/code/latency-sensitive-bench +run_root_dir: /mnt/results/latency-sensitive-bench/profile-latency-new/walker2d/20260912T092949Z-walker2d/vla +seed: 42 +wandb_entity: dongqianyu99-zhejiang-university +wandb_project: latency-sensitive-bench +auth: + env_file: null + hf_token_env: HF_TOKEN + wandb_api_key_env: WANDB_API_KEY +paths: + run_root_dir: /mnt/results/latency-sensitive-bench/profile-latency-new/walker2d/20260912T092949Z-walker2d/vla + dataset_local_dir: /mnt/results/latency-sensitive-bench/profile-latency-new/walker2d/20260912T092949Z-walker2d/converted + dataset_cache_dir: null + base_model_dir: /mnt/checkpoints/latency-sensitive-bench/openvla/Qwen3-VL-4B-Instruct + accelerate_config: starVLA/config/deepseeds/deepspeed_zero2_cpu_offload.yaml +launch: + use_accelerate: true + gpus: null + num_processes: 1 + dry_run: false +conda: + enabled: true + env_name: null +rl_games: + model_alias: openvla + env_eval: + image_size: 224 + frameskip: 1 + image_transform: raw_rgb + prompt_mode: raw + ghost_trail: + history_frames: 5 + gamma: 1.3 + min_alpha: 35 + scroll_px_per_step: 4.0 + ground_fraction: 0.22 + seed: 42 + fixed_episode_seeds: true + latency_seed_stride: 0 + task_seed_stride: 0 + task_description: '' + eval_parallel_envs: 5 + action_chunk_execution: + enabled: false + chunk_size: null + enabled: false + gymnasium: + make_kwargs: {} + mid_train: + enabled: false + interval_steps: 100 + latencies: + - 0 + num_episodes: 5 + max_steps_per_episode: 3600 + post_train: + enabled: false + latencies: + - 0 + - 1 + - 2 + - 3 + - 4 + num_episodes: 5 + max_steps_per_episode: 3600 + eval_backend: latency_bench + distributed_mode: none + vectorized: + enabled: false + batch_size: 1 + latency: + prompt_map_path: /mnt/results/latency-sensitive-bench/profile-latency-new/walker2d/20260912T092949Z-walker2d/converted/walker2d_rgb_state_profile_20260912T092949Z/latency_prompt_map.json + mode: single + values: + - 1 + - 2 + task: gymnasium + gymnasium: + task_contract: + action_space: + dtype: float32 + high: + - 1.0 + - 1.0 + - 1.0 + - 1.0 + - 1.0 + - 1.0 + labels: + - right_thigh_torque + - right_leg_torque + - right_foot_torque + - left_thigh_torque + - left_leg_torque + - left_foot_torque + low: + - -1.0 + - -1.0 + - -1.0 + - -1.0 + - -1.0 + - -1.0 + type: box + base_prompt: Move the Walker2d robot forward while keeping its torso upright. + Predict six continuous torques in [-1, 1] ordered as right thigh, right leg, + right foot, left thigh, left leg, and left foot. + env_fps: 10.0 + env_id: LatencyBench/Walker2dContinuous-v0 + frame_stack: 1 + make_kwargs: + base_env_id: Walker2d-v4 + base_make_kwargs: + ctrl_cost_weight: 0.001 + exclude_current_positions_from_observation: true + forward_reward_weight: 1.0 + healthy_reward: 1.0 + reset_noise_scale: 0.005 + terminate_when_unhealthy: true + render_mode: rgb_array + noop_action: + - 0.0 + - 0.0 + - 0.0 + - 0.0 + - 0.0 + - 0.0 + obs_fps: 10.0 + registration_imports: + - latency_bench.envs.gymnasium_walker2d + state_space: + labels: + - torso_height + - torso_angle + - right_thigh_angle + - right_leg_angle + - right_foot_angle + - left_thigh_angle + - left_leg_angle + - left_foot_angle + - torso_x_velocity + - torso_z_velocity + - torso_angular_velocity + - right_thigh_angular_velocity + - right_leg_angular_velocity + - right_foot_angular_velocity + - left_thigh_angular_velocity + - left_leg_angular_velocity + - left_foot_angular_velocity + task_name: walker2d_rgb_state + task_name: walker2d_rgb_state + initialization_mode: scratch + action_carrier: native +model: openvla +env: gymnasium +init: scratch +mode: single +checkpoint: + load: none + hf_repo_id: null + save_best_model: false + save_final_model: true + save_pt_file: true + save_training_state: false + save_safetensors_file: false + local: + keep_last_n: 1 + sync: + enabled: false + repo_id: null + keep_last_n: 0 + sync_every_n_checkpoints: 1 + resume_policy: local_latest +run_id: walker2d_rgb_state_profile_20260912T092949Z_openvla_native_continuous_projector_sft_5k +output_dir: /mnt/results/latency-sensitive-bench/profile-latency-new/walker2d/20260912T092949Z-walker2d/vla/walker2d_rgb_state_profile_20260912T092949Z_openvla_native_continuous_projector_sft_5k +config_yaml: /mnt/results/latency-sensitive-bench/profile-latency-new/walker2d/20260912T092949Z-walker2d/h100/vla_train.yaml +is_debug: false +version_id: '0.21' diff --git a/latency-aware/walker2d/vla/starvla-qwenoft-h1/walker2d_rgb_state_profile_20260912T092949Z_openvla_native_continuous_projector_sft_5k/config.yaml b/latency-aware/walker2d/vla/starvla-qwenoft-h1/walker2d_rgb_state_profile_20260912T092949Z_openvla_native_continuous_projector_sft_5k/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..3bdcc2d2be445b159a5f3300081f7a9f588b8529 --- /dev/null +++ b/latency-aware/walker2d/vla/starvla-qwenoft-h1/walker2d_rgb_state_profile_20260912T092949Z_openvla_native_continuous_projector_sft_5k/config.yaml @@ -0,0 +1,396 @@ +framework: + name: QwenOFT + qwenvl: + base_vlm: /mnt/checkpoints/latency-sensitive-bench/openvla/Qwen3-VL-4B-Instruct + attn_implementation: flash_attention_2 + enable_gradient_checkpointing: true + action_model: + action_model_type: MLP + action_dim: 6 + state_dim: 17 + action_hidden_dim: 2560 + future_action_window_size: 0 + past_action_window_size: 0 + loss_type: l1 + state_encoding: continuous_projector + task_objective: null + action_horizon: 1 + action_env_dim: 6 + kv_memory: + enabled: false + window: 4 + rollout_len: 8 + packed_train: false + rebased_sink: true +datasets: + vla_data: + dataset_py: lerobot_datasets + include_state: true + data_root_dir: /mnt/results/latency-sensitive-bench/profile-latency-new/walker2d/20260912T092949Z-walker2d/converted + data_mix: walker2d_rgb_state_profile_20260912T092949Z + eval_data_mix: walker2d_rgb_state_profile_20260912T092949Z__val + custom_mixtures_path: /mnt/results/latency-sensitive-bench/profile-latency-new/walker2d/20260912T092949Z-walker2d/converted/_generated_mixtures/walker2d_rgb_state_profile_20260912T092949Z.json + action_type: discrete + sequential_step_sampling: false + eval_sequential_step_sampling: null + num_workers: 8 + eval_num_workers: 8 + prefetch_factor: 4 + persistent_workers: true + pin_memory: true + shuffle: true + action_balance: + enabled: false + strategy: balanced_epoch + action_key: action_id + target_flap_fraction: 0.3 + noop_id: 0 + flap_id: 1 + latency_curriculum: + enabled: false + strategy: exclusive + latencies: null + phase_steps: null + phase_distributions: null + new_latency_passes: 1.0 + replay_passes: 0.25 + target_total_passes: 2.0 + final_equalization: true + step_budget_mode: auto + eval_at_phase_end: false + save_at_phase_end: false + computed_plan: null + per_device_batch_size: 16 + load_all_data_for_training: true + num_obs_frames: 1 + image_mode: single + prompt_mode: raw + stitch_grid: + - 2 + - 2 + obs_image_size: null + video_backend: torchvision_av + gymnasium_task_contract: + action_space: + dtype: float32 + high: + - 1.0 + - 1.0 + - 1.0 + - 1.0 + - 1.0 + - 1.0 + labels: + - right_thigh_torque + - right_leg_torque + - right_foot_torque + - left_thigh_torque + - left_leg_torque + - left_foot_torque + low: + - -1.0 + - -1.0 + - -1.0 + - -1.0 + - -1.0 + - -1.0 + type: box + base_prompt: Move the Walker2d robot forward while keeping its torso upright. + Predict six continuous torques in [-1, 1] ordered as right thigh, right leg, + right foot, left thigh, left leg, and left foot. + env_fps: 10.0 + env_id: LatencyBench/Walker2dContinuous-v0 + frame_stack: 1 + make_kwargs: + base_env_id: Walker2d-v4 + base_make_kwargs: + ctrl_cost_weight: 0.001 + exclude_current_positions_from_observation: true + forward_reward_weight: 1.0 + healthy_reward: 1.0 + reset_noise_scale: 0.005 + terminate_when_unhealthy: true + render_mode: rgb_array + noop_action: + - 0.0 + - 0.0 + - 0.0 + - 0.0 + - 0.0 + - 0.0 + obs_fps: 10.0 + registration_imports: + - latency_bench.envs.gymnasium_walker2d + state_space: + labels: + - torso_height + - torso_angle + - right_thigh_angle + - right_leg_angle + - right_foot_angle + - left_thigh_angle + - left_leg_angle + - left_foot_angle + - torso_x_velocity + - torso_z_velocity + - torso_angular_velocity + - right_thigh_angular_velocity + - right_leg_angular_velocity + - right_foot_angular_velocity + - left_thigh_angular_velocity + - left_leg_angular_velocity + - left_foot_angular_velocity + task_name: walker2d_rgb_state + active_action_dim: 6 +dataset: + source_hf: '' + config_name: null + source_subdir: null + converted_name: walker2d_rgb_state_profile_20260912T092949Z + single_source_hf: '' + mixed_source_hf: '' + single_converted_name: walker2d_rgb_state_profile_20260912T092949Z + mixed_converted_name: walker2d_rgb_state_profile_20260912T092949Z + single_latency_filter: null + mixed_latency_filter: null + force_download: false + setup_force: false + skip_verification: false + target_latency_unit: observation_steps + verify_rows: 200 + max_episodes: null + episodes_per_latency: null + latency_filter: null + debug_subset: + enabled: false + max_episodes: 5 + suffix: debug +base_model: + repo_id: Qwen/Qwen3-VL-4B-Instruct +initialization: + checkpoint_local_dir: null + checkpoint_hf_repo_id: null + checkpoint_filename: null +trainer: + max_train_steps: 5000 + num_warmup_steps: 100 + save_interval: 500 + eval_interval: 500 + eval_num_batches: 200 + per_latency_eval_num_batches: null + eval_action_classification: false + eval_action_classification_interval: null + cc_f1_tolerance: 1 + learning_rate: + base: 2.0e-05 + qwen_vl_interface: 1.0e-05 + action_model: 0.0001 + lr_scheduler_type: cosine_with_min_lr + scheduler_specific_kwargs: + min_lr: 1.0e-06 + freeze_modules: '' + freeze_vit: false + freeze_tied_embedding: false + freeze_llm_layers: [] + loss_scale: + vla: 1.0 + vlm: 0.1 + max_grad_norm: 1.0 + weight_decay: 0.0 + logging_frequency: 1 + profile_timing: + enabled: true + log_interval: 10 + gradient_clipping: 1.0 + gradient_accumulation_steps: 1 + distributed_backend: deepspeed + is_resume: false + pretrained_checkpoint: null + resume_step: 0 + reload_modules: null + optimizer: + name: AdamW + betas: + - 0.9 + - 0.95 + eps: 1.0e-08 + weight_decay: 1.0e-08 + fused: true + save_format: pt +workspace_dir: /home/lzj/code/latency-sensitive-bench +run_root_dir: /mnt/results/latency-sensitive-bench/profile-latency-new/walker2d/20260912T092949Z-walker2d/vla +seed: 42 +wandb_entity: dongqianyu99-zhejiang-university +wandb_project: latency-sensitive-bench +auth: + env_file: null + hf_token_env: HF_TOKEN + wandb_api_key_env: WANDB_API_KEY +paths: + run_root_dir: /mnt/results/latency-sensitive-bench/profile-latency-new/walker2d/20260912T092949Z-walker2d/vla + dataset_local_dir: /mnt/results/latency-sensitive-bench/profile-latency-new/walker2d/20260912T092949Z-walker2d/converted + dataset_cache_dir: null + base_model_dir: /mnt/checkpoints/latency-sensitive-bench/openvla/Qwen3-VL-4B-Instruct + accelerate_config: starVLA/config/deepseeds/deepspeed_zero2_cpu_offload.yaml +launch: + use_accelerate: true + gpus: null + num_processes: 1 + dry_run: false +conda: + enabled: true + env_name: null +rl_games: + model_alias: openvla + env_eval: + image_size: 224 + frameskip: 1 + image_transform: raw_rgb + prompt_mode: raw + ghost_trail: + history_frames: 5 + gamma: 1.3 + min_alpha: 35 + scroll_px_per_step: 4.0 + ground_fraction: 0.22 + seed: 42 + fixed_episode_seeds: true + latency_seed_stride: 0 + task_seed_stride: 0 + task_description: '' + eval_parallel_envs: 5 + action_chunk_execution: + enabled: false + chunk_size: null + enabled: false + gymnasium: + make_kwargs: {} + mid_train: + enabled: false + interval_steps: 100 + latencies: + - 0 + num_episodes: 5 + max_steps_per_episode: 3600 + post_train: + enabled: false + latencies: + - 0 + - 1 + - 2 + - 3 + - 4 + num_episodes: 5 + max_steps_per_episode: 3600 + eval_backend: latency_bench + distributed_mode: none + vectorized: + enabled: false + batch_size: 1 + latency: + prompt_map_path: /mnt/results/latency-sensitive-bench/profile-latency-new/walker2d/20260912T092949Z-walker2d/converted/walker2d_rgb_state_profile_20260912T092949Z/latency_prompt_map.json + mode: single + values: + - 1 + - 2 + task: gymnasium + gymnasium: + task_contract: + action_space: + dtype: float32 + high: + - 1.0 + - 1.0 + - 1.0 + - 1.0 + - 1.0 + - 1.0 + labels: + - right_thigh_torque + - right_leg_torque + - right_foot_torque + - left_thigh_torque + - left_leg_torque + - left_foot_torque + low: + - -1.0 + - -1.0 + - -1.0 + - -1.0 + - -1.0 + - -1.0 + type: box + base_prompt: Move the Walker2d robot forward while keeping its torso upright. + Predict six continuous torques in [-1, 1] ordered as right thigh, right leg, + right foot, left thigh, left leg, and left foot. + env_fps: 10.0 + env_id: LatencyBench/Walker2dContinuous-v0 + frame_stack: 1 + make_kwargs: + base_env_id: Walker2d-v4 + base_make_kwargs: + ctrl_cost_weight: 0.001 + exclude_current_positions_from_observation: true + forward_reward_weight: 1.0 + healthy_reward: 1.0 + reset_noise_scale: 0.005 + terminate_when_unhealthy: true + render_mode: rgb_array + noop_action: + - 0.0 + - 0.0 + - 0.0 + - 0.0 + - 0.0 + - 0.0 + obs_fps: 10.0 + registration_imports: + - latency_bench.envs.gymnasium_walker2d + state_space: + labels: + - torso_height + - torso_angle + - right_thigh_angle + - right_leg_angle + - right_foot_angle + - left_thigh_angle + - left_leg_angle + - left_foot_angle + - torso_x_velocity + - torso_z_velocity + - torso_angular_velocity + - right_thigh_angular_velocity + - right_leg_angular_velocity + - right_foot_angular_velocity + - left_thigh_angular_velocity + - left_leg_angular_velocity + - left_foot_angular_velocity + task_name: walker2d_rgb_state + task_name: walker2d_rgb_state + initialization_mode: scratch + action_carrier: native +model: openvla +env: gymnasium +init: scratch +mode: single +checkpoint: + load: none + hf_repo_id: null + save_best_model: false + save_final_model: true + save_pt_file: true + save_training_state: false + save_safetensors_file: false + local: + keep_last_n: 1 + sync: + enabled: false + repo_id: null + keep_last_n: 0 + sync_every_n_checkpoints: 1 + resume_policy: local_latest +run_id: walker2d_rgb_state_profile_20260912T092949Z_openvla_native_continuous_projector_sft_5k +output_dir: /mnt/results/latency-sensitive-bench/profile-latency-new/walker2d/20260912T092949Z-walker2d/vla/walker2d_rgb_state_profile_20260912T092949Z_openvla_native_continuous_projector_sft_5k +config_yaml: /mnt/results/latency-sensitive-bench/profile-latency-new/walker2d/20260912T092949Z-walker2d/h100/vla_train.yaml +is_debug: false +version_id: '0.21' diff --git a/latency-aware/walker2d/vla/starvla-qwenoft-h1/walker2d_rgb_state_profile_20260912T092949Z_openvla_native_continuous_projector_sft_5k/dataset_statistics.json b/latency-aware/walker2d/vla/starvla-qwenoft-h1/walker2d_rgb_state_profile_20260912T092949Z_openvla_native_continuous_projector_sft_5k/dataset_statistics.json new file mode 100644 index 0000000000000000000000000000000000000000..3c992bae026cb655c540048e0dcbb0dfb423bf58 --- /dev/null +++ b/latency-aware/walker2d/vla/starvla-qwenoft-h1/walker2d_rgb_state_profile_20260912T092949Z_openvla_native_continuous_projector_sft_5k/dataset_statistics.json @@ -0,0 +1,180 @@ +{ + "new_embodiment": { + "action": { + "mean": [ + 0.4944669008255005, + 0.6347032785415649, + 0.5284651517868042, + 0.982084333896637, + 0.7498865127563477, + 0.3223438262939453 + ], + "std": [ + 0.7232871651649475, + 0.6772997975349426, + 0.7492502331733704, + 0.08576280623674361, + 0.43449321389198303, + 0.8466201424598694 + ], + "max": [ + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0 + ], + "min": [ + -1.0, + -1.0, + -1.0, + 0.0, + -1.0, + -1.0 + ], + "q01": [ + -1.0, + -1.0, + -1.0, + 0.5344227284193039, + -0.6786504459381103, + -1.0 + ], + "q99": [ + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0 + ], + "mask": [ + true, + true, + true, + true, + true, + true + ] + }, + "state": { + "mean": [ + 0.09925076365470886, + -0.038648512214422226, + 0.5415859222412109, + 0.5668906569480896, + 0.39615780115127563, + 0.3135105073451996, + 0.6150383353233337, + 0.3208809494972229, + -0.16915184259414673, + -0.031262025237083435, + -0.013657420873641968, + 0.009614686481654644, + -0.09996035695075989, + -0.07044283300638199, + -0.016576984897255898, + -0.026138588786125183, + -0.1820809543132782 + ], + "std": [ + 0.2604527175426483, + 0.3017294406890869, + 0.39846158027648926, + 0.3285336196422577, + 0.4541752338409424, + 0.3007560670375824, + 0.2675665616989136, + 0.4651698172092438, + 0.33045694231987, + 0.3108684718608856, + 0.40756914019584656, + 0.4765605628490448, + 0.5269289612770081, + 0.5424656271934509, + 0.32406172156333923, + 0.4408577084541321, + 0.5630635619163513 + ], + "max": [ + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0 + ], + "min": [ + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0 + ], + "q01": [ + -0.4896830064058304, + -0.719413949251175, + -0.669249347448349, + -0.36913287043571474, + -0.7948353081941605, + -0.6900431627035141, + -0.2929621422290802, + -0.7326109766960144, + -0.9386063790321351, + -0.5822515177726746, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0 + ], + "q99": [ + 0.6999427938461306, + 0.5288391351699846, + 0.9718325519561768, + 0.9274609363079073, + 0.9418661379814148, + 0.7052002274990088, + 0.8861017107963565, + 0.9377889406681061, + 0.7605115771293646, + 0.68762526512146, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0 + ] + }, + "num_transitions": 87052, + "num_trajectories": 90 + } +} \ No newline at end of file diff --git a/latency-aware/walker2d/vla/starvla-qwenoft-h1/walker2d_rgb_state_profile_20260912T092949Z_openvla_native_continuous_projector_sft_5k/dataset_statistics_eval.json b/latency-aware/walker2d/vla/starvla-qwenoft-h1/walker2d_rgb_state_profile_20260912T092949Z_openvla_native_continuous_projector_sft_5k/dataset_statistics_eval.json new file mode 100644 index 0000000000000000000000000000000000000000..bacad4dd9438b558720c6060d2d7d966f9283b36 --- /dev/null +++ b/latency-aware/walker2d/vla/starvla-qwenoft-h1/walker2d_rgb_state_profile_20260912T092949Z_openvla_native_continuous_projector_sft_5k/dataset_statistics_eval.json @@ -0,0 +1,180 @@ +{ + "new_embodiment": { + "action": { + "mean": [ + 0.4962558448314667, + 0.626217246055603, + 0.5222503542900085, + 0.9820914268493652, + 0.7401354312896729, + 0.3194097578525543 + ], + "std": [ + 0.7217729091644287, + 0.6857077479362488, + 0.7502380013465881, + 0.08482990413904157, + 0.44005393981933594, + 0.8470587730407715 + ], + "max": [ + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0 + ], + "min": [ + -1.0, + -1.0, + -1.0, + 0.0, + -1.0, + -1.0 + ], + "q01": [ + -1.0, + -1.0, + -1.0, + 0.5662413769960404, + -0.666657361984253, + -1.0 + ], + "q99": [ + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0 + ], + "mask": [ + true, + true, + true, + true, + true, + true + ] + }, + "state": { + "mean": [ + 0.09460366517305374, + -0.03752301260828972, + 0.5487568974494934, + 0.562300980091095, + 0.39412277936935425, + 0.3122585117816925, + 0.6065791845321655, + 0.31886041164398193, + -0.1585373878479004, + -0.03125736489892006, + -0.011883462779223919, + 0.006754553411155939, + -0.09752458333969116, + -0.07064453512430191, + -0.016498375684022903, + -0.025585465133190155, + -0.18268170952796936 + ], + "std": [ + 0.25703251361846924, + 0.306450754404068, + 0.3891076445579529, + 0.3312922716140747, + 0.45703452825546265, + 0.2963048815727234, + 0.276064395904541, + 0.4649234414100647, + 0.33390170335769653, + 0.3093952238559723, + 0.41489359736442566, + 0.4803824722766876, + 0.5347743630409241, + 0.5472173094749451, + 0.32215026021003723, + 0.4493107497692108, + 0.5645248293876648 + ], + "max": [ + 0.790878176689148, + 0.9916481971740723, + 1.0, + 0.980484127998352, + 0.9965674877166748, + 0.8545539379119873, + 0.9251037836074829, + 0.9847626686096191, + 0.987815260887146, + 0.8139623403549194, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0 + ], + "min": [ + -0.6960445642471313, + -0.8045408725738525, + -0.9902843236923218, + -0.6551779508590698, + -0.9393892288208008, + -0.9327083230018616, + -0.6576221585273743, + -0.8119556903839111, + -0.986156165599823, + -0.7416363954544067, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0 + ], + "q01": [ + -0.4923044937849045, + -0.708883695602417, + -0.5382996904850006, + -0.36911166667938233, + -0.7940302550792694, + -0.6420850110054016, + -0.28745206773281096, + -0.7374403268098831, + -0.9400115501880646, + -0.5844572883844376, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0 + ], + "q99": [ + 0.6791199707984922, + 0.546788519620895, + 0.9726598012447357, + 0.9238831114768981, + 0.944482626914978, + 0.6942272436618804, + 0.8820822370052336, + 0.9389757359027862, + 0.7519717931747423, + 0.6876130568981168, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0 + ] + }, + "num_transitions": 9678, + "num_trajectories": 10 + } +} \ No newline at end of file diff --git a/latency-aware/walker2d/vla/starvla-qwenoft-h1/walker2d_rgb_state_profile_20260912T092949Z_openvla_native_continuous_projector_sft_5k/latency_prompt_map.json b/latency-aware/walker2d/vla/starvla-qwenoft-h1/walker2d_rgb_state_profile_20260912T092949Z_openvla_native_continuous_projector_sft_5k/latency_prompt_map.json new file mode 100644 index 0000000000000000000000000000000000000000..4c8c5420e53c751882b7c759ae17715989cf8c49 --- /dev/null +++ b/latency-aware/walker2d/vla/starvla-qwenoft-h1/walker2d_rgb_state_profile_20260912T092949Z_openvla_native_continuous_projector_sft_5k/latency_prompt_map.json @@ -0,0 +1,12 @@ +{ + "1": { + "prompt": "Move the Walker2d robot forward while keeping its torso upright. Predict six continuous torques in [-1, 1] ordered as right thigh, right leg, right foot, left thigh, left leg, and left foot. Current action latency is 1 raw frames (100.00 ms). The environment runs at 10 FPS and observations are emitted at 10 FPS. Choose the best next action.", + "latency_raw_frames": 1, + "latency_ms": 100.0 + }, + "2": { + "prompt": "Move the Walker2d robot forward while keeping its torso upright. Predict six continuous torques in [-1, 1] ordered as right thigh, right leg, right foot, left thigh, left leg, and left foot. Current action latency is 2 raw frames (200.00 ms). The environment runs at 10 FPS and observations are emitted at 10 FPS. Choose the best next action.", + "latency_raw_frames": 2, + "latency_ms": 200.0 + } +} \ No newline at end of file diff --git a/latency-aware/walker2d/vla/starvla-qwenoft-h1/walker2d_rgb_state_profile_20260912T092949Z_openvla_native_continuous_projector_sft_5k/manifest.json b/latency-aware/walker2d/vla/starvla-qwenoft-h1/walker2d_rgb_state_profile_20260912T092949Z_openvla_native_continuous_projector_sft_5k/manifest.json new file mode 100644 index 0000000000000000000000000000000000000000..1be0117b7ac9b78a5d4717ad260b94bd8265224a --- /dev/null +++ b/latency-aware/walker2d/vla/starvla-qwenoft-h1/walker2d_rgb_state_profile_20260912T092949Z_openvla_native_continuous_projector_sft_5k/manifest.json @@ -0,0 +1,185 @@ +{ + "dataset_name": "walker2d_rgb_state_profile_20260912T092949Z", + "env_name": "walker2d_rgb_state", + "episodes": 90, + "frames": 87052, + "task_prompts": [ + "Move the Walker2d robot forward while keeping its torso upright. Predict six continuous torques in [-1, 1] ordered as right thigh, right leg, right foot, left thigh, left leg, and left foot. Current action latency is 1 raw frames (100.00 ms). The environment runs at 10 FPS and observations are emitted at 10 FPS. Choose the best next action.", + "Move the Walker2d robot forward while keeping its torso upright. Predict six continuous torques in [-1, 1] ordered as right thigh, right leg, right foot, left thigh, left leg, and left foot. Current action latency is 2 raw frames (200.00 ms). The environment runs at 10 FPS and observations are emitted at 10 FPS. Choose the best next action." + ], + "format": "lerobot_v2_image_parquet", + "source": "/mnt/results/latency-sensitive-bench/profile-latency-new/walker2d/20260912T092949Z-walker2d/raw", + "integration_name": "gymnasium", + "task_name": "walker2d_rgb_state", + "action_layout": "gymnasium_continuous_v1", + "action_labels": [ + "right_thigh_torque", + "right_leg_torque", + "right_foot_torque", + "left_thigh_torque", + "left_leg_torque", + "left_foot_torque" + ], + "carrier_action_labels": [ + "right_thigh_torque", + "right_leg_torque", + "right_foot_torque", + "left_thigh_torque", + "left_leg_torque", + "left_foot_torque" + ], + "action_dim": 6, + "active_action_dim": 6, + "action_carrier": "native", + "rows_unit": "decision_step", + "obs_fps": 10.0, + "obs_stride_raw_frames": 1, + "uses_state": true, + "state_dim": 17, + "state_labels": [ + "torso_height", + "torso_angle", + "right_thigh_angle", + "right_leg_angle", + "right_foot_angle", + "left_thigh_angle", + "left_leg_angle", + "left_foot_angle", + "torso_x_velocity", + "torso_z_velocity", + "torso_angular_velocity", + "right_thigh_angular_velocity", + "right_leg_angular_velocity", + "right_foot_angular_velocity", + "left_thigh_angular_velocity", + "left_leg_angular_velocity", + "left_foot_angular_velocity" + ], + "state_normalization": { + "type": "min_max", + "min": [ + 0.8002559542655945, + -0.4052939713001251, + -2.721090078353882, + -2.743088960647583, + -1.4680683612823486, + -0.7136482000350952, + -1.6744592189788818, + -1.3974136114120483, + -0.23580805957317352, + -3.591466188430786, + -10.0, + -10.0, + -10.0, + -10.0, + -10.0, + -10.0, + -10.0 + ], + "max": [ + 1.6685550212860107, + 0.997637927532196, + 0.1443042755126953, + 0.4510042369365692, + 1.3804264068603516, + 0.32474857568740845, + 0.28670230507850647, + 1.4149495363235474, + 8.862671852111816, + 3.8120992183685303, + 10.0, + 10.0, + 10.0, + 10.0, + 10.0, + 10.0, + 10.0 + ] + }, + "robot_type": "rl_games_gymnasium", + "latency_prompt_map_path": "/mnt/results/latency-sensitive-bench/profile-latency-new/walker2d/20260912T092949Z-walker2d/converted/walker2d_rgb_state_profile_20260912T092949Z/latency_prompt_map.json", + "custom_mixtures_path": "/mnt/results/latency-sensitive-bench/profile-latency-new/walker2d/20260912T092949Z-walker2d/converted/_generated_mixtures/walker2d_rgb_state_profile_20260912T092949Z.json", + "gymnasium_task": { + "action_space": { + "dtype": "float32", + "high": [ + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0 + ], + "labels": [ + "right_thigh_torque", + "right_leg_torque", + "right_foot_torque", + "left_thigh_torque", + "left_leg_torque", + "left_foot_torque" + ], + "low": [ + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0 + ], + "type": "box" + }, + "base_prompt": "Move the Walker2d robot forward while keeping its torso upright. Predict six continuous torques in [-1, 1] ordered as right thigh, right leg, right foot, left thigh, left leg, and left foot.", + "env_fps": 10.0, + "env_id": "LatencyBench/Walker2dContinuous-v0", + "frame_stack": 1, + "make_kwargs": { + "base_env_id": "Walker2d-v4", + "base_make_kwargs": { + "ctrl_cost_weight": 0.001, + "exclude_current_positions_from_observation": true, + "forward_reward_weight": 1.0, + "healthy_reward": 1.0, + "reset_noise_scale": 0.005, + "terminate_when_unhealthy": true + }, + "render_mode": "rgb_array" + }, + "noop_action": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ], + "obs_fps": 10.0, + "registration_imports": [ + "latency_bench.envs.gymnasium_walker2d" + ], + "state_space": { + "labels": [ + "torso_height", + "torso_angle", + "right_thigh_angle", + "right_leg_angle", + "right_foot_angle", + "left_thigh_angle", + "left_leg_angle", + "left_foot_angle", + "torso_x_velocity", + "torso_z_velocity", + "torso_angular_velocity", + "right_thigh_angular_velocity", + "right_leg_angular_velocity", + "right_foot_angular_velocity", + "left_thigh_angular_velocity", + "left_leg_angular_velocity", + "left_foot_angular_velocity" + ] + }, + "task_name": "walker2d_rgb_state" + }, + "validation_dataset_name": "walker2d_rgb_state_profile_20260912T092949Z__val", + "validation_episodes": 10, + "validation_frames": 9678 +} \ No newline at end of file diff --git a/latency-aware/walker2d/vla/starvla-qwenoft-h1/walker2d_rgb_state_profile_20260912T092949Z_openvla_native_continuous_projector_sft_5k/profile/DONE b/latency-aware/walker2d/vla/starvla-qwenoft-h1/walker2d_rgb_state_profile_20260912T092949Z_openvla_native_continuous_projector_sft_5k/profile/DONE new file mode 100644 index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391 diff --git a/latency-aware/walker2d/vla/starvla-qwenoft-h1/walker2d_rgb_state_profile_20260912T092949Z_openvla_native_continuous_projector_sft_5k/profile/hardware.json b/latency-aware/walker2d/vla/starvla-qwenoft-h1/walker2d_rgb_state_profile_20260912T092949Z_openvla_native_continuous_projector_sft_5k/profile/hardware.json new file mode 100644 index 0000000000000000000000000000000000000000..e8204e81807044975c0e0a0953d16d6835192f31 --- /dev/null +++ b/latency-aware/walker2d/vla/starvla-qwenoft-h1/walker2d_rgb_state_profile_20260912T092949Z_openvla_native_continuous_projector_sft_5k/profile/hardware.json @@ -0,0 +1,40 @@ +{ + "driver_version": "580.173.02", + "gpu_class": "1x-rtx3090", + "gpus": [ + { + "name": "NVIDIA GeForce RTX 3090", + "slot": 0 + } + ], + "instance_id": "instance_859cf1e47bca6046", + "topology_links": [], + "torch": { + "backends": { + "cuda_cudnn_sdp_enabled": true, + "cuda_flash_sdp_enabled": true, + "cuda_math_sdp_enabled": true, + "cuda_matmul_allow_tf32": false, + "cuda_mem_efficient_sdp_enabled": true, + "cudnn_allow_tf32": true, + "cudnn_benchmark": false + }, + "cuda_available": true, + "cuda_device_count": 1, + "cuda_version": "12.8", + "current_device": 0, + "current_device_name": "NVIDIA GeForce RTX 3090", + "device_properties": [ + { + "index": 0, + "major": 8, + "minor": 6, + "multi_processor_count": 82, + "name": "NVIDIA GeForce RTX 3090", + "total_memory": 25295257600 + } + ], + "float32_matmul_precision": "highest", + "version": "2.11.0+cu128" + } +} diff --git a/latency-aware/walker2d/vla/starvla-qwenoft-h1/walker2d_rgb_state_profile_20260912T092949Z_openvla_native_continuous_projector_sft_5k/profile/latency_burst_model.json b/latency-aware/walker2d/vla/starvla-qwenoft-h1/walker2d_rgb_state_profile_20260912T092949Z_openvla_native_continuous_projector_sft_5k/profile/latency_burst_model.json new file mode 100644 index 0000000000000000000000000000000000000000..e1e6d9259f92d6d826441e270750e7712bf1e60a --- /dev/null +++ b/latency-aware/walker2d/vla/starvla-qwenoft-h1/walker2d_rgb_state_profile_20260912T092949Z_openvla_native_continuous_projector_sft_5k/profile/latency_burst_model.json @@ -0,0 +1,1322 @@ +{ + "burst_dwell_distribution": { + "distribution_type": "inverse_cdf", + "latency_ms": [ + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0 + ], + "quantile_levels": [ + 0.0, + 0.0001, + 0.0005, + 0.001, + 0.002, + 0.003, + 0.004, + 0.005, + 0.006, + 0.007, + 0.008, + 0.009, + 0.01, + 0.02, + 0.03, + 0.04, + 0.05, + 0.06, + 0.07, + 0.08, + 0.09, + 0.1, + 0.11, + 0.12, + 0.13, + 0.14, + 0.15, + 0.16, + 0.17, + 0.18, + 0.19, + 0.2, + 0.21, + 0.22, + 0.23, + 0.24, + 0.25, + 0.26, + 0.27, + 0.28, + 0.29, + 0.3, + 0.31, + 0.32, + 0.33, + 0.34, + 0.35, + 0.36, + 0.37, + 0.38, + 0.39, + 0.4, + 0.41, + 0.42, + 0.43, + 0.44, + 0.45, + 0.46, + 0.47, + 0.48, + 0.49, + 0.5, + 0.51, + 0.52, + 0.53, + 0.54, + 0.55, + 0.56, + 0.57, + 0.58, + 0.59, + 0.6, + 0.61, + 0.62, + 0.63, + 0.64, + 0.65, + 0.66, + 0.67, + 0.68, + 0.69, + 0.7, + 0.71, + 0.72, + 0.73, + 0.74, + 0.75, + 0.76, + 0.77, + 0.78, + 0.79, + 0.8, + 0.81, + 0.82, + 0.83, + 0.84, + 0.85, + 0.86, + 0.87, + 0.88, + 0.89, + 0.9, + 0.91, + 0.92, + 0.93, + 0.94, + 0.95, + 0.96, + 0.97, + 0.98, + 0.99, + 0.991, + 0.992, + 0.993, + 0.994, + 0.995, + 0.996, + 0.997, + 0.998, + 0.999, + 0.9995, + 0.9999, + 1.0 + ] + }, + "burst_dwell_lengths": [ + 1, + 1, + 1 + ], + "burst_merge_gap_records": 30, + "burst_rank_processes": [ + { + "draw_count": 3, + "dwell_length_spearman_rho": 0.0, + "level_ms": { + "distribution_type": "inverse_cdf", + "latency_ms": [ + 86.64467597007751, + 86.64467597007751, + 86.64467597007751, + 86.64467597007751, + 86.64467597007751, + 86.64467597007751, + 86.64467597007751, + 86.64467597007751, + 86.64467597007751, + 86.64467597007751, + 86.64467597007751, + 86.64467597007751, + 86.64467597007751, + 86.64467597007751, + 86.64467597007751, + 86.64467597007751, + 86.64467597007751, + 86.64467597007751, + 86.64467597007751, + 86.64467597007751, + 86.64467597007751, + 86.64467597007751, + 86.64467597007751, + 86.64467597007751, + 86.64467597007751, + 86.64467597007751, + 86.64467597007751, + 86.64467597007751, + 86.734783680439, + 87.00510681152343, + 87.27542994260789, + 87.54575307369232, + 87.81607620477676, + 88.0863993358612, + 88.35672246694566, + 88.6270455980301, + 88.89736872911453, + 89.16769186019897, + 89.43801499128341, + 89.70833812236786, + 89.9786612534523, + 90.24898438453674, + 90.51930751562118, + 90.78963064670563, + 91.05995377779007, + 91.33027690887451, + 91.60060003995895, + 91.8709231710434, + 92.14124630212784, + 92.41156943321228, + 92.68189256429672, + 92.95221569538117, + 93.22253882646561, + 93.49286195755005, + 93.76318508863449, + 94.03350821971894, + 94.30383135080338, + 94.57415448188782, + 94.84447761297226, + 95.11480074405671, + 95.38512387514115, + 95.65544700622559, + 95.80195389032365, + 95.94846077442169, + 96.09496765851975, + 96.2414745426178, + 96.38798142671585, + 96.5344883108139, + 96.68099519491196, + 96.82750207901, + 96.97400896310806, + 97.12051584720612, + 97.26702273130417, + 97.41352961540223, + 97.56003649950027, + 97.70654338359833, + 97.85305026769637, + 97.99955715179443, + 98.1460640358925, + 98.29257091999054, + 98.4390778040886, + 98.58558468818664, + 98.7320915722847, + 98.87859845638275, + 99.0251053404808, + 99.17161222457885, + 99.31811910867691, + 99.46462599277497, + 99.61113287687301, + 99.75763976097107, + 99.90414664506912, + 100.05065352916718, + 100.19716041326522, + 100.34366729736328, + 100.49017418146133, + 100.53900980949402, + 100.53900980949402, + 100.53900980949402, + 100.53900980949402, + 100.53900980949402, + 100.53900980949402, + 100.53900980949402, + 100.53900980949402, + 100.53900980949402, + 100.53900980949402, + 100.53900980949402, + 100.53900980949402, + 100.53900980949402, + 100.53900980949402, + 100.53900980949402, + 100.53900980949402, + 100.53900980949402, + 100.53900980949402, + 100.53900980949402, + 100.53900980949402, + 100.53900980949402, + 100.53900980949402, + 100.53900980949402, + 100.53900980949402, + 100.53900980949402, + 100.53900980949402, + 100.53900980949402, + 100.53900980949402 + ], + "quantile_levels": [ + 0.0, + 0.0001, + 0.0005, + 0.001, + 0.002, + 0.003, + 0.004, + 0.005, + 0.006, + 0.007, + 0.008, + 0.009, + 0.01, + 0.02, + 0.03, + 0.04, + 0.05, + 0.06, + 0.07, + 0.08, + 0.09, + 0.1, + 0.11, + 0.12, + 0.13, + 0.14, + 0.15, + 0.16, + 0.17, + 0.18, + 0.19, + 0.2, + 0.21, + 0.22, + 0.23, + 0.24, + 0.25, + 0.26, + 0.27, + 0.28, + 0.29, + 0.3, + 0.31, + 0.32, + 0.33, + 0.34, + 0.35, + 0.36, + 0.37, + 0.38, + 0.39, + 0.4, + 0.41, + 0.42, + 0.43, + 0.44, + 0.45, + 0.46, + 0.47, + 0.48, + 0.49, + 0.5, + 0.51, + 0.52, + 0.53, + 0.54, + 0.55, + 0.56, + 0.57, + 0.58, + 0.59, + 0.6, + 0.61, + 0.62, + 0.63, + 0.64, + 0.65, + 0.66, + 0.67, + 0.68, + 0.69, + 0.7, + 0.71, + 0.72, + 0.73, + 0.74, + 0.75, + 0.76, + 0.77, + 0.78, + 0.79, + 0.8, + 0.81, + 0.82, + 0.83, + 0.84, + 0.85, + 0.86, + 0.87, + 0.88, + 0.89, + 0.9, + 0.91, + 0.92, + 0.93, + 0.94, + 0.95, + 0.96, + 0.97, + 0.98, + 0.99, + 0.991, + 0.992, + 0.993, + 0.994, + 0.995, + 0.996, + 0.997, + 0.998, + 0.999, + 0.9995, + 0.9999, + 1.0 + ] + }, + "level_residual_ms": { + "distribution_type": "inverse_cdf", + "latency_ms": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ], + "quantile_levels": [ + 0.0, + 0.0001, + 0.0005, + 0.001, + 0.002, + 0.003, + 0.004, + 0.005, + 0.006, + 0.007, + 0.008, + 0.009, + 0.01, + 0.02, + 0.03, + 0.04, + 0.05, + 0.06, + 0.07, + 0.08, + 0.09, + 0.1, + 0.11, + 0.12, + 0.13, + 0.14, + 0.15, + 0.16, + 0.17, + 0.18, + 0.19, + 0.2, + 0.21, + 0.22, + 0.23, + 0.24, + 0.25, + 0.26, + 0.27, + 0.28, + 0.29, + 0.3, + 0.31, + 0.32, + 0.33, + 0.34, + 0.35, + 0.36, + 0.37, + 0.38, + 0.39, + 0.4, + 0.41, + 0.42, + 0.43, + 0.44, + 0.45, + 0.46, + 0.47, + 0.48, + 0.49, + 0.5, + 0.51, + 0.52, + 0.53, + 0.54, + 0.55, + 0.56, + 0.57, + 0.58, + 0.59, + 0.6, + 0.61, + 0.62, + 0.63, + 0.64, + 0.65, + 0.66, + 0.67, + 0.68, + 0.69, + 0.7, + 0.71, + 0.72, + 0.73, + 0.74, + 0.75, + 0.76, + 0.77, + 0.78, + 0.79, + 0.8, + 0.81, + 0.82, + 0.83, + 0.84, + 0.85, + 0.86, + 0.87, + 0.88, + 0.89, + 0.9, + 0.91, + 0.92, + 0.93, + 0.94, + 0.95, + 0.96, + 0.97, + 0.98, + 0.99, + 0.991, + 0.992, + 0.993, + 0.994, + 0.995, + 0.996, + 0.997, + 0.998, + 0.999, + 0.9995, + 0.9999, + 1.0 + ] + }, + "severity": { + "distribution_type": "inverse_cdf", + "latency_ms": [ + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0 + ], + "quantile_levels": [ + 0.0, + 0.0001, + 0.0005, + 0.001, + 0.002, + 0.003, + 0.004, + 0.005, + 0.006, + 0.007, + 0.008, + 0.009, + 0.01, + 0.02, + 0.03, + 0.04, + 0.05, + 0.06, + 0.07, + 0.08, + 0.09, + 0.1, + 0.11, + 0.12, + 0.13, + 0.14, + 0.15, + 0.16, + 0.17, + 0.18, + 0.19, + 0.2, + 0.21, + 0.22, + 0.23, + 0.24, + 0.25, + 0.26, + 0.27, + 0.28, + 0.29, + 0.3, + 0.31, + 0.32, + 0.33, + 0.34, + 0.35, + 0.36, + 0.37, + 0.38, + 0.39, + 0.4, + 0.41, + 0.42, + 0.43, + 0.44, + 0.45, + 0.46, + 0.47, + 0.48, + 0.49, + 0.5, + 0.51, + 0.52, + 0.53, + 0.54, + 0.55, + 0.56, + 0.57, + 0.58, + 0.59, + 0.6, + 0.61, + 0.62, + 0.63, + 0.64, + 0.65, + 0.66, + 0.67, + 0.68, + 0.69, + 0.7, + 0.71, + 0.72, + 0.73, + 0.74, + 0.75, + 0.76, + 0.77, + 0.78, + 0.79, + 0.8, + 0.81, + 0.82, + 0.83, + 0.84, + 0.85, + 0.86, + 0.87, + 0.88, + 0.89, + 0.9, + 0.91, + 0.92, + 0.93, + 0.94, + 0.95, + 0.96, + 0.97, + 0.98, + 0.99, + 0.991, + 0.992, + 0.993, + 0.994, + 0.995, + 0.996, + 0.997, + 0.998, + 0.999, + 0.9995, + 0.9999, + 1.0 + ] + }, + "spike_count": 3 + } + ], + "burst_slot_rank_templates": [ + { + "dwell_length": 1, + "slot_to_rank": { + "0": 0 + } + }, + { + "dwell_length": 1, + "slot_to_rank": { + "0": 0 + } + }, + { + "dwell_length": 1, + "slot_to_rank": { + "0": 0 + } + } + ], + "model_type": "hidden_regime", + "pre_worker_time_ms": { + "distribution_type": "inverse_cdf", + "latency_ms": [ + 0.2871279716491699, + 0.2871279716491699, + 0.3000966912508011, + 0.3125195586681366, + 0.3222240471839905, + 0.3309027588367462, + 0.33220982551574707, + 0.33615071773529054, + 0.34000066518783567, + 0.3445212626457214, + 0.35144575595855715, + 0.3605480372905731, + 0.36516556739807127, + 0.417451810836792, + 0.5077130913734436, + 0.5416164875030518, + 0.5575604438781738, + 0.5679127693176269, + 0.5778009295463562, + 0.5856747627258301, + 0.5923689246177674, + 0.5970578193664551, + 0.6007380723953247, + 0.6050690412521362, + 0.6085355162620545, + 0.6125645637512207, + 0.6163313388824463, + 0.6203382730484008, + 0.6243826389312744, + 0.6274452209472656, + 0.6306731224060058, + 0.633169412612915, + 0.635418975353241, + 0.6376892566680908, + 0.6397212743759155, + 0.6422977685928345, + 0.6441904902458191, + 0.6462940216064453, + 0.6483324766159058, + 0.6513290643692017, + 0.6537760257720947, + 0.657095193862915, + 0.6599553108215332, + 0.6619826078414917, + 0.6641056776046753, + 0.6661466121673584, + 0.6684087514877319, + 0.6695426940917969, + 0.6714422941207886, + 0.6728639125823974, + 0.6742174506187439, + 0.6758885383605957, + 0.6775458693504334, + 0.6791774749755859, + 0.6806173920631409, + 0.6820759773254395, + 0.6841613054275513, + 0.6858915328979492, + 0.6874943017959595, + 0.6891242265701294, + 0.6925053238868714, + 0.6948971748352051, + 0.6978367328643799, + 0.7002923250198364, + 0.7020790815353394, + 0.7040354251861572, + 0.705716609954834, + 0.707164192199707, + 0.7092588067054748, + 0.7109628677368164, + 0.7128337740898132, + 0.7143014669418335, + 0.716428017616272, + 0.718038272857666, + 0.719979465007782, + 0.722154974937439, + 0.7243303060531616, + 0.7261616230010987, + 0.7283614039421081, + 0.7306269884109498, + 0.7336044073104858, + 0.7368490695953369, + 0.7397679209709167, + 0.7427667856216431, + 0.7457111358642579, + 0.7481389045715332, + 0.751095712184906, + 0.7529307126998901, + 0.7558730483055115, + 0.7581016540527343, + 0.7609253048896789, + 0.7642731666564941, + 0.767168390750885, + 0.7698051929473877, + 0.7736105918884277, + 0.7770744323730469, + 0.7799339890480042, + 0.7838371276855468, + 0.7889614105224609, + 0.7920980215072632, + 0.7968101620674133, + 0.8028519153594971, + 0.8096318721771241, + 0.8171645402908325, + 0.8260717749595643, + 0.8360280990600586, + 0.8494804501533508, + 0.8625362396240234, + 0.8821966767311097, + 0.919889736175537, + 0.976528728008272, + 0.9891926014423378, + 1.0017096090316777, + 1.0116808032989506, + 1.0215084385871889, + 1.0286476433277132, + 1.0351726627349853, + 1.0525703740119934, + 1.082165348529816, + 1.0893350529670716, + 1.1060434979200409, + 1.1279549598693848, + 1.1279549598693848 + ], + "quantile_levels": [ + 0.0, + 0.0001, + 0.0005, + 0.001, + 0.002, + 0.003, + 0.004, + 0.005, + 0.006, + 0.007, + 0.008, + 0.009, + 0.01, + 0.02, + 0.03, + 0.04, + 0.05, + 0.06, + 0.07, + 0.08, + 0.09, + 0.1, + 0.11, + 0.12, + 0.13, + 0.14, + 0.15, + 0.16, + 0.17, + 0.18, + 0.19, + 0.2, + 0.21, + 0.22, + 0.23, + 0.24, + 0.25, + 0.26, + 0.27, + 0.28, + 0.29, + 0.3, + 0.31, + 0.32, + 0.33, + 0.34, + 0.35, + 0.36, + 0.37, + 0.38, + 0.39, + 0.4, + 0.41, + 0.42, + 0.43, + 0.44, + 0.45, + 0.46, + 0.47, + 0.48, + 0.49, + 0.5, + 0.51, + 0.52, + 0.53, + 0.54, + 0.55, + 0.56, + 0.57, + 0.58, + 0.59, + 0.6, + 0.61, + 0.62, + 0.63, + 0.64, + 0.65, + 0.66, + 0.67, + 0.68, + 0.69, + 0.7, + 0.71, + 0.72, + 0.73, + 0.74, + 0.75, + 0.76, + 0.77, + 0.78, + 0.79, + 0.8, + 0.81, + 0.82, + 0.83, + 0.84, + 0.85, + 0.86, + 0.87, + 0.88, + 0.89, + 0.9, + 0.91, + 0.92, + 0.93, + 0.94, + 0.95, + 0.96, + 0.97, + 0.98, + 0.99, + 0.991, + 0.992, + 0.993, + 0.994, + 0.995, + 0.996, + 0.997, + 0.998, + 0.999, + 0.9995, + 0.9999, + 1.0 + ] + }, + "regime_step_counts": { + "burst": 3, + "calm": 3892 + }, + "regime_transition_counts": { + "burst": { + "burst": 0, + "calm": 3 + }, + "calm": { + "burst": 3, + "calm": 3884 + } + }, + "reset_scope": "session", + "schema_version": 12, + "spike_median_multiplier": 1.25, + "spike_threshold_ms_by_worker_slot": { + "0": 86.57148361206055 + }, + "worker_count": 1 +} diff --git a/latency-aware/walker2d/vla/starvla-qwenoft-h1/walker2d_rgb_state_profile_20260912T092949Z_openvla_native_continuous_projector_sft_5k/profile/latency_distribution.json b/latency-aware/walker2d/vla/starvla-qwenoft-h1/walker2d_rgb_state_profile_20260912T092949Z_openvla_native_continuous_projector_sft_5k/profile/latency_distribution.json new file mode 100644 index 0000000000000000000000000000000000000000..b38e74471cbc78e0c5b711175a994369bd71e08a --- /dev/null +++ b/latency-aware/walker2d/vla/starvla-qwenoft-h1/walker2d_rgb_state_profile_20260912T092949Z_openvla_native_continuous_projector_sft_5k/profile/latency_distribution.json @@ -0,0 +1,1027 @@ +{ + "schema_version": 3, + "worker_slots": { + "0": { + "all": { + "count": 3895, + "observation_to_action_latency_ms": { + "distribution_type": "inverse_cdf", + "latency_ms": [ + 64.21888089179993, + 64.21888089179993, + 64.26744388699531, + 64.46230773687363, + 64.66379324674607, + 64.89459874749184, + 65.04965202331543, + 65.14420080184937, + 65.27448462486267, + 65.32612436056137, + 65.44193125247955, + 65.54984476804734, + 65.60522656440735, + 66.6580442905426, + 67.43931007385254, + 68.07540552616119, + 68.45212668180466, + 68.74475541114808, + 68.93804426193238, + 69.10382010936738, + 69.19791586399079, + 69.27610993385315, + 69.38696476221085, + 69.47144668102264, + 69.53326894044876, + 69.58634305000305, + 69.65139049291611, + 69.70361912250519, + 69.74212255477906, + 69.77380666732788, + 69.82166377305984, + 69.85122108459473, + 69.88792041540145, + 69.91167421340943, + 69.94627538919448, + 69.97577981948852, + 70.00295335054398, + 70.03254795074463, + 70.0596283197403, + 70.08194353580475, + 70.10447882413864, + 70.1274299621582, + 70.14408979415893, + 70.16187624931335, + 70.18096616268159, + 70.20358176231385, + 70.22562551498413, + 70.24441020488739, + 70.25932915210724, + 70.28569169044495, + 70.31029176712036, + 70.33195090293884, + 70.35108722448349, + 70.37104334831238, + 70.40294820070267, + 70.42530269622803, + 70.44491225481033, + 70.46375651359558, + 70.47658450603485, + 70.49450085163116, + 70.51512625217438, + 70.53450798988342, + 70.55166764259339, + 70.57121002674103, + 70.58782054185868, + 70.60449481010437, + 70.62178468704224, + 70.64538116455078, + 70.6610111117363, + 70.68199644088745, + 70.70090284347535, + 70.71804141998291, + 70.73997579813003, + 70.76166491508484, + 70.78612376451493, + 70.81402132511138, + 70.84088504314423, + 70.87055168151855, + 70.90500563383102, + 70.93476188182831, + 70.96168086528778, + 70.98515176773071, + 71.0100483417511, + 71.03640780448913, + 71.05892330408096, + 71.08110203742982, + 71.10168570280075, + 71.1350690126419, + 71.16858087778091, + 71.19318833351136, + 71.22100743055344, + 71.25229752063751, + 71.28433256149292, + 71.30739164352417, + 71.33299107551575, + 71.3687974691391, + 71.39627569913864, + 71.435498046875, + 71.47914789915085, + 71.52327086925507, + 71.56817677021027, + 71.63192391395569, + 71.68392897844315, + 71.7507515668869, + 71.84099113941193, + 71.91955304145813, + 72.06988722085953, + 72.29740436077118, + 72.64941691160202, + 73.50347318649293, + 74.43839849233628, + 74.50578736782074, + 74.56680265903474, + 74.60346013188362, + 74.66835722446442, + 74.76408451199532, + 74.85193955421448, + 75.35295803785326, + 75.85356700181961, + 84.02688551187528, + 92.33079797923779, + 101.24016690254211, + 101.24016690254211 + ], + "quantile_levels": [ + 0.0, + 0.0001, + 0.0005, + 0.001, + 0.002, + 0.003, + 0.004, + 0.005, + 0.006, + 0.007, + 0.008, + 0.009, + 0.01, + 0.02, + 0.03, + 0.04, + 0.05, + 0.06, + 0.07, + 0.08, + 0.09, + 0.1, + 0.11, + 0.12, + 0.13, + 0.14, + 0.15, + 0.16, + 0.17, + 0.18, + 0.19, + 0.2, + 0.21, + 0.22, + 0.23, + 0.24, + 0.25, + 0.26, + 0.27, + 0.28, + 0.29, + 0.3, + 0.31, + 0.32, + 0.33, + 0.34, + 0.35, + 0.36, + 0.37, + 0.38, + 0.39, + 0.4, + 0.41, + 0.42, + 0.43, + 0.44, + 0.45, + 0.46, + 0.47, + 0.48, + 0.49, + 0.5, + 0.51, + 0.52, + 0.53, + 0.54, + 0.55, + 0.56, + 0.57, + 0.58, + 0.59, + 0.6, + 0.61, + 0.62, + 0.63, + 0.64, + 0.65, + 0.66, + 0.67, + 0.68, + 0.69, + 0.7, + 0.71, + 0.72, + 0.73, + 0.74, + 0.75, + 0.76, + 0.77, + 0.78, + 0.79, + 0.8, + 0.81, + 0.82, + 0.83, + 0.84, + 0.85, + 0.86, + 0.87, + 0.88, + 0.89, + 0.9, + 0.91, + 0.92, + 0.93, + 0.94, + 0.95, + 0.96, + 0.97, + 0.98, + 0.99, + 0.991, + 0.992, + 0.993, + 0.994, + 0.995, + 0.996, + 0.997, + 0.998, + 0.999, + 0.9995, + 0.9999, + 1.0 + ] + }, + "spearman_rho": 0.9962526864986927, + "worker_service_time_ms": { + "distribution_type": "inverse_cdf", + "latency_ms": [ + 63.82746982574463, + 63.82746982574463, + 63.92493312239647, + 64.06386149644851, + 64.26113409757615, + 64.53079591751099, + 64.67386532783509, + 64.76888129115105, + 64.8894479894638, + 64.96433173656463, + 65.04352538585663, + 65.13859939455986, + 65.21461226940156, + 66.16340847015381, + 66.85050959587097, + 67.493039393425, + 67.849141061306, + 68.12691159248352, + 68.3059456706047, + 68.43858327865601, + 68.54641152620316, + 68.63038969039917, + 68.72071390151977, + 68.80180246829987, + 68.86677287817001, + 68.91922354698181, + 68.97205686569214, + 69.01177797317504, + 69.06502404212952, + 69.10255327224732, + 69.13925127983093, + 69.1760745048523, + 69.20698500871659, + 69.23804593086243, + 69.2644897222519, + 69.29399082660674, + 69.32176500558853, + 69.34696702957153, + 69.37006133794785, + 69.39575870037079, + 69.4202886223793, + 69.44288921356201, + 69.46471593379974, + 69.48130025863648, + 69.5000395655632, + 69.51833429336548, + 69.5362491607666, + 69.5574676990509, + 69.57567809820175, + 69.59754242897034, + 69.62273967266083, + 69.63747501373291, + 69.66297578811646, + 69.68167729377747, + 69.70806875228882, + 69.72895283699036, + 69.7458103299141, + 69.76441006660461, + 69.77901536226273, + 69.79512314796447, + 69.81000846624374, + 69.82767796516418, + 69.84250817298889, + 69.86159727573394, + 69.87757424116134, + 69.89953441619873, + 69.919504404068, + 69.9412291765213, + 69.95989866256714, + 69.97936544418334, + 69.99937173128129, + 70.01988303661346, + 70.03974276781082, + 70.06396021842957, + 70.08396409749984, + 70.1135217666626, + 70.13642358779907, + 70.15888624191284, + 70.18657667636872, + 70.21063661575317, + 70.23414570093155, + 70.2652428150177, + 70.28892891407013, + 70.32105104923248, + 70.33835902214051, + 70.36597990989685, + 70.3919090628624, + 70.41169013977051, + 70.4380200624466, + 70.46836476325988, + 70.49924392700196, + 70.51999247074127, + 70.55594639778137, + 70.5849479675293, + 70.6114976644516, + 70.64248011112213, + 70.67629051208496, + 70.72400040626526, + 70.76106050014496, + 70.80091683864593, + 70.84916614294052, + 70.89379096031189, + 70.94596419334411, + 71.01326558589935, + 71.08510189056396, + 71.18451704978942, + 71.31417214870453, + 71.499387383461, + 71.86696890592575, + 72.52735633850098, + 73.49062955379486, + 73.56633122086525, + 73.62575006008149, + 73.70218905806541, + 73.82927883386613, + 73.89994262456894, + 74.03089104652405, + 74.42580151557922, + 74.97561591863632, + 83.3456894540788, + 91.62312696755146, + 100.53900980949402, + 100.53900980949402 + ], + "quantile_levels": [ + 0.0, + 0.0001, + 0.0005, + 0.001, + 0.002, + 0.003, + 0.004, + 0.005, + 0.006, + 0.007, + 0.008, + 0.009, + 0.01, + 0.02, + 0.03, + 0.04, + 0.05, + 0.06, + 0.07, + 0.08, + 0.09, + 0.1, + 0.11, + 0.12, + 0.13, + 0.14, + 0.15, + 0.16, + 0.17, + 0.18, + 0.19, + 0.2, + 0.21, + 0.22, + 0.23, + 0.24, + 0.25, + 0.26, + 0.27, + 0.28, + 0.29, + 0.3, + 0.31, + 0.32, + 0.33, + 0.34, + 0.35, + 0.36, + 0.37, + 0.38, + 0.39, + 0.4, + 0.41, + 0.42, + 0.43, + 0.44, + 0.45, + 0.46, + 0.47, + 0.48, + 0.49, + 0.5, + 0.51, + 0.52, + 0.53, + 0.54, + 0.55, + 0.56, + 0.57, + 0.58, + 0.59, + 0.6, + 0.61, + 0.62, + 0.63, + 0.64, + 0.65, + 0.66, + 0.67, + 0.68, + 0.69, + 0.7, + 0.71, + 0.72, + 0.73, + 0.74, + 0.75, + 0.76, + 0.77, + 0.78, + 0.79, + 0.8, + 0.81, + 0.82, + 0.83, + 0.84, + 0.85, + 0.86, + 0.87, + 0.88, + 0.89, + 0.9, + 0.91, + 0.92, + 0.93, + 0.94, + 0.95, + 0.96, + 0.97, + 0.98, + 0.99, + 0.991, + 0.992, + 0.993, + 0.994, + 0.995, + 0.996, + 0.997, + 0.998, + 0.999, + 0.9995, + 0.9999, + 1.0 + ] + } + }, + "steady": { + "count": 3892, + "observation_to_action_latency_ms": { + "distribution_type": "inverse_cdf", + "latency_ms": [ + 64.21888089179993, + 64.21888089179993, + 64.26729849910735, + 64.46222882843017, + 64.66291616725921, + 64.89447894287109, + 65.04953873825073, + 65.14368352890014, + 65.27354646110534, + 65.32604718589782, + 65.44180738735199, + 65.54967481040954, + 65.60513632774354, + 66.65711528778075, + 67.435560131073, + 68.07530255794525, + 68.4492192029953, + 68.74420424938202, + 68.93682938575745, + 69.10214092254638, + 69.19769524097443, + 69.27595851421356, + 69.38678356647492, + 69.47117605686188, + 69.5328344297409, + 69.58632252216339, + 69.65088012218476, + 69.70318710803986, + 69.74202722549438, + 69.77315920352936, + 69.82142517089844, + 69.85107939243316, + 69.88772933483123, + 69.9112668466568, + 69.94601192474366, + 69.97563450813294, + 70.00201094150543, + 70.0318727684021, + 70.05896360397338, + 70.08169354915618, + 70.10433425903321, + 70.1270259141922, + 70.14405165672302, + 70.16166516304015, + 70.1808714723587, + 70.20343802928925, + 70.22512376308441, + 70.24407590866089, + 70.25929422855377, + 70.28514323234558, + 70.30990863323211, + 70.33184719085693, + 70.3501536655426, + 70.37075836658478, + 70.40146433830262, + 70.42517745494843, + 70.44485967159271, + 70.46351621627808, + 70.47651810169219, + 70.4935804605484, + 70.51407109737396, + 70.53322803974152, + 70.55151409626006, + 70.5701886844635, + 70.58723092556, + 70.60385684967041, + 70.62133185863495, + 70.6435671710968, + 70.660171251297, + 70.68115920066833, + 70.7000481414795, + 70.71734981536865, + 70.73964700698852, + 70.76134051322937, + 70.78443279743195, + 70.81122475147248, + 70.83893203735352, + 70.86912979602813, + 70.90435886859893, + 70.93248003482819, + 70.96037925243378, + 70.9836455821991, + 71.00937434196472, + 71.0327023601532, + 71.05814930438996, + 71.0794553899765, + 71.10040259361267, + 71.13344577789307, + 71.16721335411071, + 71.19253223896027, + 71.22043813228608, + 71.25034034252167, + 71.28195040225982, + 71.3037918806076, + 71.33070304393769, + 71.36692314624786, + 71.39335422515869, + 71.43432743549347, + 71.4773496055603, + 71.52164331912995, + 71.56488280773164, + 71.6233865261078, + 71.67863788127899, + 71.74672186851501, + 71.83145132541657, + 71.9102623796463, + 72.05838644504547, + 72.25901246070862, + 72.62537901878358, + 73.4486136007309, + 74.34915796756744, + 74.44829755020142, + 74.52339792919159, + 74.5730178489685, + 74.61976903152465, + 74.68462880134582, + 74.80154452991485, + 74.93210744476319, + 75.54560989379881, + 76.3117787475587, + 79.51470895004282, + 86.64668917655945, + 86.64668917655945 + ], + "quantile_levels": [ + 0.0, + 0.0001, + 0.0005, + 0.001, + 0.002, + 0.003, + 0.004, + 0.005, + 0.006, + 0.007, + 0.008, + 0.009, + 0.01, + 0.02, + 0.03, + 0.04, + 0.05, + 0.06, + 0.07, + 0.08, + 0.09, + 0.1, + 0.11, + 0.12, + 0.13, + 0.14, + 0.15, + 0.16, + 0.17, + 0.18, + 0.19, + 0.2, + 0.21, + 0.22, + 0.23, + 0.24, + 0.25, + 0.26, + 0.27, + 0.28, + 0.29, + 0.3, + 0.31, + 0.32, + 0.33, + 0.34, + 0.35, + 0.36, + 0.37, + 0.38, + 0.39, + 0.4, + 0.41, + 0.42, + 0.43, + 0.44, + 0.45, + 0.46, + 0.47, + 0.48, + 0.49, + 0.5, + 0.51, + 0.52, + 0.53, + 0.54, + 0.55, + 0.56, + 0.57, + 0.58, + 0.59, + 0.6, + 0.61, + 0.62, + 0.63, + 0.64, + 0.65, + 0.66, + 0.67, + 0.68, + 0.69, + 0.7, + 0.71, + 0.72, + 0.73, + 0.74, + 0.75, + 0.76, + 0.77, + 0.78, + 0.79, + 0.8, + 0.81, + 0.82, + 0.83, + 0.84, + 0.85, + 0.86, + 0.87, + 0.88, + 0.89, + 0.9, + 0.91, + 0.92, + 0.93, + 0.94, + 0.95, + 0.96, + 0.97, + 0.98, + 0.99, + 0.991, + 0.992, + 0.993, + 0.994, + 0.995, + 0.996, + 0.997, + 0.998, + 0.999, + 0.9995, + 0.9999, + 1.0 + ] + }, + "spearman_rho": 0.9962440143954173, + "worker_service_time_ms": { + "distribution_type": "inverse_cdf", + "latency_ms": [ + 63.82746982574463, + 63.82746982574463, + 63.924849551200865, + 64.06365878295898, + 64.25989377117158, + 64.53052592849731, + 64.67336368179322, + 64.76854853630066, + 64.88874032402039, + 64.9639313735962, + 65.04222708225251, + 65.13830837059021, + 65.21415251731872, + 66.16300260543824, + 66.84763409614563, + 67.49238602161408, + 67.84865987300873, + 68.12611276626586, + 68.30496686458588, + 68.43773768901825, + 68.54610390186309, + 68.6302877664566, + 68.7196520614624, + 68.8012242269516, + 68.8665434885025, + 68.9189061164856, + 68.97146592140197, + 69.01117464065551, + 69.06471957206726, + 69.10236324310303, + 69.13914971351623, + 69.17492520809174, + 69.20655922412872, + 69.23778124332428, + 69.26441912174225, + 69.29358471870422, + 69.32019448280334, + 69.34652063846588, + 69.3698172712326, + 69.39486378192902, + 69.41827408790589, + 69.44232745170594, + 69.46447602272033, + 69.48072097301483, + 69.49991744995117, + 69.51753581523896, + 69.53571593761444, + 69.55461455821991, + 69.57534356594086, + 69.59614387512207, + 69.62158482074737, + 69.63672277927398, + 69.66229291915893, + 69.68103442192077, + 69.70800340175629, + 69.72816224575043, + 69.7450707912445, + 69.76414365768433, + 69.77862896442413, + 69.79452700138091, + 69.80988729000092, + 69.8276025056839, + 69.84054414749146, + 69.86028493881226, + 69.87702286720275, + 69.89840402126312, + 69.9181250333786, + 69.94080069541931, + 69.95955774307251, + 69.97868758678436, + 69.99920504570008, + 70.01864914894104, + 70.03822797775268, + 70.06183682441711, + 70.08342313289643, + 70.11323342323303, + 70.1355171918869, + 70.15791974544526, + 70.18509419441223, + 70.20979231834411, + 70.23125278472901, + 70.26408920288085, + 70.28767484664917, + 70.31905544281005, + 70.33731842041016, + 70.36446030139923, + 70.38798296451569, + 70.41008462905884, + 70.43654942035676, + 70.46712791919708, + 70.49677614212037, + 70.51846179962158, + 70.55446595668792, + 70.58218722820281, + 70.60933103084564, + 70.64008875370025, + 70.67384169101715, + 70.72326898097992, + 70.75816891670227, + 70.79773238182068, + 70.84490663528443, + 70.89026074409485, + 70.94181303977966, + 71.01094079017639, + 71.08044451713562, + 71.17663326740265, + 71.31091842651367, + 71.48311145305634, + 71.83137361526488, + 72.4928621673584, + 73.34201078414917, + 73.49197626876831, + 73.5695899362564, + 73.6369322757721, + 73.70509721183777, + 73.8460913658142, + 73.91988869094848, + 74.07588002777099, + 74.60754434394833, + 75.38980871009836, + 78.84191055488596, + 85.96448826789856, + 85.96448826789856 + ], + "quantile_levels": [ + 0.0, + 0.0001, + 0.0005, + 0.001, + 0.002, + 0.003, + 0.004, + 0.005, + 0.006, + 0.007, + 0.008, + 0.009, + 0.01, + 0.02, + 0.03, + 0.04, + 0.05, + 0.06, + 0.07, + 0.08, + 0.09, + 0.1, + 0.11, + 0.12, + 0.13, + 0.14, + 0.15, + 0.16, + 0.17, + 0.18, + 0.19, + 0.2, + 0.21, + 0.22, + 0.23, + 0.24, + 0.25, + 0.26, + 0.27, + 0.28, + 0.29, + 0.3, + 0.31, + 0.32, + 0.33, + 0.34, + 0.35, + 0.36, + 0.37, + 0.38, + 0.39, + 0.4, + 0.41, + 0.42, + 0.43, + 0.44, + 0.45, + 0.46, + 0.47, + 0.48, + 0.49, + 0.5, + 0.51, + 0.52, + 0.53, + 0.54, + 0.55, + 0.56, + 0.57, + 0.58, + 0.59, + 0.6, + 0.61, + 0.62, + 0.63, + 0.64, + 0.65, + 0.66, + 0.67, + 0.68, + 0.69, + 0.7, + 0.71, + 0.72, + 0.73, + 0.74, + 0.75, + 0.76, + 0.77, + 0.78, + 0.79, + 0.8, + 0.81, + 0.82, + 0.83, + 0.84, + 0.85, + 0.86, + 0.87, + 0.88, + 0.89, + 0.9, + 0.91, + 0.92, + 0.93, + 0.94, + 0.95, + 0.96, + 0.97, + 0.98, + 0.99, + 0.991, + 0.992, + 0.993, + 0.994, + 0.995, + 0.996, + 0.997, + 0.998, + 0.999, + 0.9995, + 0.9999, + 1.0 + ] + } + } + } + } +} diff --git a/latency-aware/walker2d/vla/starvla-qwenoft-h1/walker2d_rgb_state_profile_20260912T092949Z_openvla_native_continuous_projector_sft_5k/profile/latency_profile.png b/latency-aware/walker2d/vla/starvla-qwenoft-h1/walker2d_rgb_state_profile_20260912T092949Z_openvla_native_continuous_projector_sft_5k/profile/latency_profile.png new file mode 100644 index 0000000000000000000000000000000000000000..9391681923c8583792d0d92cd799fc0464b0b68c Binary files /dev/null and b/latency-aware/walker2d/vla/starvla-qwenoft-h1/walker2d_rgb_state_profile_20260912T092949Z_openvla_native_continuous_projector_sft_5k/profile/latency_profile.png differ diff --git a/latency-aware/walker2d/vla/starvla-qwenoft-h1/walker2d_rgb_state_profile_20260912T092949Z_openvla_native_continuous_projector_sft_5k/profile/profile.json b/latency-aware/walker2d/vla/starvla-qwenoft-h1/walker2d_rgb_state_profile_20260912T092949Z_openvla_native_continuous_projector_sft_5k/profile/profile.json new file mode 100644 index 0000000000000000000000000000000000000000..4877e60530800330ac998c3f9aa83070a936702d --- /dev/null +++ b/latency-aware/walker2d/vla/starvla-qwenoft-h1/walker2d_rgb_state_profile_20260912T092949Z_openvla_native_continuous_projector_sft_5k/profile/profile.json @@ -0,0 +1,66 @@ +{ + "burst_model_path": "latency_burst_model.json", + "distribution_path": "latency_distribution.json", + "env_fps": 10, + "frame_ms": 100.0, + "gpu_class": "1x-rtx3090", + "instance_id": "instance_859cf1e47bca6046", + "latency_kind": "observation_to_action_latency", + "latency_method": "temporal", + "model_id": "qwenoft", + "n_admitted_observations": 3895, + "n_capacity_drops": 0, + "n_observation_attempts": 3895, + "per_slot_summary": { + "0": { + "admitted_count": 3895, + "mean_observation_to_action_latency_ms": 70.49593999162413, + "mean_worker_service_time_ms": 69.79960327613645, + "p95_observation_to_action_latency_ms": 72.06934230327606, + "p95_worker_service_time_ms": 71.31323595046997, + "p99_worker_service_time_ms": 73.49043211936952 + } + }, + "provenance": { + "base_config": "/workspace/tasks/20260912T092949Z-walker2d/profile.yaml", + "checkpoint_kind": "best", + "model_artifact": { + "checkpoint": "checkpoints/steps_5000_pytorch_model.pt", + "model_config": "config.full.yaml", + "path_in_repo": "OpenVLA/zero-latency/walker2d_rgb_state_l0_return_gt3000_100ep_openvla_native_continuous_projector_sft_5k", + "repo_id": "latency-sensitive-bench/extra-envs-checkpoints", + "source": "local" + }, + "session_ids": [ + 0, + 1, + 2, + 3, + 4 + ] + }, + "sample_model_type": "hidden_regime", + "source_run_id": "20260912T093545326611Z", + "summary": { + "frame_ms": 100.0, + "max_ms": 101.24016690254211, + "mean_effective_frames": 0.7049593999162412, + "mean_ms": 70.49593999162413, + "min_ms": 64.21888089179993, + "n_samples": 3895, + "p50_frames": 0.7053450798988342, + "p50_ms": 70.53450798988342, + "p90_frames": 0.7163049759864808, + "p90_ms": 71.63049759864808, + "p95_frames": 0.7206934230327606, + "p95_ms": 72.06934230327606, + "p99_frames": 0.7443600579738617, + "p99_ms": 74.43600579738617, + "prob_latency_gt_1_frame": 0.00025673940949935817, + "prob_latency_gt_2_frames": 0.0, + "prob_latency_gt_3_frames": 0.0, + "std_ms": 1.483596848515007 + }, + "visualization_path": "latency_profile.png", + "workload_id": "walker2d" +} diff --git a/latency-aware/walker2d/vla/starvla-qwenoft-h1/walker2d_rgb_state_profile_20260912T092949Z_openvla_native_continuous_projector_sft_5k/provenance.json b/latency-aware/walker2d/vla/starvla-qwenoft-h1/walker2d_rgb_state_profile_20260912T092949Z_openvla_native_continuous_projector_sft_5k/provenance.json new file mode 100644 index 0000000000000000000000000000000000000000..01971aee32d2e0f569bdb7beda56d23e724d1b49 --- /dev/null +++ b/latency-aware/walker2d/vla/starvla-qwenoft-h1/walker2d_rgb_state_profile_20260912T092949Z_openvla_native_continuous_projector_sft_5k/provenance.json @@ -0,0 +1,34 @@ +{ + "task": "walker2d", + "model": "qwenoft", + "training_condition": "latency-aware", + "training_run_id": "walker2d_rgb_state_profile_20260912T092949Z_openvla_native_continuous_projector_sft_5k", + "source": { + "repo_id": "latency-sensitive-bench/Standard-Pipeline", + "revision": "2208875f92b21e3f8ec2363a30a6705dfe56cc04", + "path": "walker2d/profile_latency/OpenVLA", + "url": "https://huggingface.co/latency-sensitive-bench/Standard-Pipeline/tree/2208875f92b21e3f8ec2363a30a6705dfe56cc04/walker2d/profile_latency/OpenVLA" + }, + "hf_repo_id": "latency-sensitive-bench/benchmark-models", + "hf_path": "latency-aware/walker2d/vla/starvla-qwenoft-h1/walker2d_rgb_state_profile_20260912T092949Z_openvla_native_continuous_projector_sft_5k", + "checkpoint": { + "source_file": "walker2d/profile_latency/OpenVLA/checkpoints/steps_5000_pytorch_model.pt", + "source_sha256": "cbd8a14cfa3fe1a97b6d131294e0264e43e0267943189f776b4191922ee4559d", + "source_bytes": 9785142689, + "file": "checkpoints/model.pt", + "selection_rule": "only published VLA checkpoint", + "method": "copy", + "sha256": "cbd8a14cfa3fe1a97b6d131294e0264e43e0267943189f776b4191922ee4559d", + "bytes": 9785142689 + }, + "config_source": "walker2d/profile_latency/OpenVLA/config.full.yaml", + "task_contract_file": "task_contract.json", + "missing_source_files": [], + "migration": { + "operation": "copy_selected_inference_bundle", + "source_modified": false, + "evaluation_performed": false, + "experiment_status": "preserved_in_source_evidence" + }, + "action_horizon": 1 +} diff --git a/latency-aware/walker2d/vla/starvla-qwenoft-h1/walker2d_rgb_state_profile_20260912T092949Z_openvla_native_continuous_projector_sft_5k/source/config.yaml b/latency-aware/walker2d/vla/starvla-qwenoft-h1/walker2d_rgb_state_profile_20260912T092949Z_openvla_native_continuous_projector_sft_5k/source/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..de220b0ee72d57420250b87504fd49623810b2e8 --- /dev/null +++ b/latency-aware/walker2d/vla/starvla-qwenoft-h1/walker2d_rgb_state_profile_20260912T092949Z_openvla_native_continuous_projector_sft_5k/source/config.yaml @@ -0,0 +1,87 @@ +checkpoint: + local: + keep_last_n: 1 + save_best_model: false + save_final_model: true + save_pt_file: true + save_safetensors_file: false + save_training_state: false + sync: + enabled: false + keep_last_n: 0 + repo_id: null +datasets: + vla_data: + data_mix: walker2d_rgb_state_profile_20260912T092949Z + dataset_py: lerobot_datasets + eval_data_mix: walker2d_rgb_state_profile_20260912T092949Z__val + latency_curriculum: + enabled: false + per_device_batch_size: 16 +framework: + action_model: + action_dim: 6 + action_env_dim: 6 + action_hidden_dim: 2560 + action_horizon: 1 + action_model_type: MLP + loss_type: l1 + state_dim: 17 + state_encoding: continuous_projector + task_objective: null + kv_memory: + enabled: false + packed_train: false + rebased_sink: true + rollout_len: 8 + window: 4 + name: QwenOFT + qwenvl: + attn_implementation: flash_attention_2 + base_vlm: /mnt/checkpoints/latency-sensitive-bench/openvla/Qwen3-VL-4B-Instruct + enable_gradient_checkpointing: true +output_dir: /mnt/results/latency-sensitive-bench/profile-latency-new/walker2d/20260912T092949Z-walker2d/vla/walker2d_rgb_state_profile_20260912T092949Z_openvla_native_continuous_projector_sft_5k +rl_games: + env_eval: + enabled: false + task: gymnasium +run_id: walker2d_rgb_state_profile_20260912T092949Z_openvla_native_continuous_projector_sft_5k +run_root_dir: /mnt/results/latency-sensitive-bench/profile-latency-new/walker2d/20260912T092949Z-walker2d/vla +seed: 42 +trainer: + distributed_backend: deepspeed + eval_action_classification: false + eval_action_classification_interval: null + eval_interval: 500 + eval_num_batches: 200 + freeze_llm_layers: [] + freeze_modules: '' + freeze_tied_embedding: false + freeze_vit: false + gradient_accumulation_steps: 1 + is_resume: false + learning_rate: + action_model: 0.0001 + base: 2.0e-05 + qwen_vl_interface: 1.0e-05 + logging_frequency: 1 + lr_scheduler_type: cosine_with_min_lr + max_train_steps: 5000 + num_warmup_steps: 100 + optimizer: + betas: + - 0.9 + - 0.95 + eps: 1.0e-08 + fused: true + weight_decay: 1.0e-08 + per_latency_eval_num_batches: null + pretrained_checkpoint: null + profile_timing: + enabled: true + log_interval: 10 + save_interval: 500 + scheduler_specific_kwargs: + min_lr: 1.0e-06 +wandb_entity: dongqianyu99-zhejiang-university +wandb_project: latency-sensitive-bench diff --git a/latency-aware/walker2d/vla/starvla-qwenoft-h1/walker2d_rgb_state_profile_20260912T092949Z_openvla_native_continuous_projector_sft_5k/source/provenance.json b/latency-aware/walker2d/vla/starvla-qwenoft-h1/walker2d_rgb_state_profile_20260912T092949Z_openvla_native_continuous_projector_sft_5k/source/provenance.json new file mode 100644 index 0000000000000000000000000000000000000000..01827063664b06956ce0b6b655c51e282dcf8825 --- /dev/null +++ b/latency-aware/walker2d/vla/starvla-qwenoft-h1/walker2d_rgb_state_profile_20260912T092949Z_openvla_native_continuous_projector_sft_5k/source/provenance.json @@ -0,0 +1,1044 @@ +{ + "run_id": "20260912T092949Z-walker2d", + "final_step": 5000, + "exit_code": 0, + "checkpoint_name": "steps_5000_pytorch_model.pt", + "checkpoint_size": 9785142689, + "checkpoint_sha256": "cbd8a14cfa3fe1a97b6d131294e0264e43e0267943189f776b4191922ee4559d", + "state_dict_entries": 732, + "state_dict_key_examples": [ + "qwen_vl_interface.model.model.visual.patch_embed.proj.weight", + "qwen_vl_interface.model.model.visual.patch_embed.proj.bias", + "qwen_vl_interface.model.model.visual.pos_embed.weight", + "qwen_vl_interface.model.model.visual.blocks.0.norm1.weight", + "qwen_vl_interface.model.model.visual.blocks.0.norm1.bias" + ], + "checkpoint_content": "model state dict only", + "wandb": { + "state": "finished", + "summary": { + "_runtime": 12916.753093822, + "_step": 5000, + "_timestamp": 1789225672.5963194, + "_wandb.runtime": 12916, + "batch/effective_tokens": 2848, + "batch/image_count": 16, + "batch/input_len_max": 178, + "batch/input_len_mean": 178, + "batch/padding_ratio": 0, + "batch/pixel_values_rows": 4096, + "batch/size": 16, + "epoch": 0.92, + "eval/action_loss/samples": 3200, + "eval/action_loss/seconds": 29.096430673002033, + "eval/latency_1/loss": 0.036322228610515594, + "eval/loss": 0.036322228610515594, + "eval/walker2d_rgb_state/latency_1/loss": 0.036322228610515594, + "eval/walker2d_rgb_state/loss": 0.036322228610515594, + "learning_rate/action_model": 5e-06, + "learning_rate/qwen_vl_interface": 5e-07, + "throughput/effective_tokens_per_sec": 1133.0846952166062, + "throughput/samples_per_sec": 6.365644355149473, + "timing/action_head_loss": 0.0017446179990656674, + "timing/backward": 0.7881747860228643, + "timing/checkpoint_total": 13.579471566015854, + "timing/data": 0.0004350999952293933, + "timing/dataloader_next": 0.0008882379916030914, + "timing/eval_action_loss_total": 29.099037520994894, + "timing/forward": 0.14932741899974644, + "timing/log_metrics_total": 0.006318414001725614, + "timing/lr_scheduler": 8.731201523914933e-05, + "timing/model": 2.513312766997842, + "timing/optimizer_step": 1.5714728670136535, + "timing/qwen_h2d": 0.0032470029836986214, + "timing/qwen_input_build_total": 0.022446599003160372, + "timing/qwen_processor": 0.018632816994795576, + "timing/train_step_total": 2.5134926030004863, + "timing/vlm_forward": 0.12444130500080064, + "train/grad_norm_pre_clip": 7.462423324584961, + "train/loss": 0.02517159841954708 + }, + "url": "https://wandb.ai/dongqianyu99-zhejiang-university/latency-sensitive-bench/runs/vmndk81h", + "exit_code": 0, + "completed_utc": "2026-09-12T15:07:57Z" + }, + "dataset_upload": [ + { + "repo_id": "latency-sensitive-bench/extra-envs-rollouts", + "repo_type": "dataset", + "prefix": "profile-latency-new/walker2d/20260912T092949Z-walker2d/raw", + "commit": "139aa8eb7e854ffe4f54a3d83e069e0fba8357db", + "url": "https://huggingface.co/datasets/latency-sensitive-bench/extra-envs-rollouts/commit/139aa8eb7e854ffe4f54a3d83e069e0fba8357db", + "verified_files": 9, + "bytes": 2138512099, + "verification": "Every file size; Git blob SHA1 for regular files and SHA256 for LFS files" + }, + { + "repo_id": "latency-sensitive-bench/extra-envs-rollouts", + "repo_type": "dataset", + "prefix": "profile-latency-new/walker2d/20260912T092949Z-walker2d/converted", + "commit": "a29d807281c68338b681a3bb6360d4cc8f497c69", + "url": "https://huggingface.co/datasets/latency-sensitive-bench/extra-envs-rollouts/commit/a29d807281c68338b681a3bb6360d4cc8f497c69", + "verified_files": 120, + "bytes": 2136992268, + "verification": "Every file size; Git blob SHA1 for regular files and SHA256 for LFS files" + } + ], + "data_validation": { + "episodes": [ + { + "split": "train", + "episode_idx": 0, + "seed": 9, + "actual_rows": 726, + "actual_return": 3208.9899393320084 + }, + { + "split": "train", + "episode_idx": 2, + "seed": 2, + "actual_rows": 863, + "actual_return": 3862.9185177087784 + }, + { + "split": "train", + "episode_idx": 4, + "seed": 1, + "actual_rows": 1000, + "actual_return": 4582.548987925053 + }, + { + "split": "train", + "episode_idx": 5, + "seed": 3, + "actual_rows": 1000, + "actual_return": 4738.329069197178 + }, + { + "split": "train", + "episode_idx": 6, + "seed": 4, + "actual_rows": 1000, + "actual_return": 4568.033828258514 + }, + { + "split": "train", + "episode_idx": 8, + "seed": 7, + "actual_rows": 1000, + "actual_return": 4388.061577618122 + }, + { + "split": "train", + "episode_idx": 9, + "seed": 8, + "actual_rows": 1000, + "actual_return": 4624.5429965257645 + }, + { + "split": "train", + "episode_idx": 10, + "seed": 10, + "actual_rows": 1000, + "actual_return": 4646.167462468147 + }, + { + "split": "train", + "episode_idx": 11, + "seed": 11, + "actual_rows": 1000, + "actual_return": 4686.4867441654205 + }, + { + "split": "train", + "episode_idx": 12, + "seed": 14, + "actual_rows": 831, + "actual_return": 3838.5751860141754 + }, + { + "split": "train", + "episode_idx": 13, + "seed": 12, + "actual_rows": 1000, + "actual_return": 4480.943344056606 + }, + { + "split": "train", + "episode_idx": 14, + "seed": 13, + "actual_rows": 1000, + "actual_return": 4412.779609620571 + }, + { + "split": "train", + "episode_idx": 15, + "seed": 15, + "actual_rows": 1000, + "actual_return": 4593.643612384796 + }, + { + "split": "train", + "episode_idx": 16, + "seed": 16, + "actual_rows": 1000, + "actual_return": 4670.15686494112 + }, + { + "split": "train", + "episode_idx": 18, + "seed": 18, + "actual_rows": 1000, + "actual_return": 4619.547595500946 + }, + { + "split": "train", + "episode_idx": 19, + "seed": 19, + "actual_rows": 1000, + "actual_return": 4702.1127672195435 + }, + { + "split": "train", + "episode_idx": 20, + "seed": 20, + "actual_rows": 1000, + "actual_return": 4361.072299003601 + }, + { + "split": "train", + "episode_idx": 21, + "seed": 21, + "actual_rows": 1000, + "actual_return": 4649.568623423576 + }, + { + "split": "train", + "episode_idx": 22, + "seed": 22, + "actual_rows": 1000, + "actual_return": 4509.352912843227 + }, + { + "split": "train", + "episode_idx": 23, + "seed": 23, + "actual_rows": 1000, + "actual_return": 4572.384602069855 + }, + { + "split": "train", + "episode_idx": 24, + "seed": 24, + "actual_rows": 1000, + "actual_return": 4632.64072316885 + }, + { + "split": "train", + "episode_idx": 26, + "seed": 27, + "actual_rows": 1000, + "actual_return": 4528.378586828709 + }, + { + "split": "train", + "episode_idx": 27, + "seed": 30, + "actual_rows": 929, + "actual_return": 4166.440938651562 + }, + { + "split": "train", + "episode_idx": 28, + "seed": 28, + "actual_rows": 1000, + "actual_return": 4701.811249375343 + }, + { + "split": "train", + "episode_idx": 30, + "seed": 31, + "actual_rows": 1000, + "actual_return": 4701.488458037376 + }, + { + "split": "train", + "episode_idx": 31, + "seed": 32, + "actual_rows": 1000, + "actual_return": 4556.560275912285 + }, + { + "split": "train", + "episode_idx": 32, + "seed": 33, + "actual_rows": 1000, + "actual_return": 4491.865241408348 + }, + { + "split": "train", + "episode_idx": 33, + "seed": 34, + "actual_rows": 1000, + "actual_return": 4547.573637366295 + }, + { + "split": "train", + "episode_idx": 34, + "seed": 35, + "actual_rows": 1000, + "actual_return": 4567.3020188212395 + }, + { + "split": "train", + "episode_idx": 35, + "seed": 36, + "actual_rows": 768, + "actual_return": 3382.415802717209 + }, + { + "split": "train", + "episode_idx": 36, + "seed": 37, + "actual_rows": 1000, + "actual_return": 4489.921771645546 + }, + { + "split": "train", + "episode_idx": 37, + "seed": 38, + "actual_rows": 1000, + "actual_return": 4609.555367290974 + }, + { + "split": "train", + "episode_idx": 38, + "seed": 39, + "actual_rows": 1000, + "actual_return": 4564.513348519802 + }, + { + "split": "train", + "episode_idx": 39, + "seed": 40, + "actual_rows": 1000, + "actual_return": 4644.1978949308395 + }, + { + "split": "train", + "episode_idx": 40, + "seed": 42, + "actual_rows": 1000, + "actual_return": 4527.426118969917 + }, + { + "split": "train", + "episode_idx": 41, + "seed": 43, + "actual_rows": 1000, + "actual_return": 4609.569993317127 + }, + { + "split": "train", + "episode_idx": 42, + "seed": 44, + "actual_rows": 1000, + "actual_return": 4618.82773822546 + }, + { + "split": "train", + "episode_idx": 43, + "seed": 45, + "actual_rows": 1000, + "actual_return": 4585.388746500015 + }, + { + "split": "train", + "episode_idx": 44, + "seed": 46, + "actual_rows": 1000, + "actual_return": 4524.564453125 + }, + { + "split": "train", + "episode_idx": 45, + "seed": 47, + "actual_rows": 1000, + "actual_return": 4480.179334104061 + }, + { + "split": "train", + "episode_idx": 46, + "seed": 48, + "actual_rows": 1000, + "actual_return": 4740.075881540775 + }, + { + "split": "train", + "episode_idx": 48, + "seed": 50, + "actual_rows": 971, + "actual_return": 4310.374913454056 + }, + { + "split": "train", + "episode_idx": 49, + "seed": 51, + "actual_rows": 1000, + "actual_return": 4453.313854634762 + }, + { + "split": "train", + "episode_idx": 50, + "seed": 53, + "actual_rows": 882, + "actual_return": 4043.457126915455 + }, + { + "split": "train", + "episode_idx": 51, + "seed": 57, + "actual_rows": 862, + "actual_return": 3862.672888338566 + }, + { + "split": "train", + "episode_idx": 52, + "seed": 54, + "actual_rows": 885, + "actual_return": 3997.717103123665 + }, + { + "split": "train", + "episode_idx": 53, + "seed": 52, + "actual_rows": 1000, + "actual_return": 4650.883183002472 + }, + { + "split": "train", + "episode_idx": 54, + "seed": 55, + "actual_rows": 1000, + "actual_return": 4628.249386608601 + }, + { + "split": "train", + "episode_idx": 55, + "seed": 56, + "actual_rows": 1000, + "actual_return": 4352.901657938957 + }, + { + "split": "train", + "episode_idx": 56, + "seed": 58, + "actual_rows": 1000, + "actual_return": 4398.224647581577 + }, + { + "split": "train", + "episode_idx": 57, + "seed": 59, + "actual_rows": 1000, + "actual_return": 4583.017430663109 + }, + { + "split": "train", + "episode_idx": 59, + "seed": 61, + "actual_rows": 1000, + "actual_return": 4559.465924978256 + }, + { + "split": "train", + "episode_idx": 60, + "seed": 63, + "actual_rows": 1000, + "actual_return": 4380.1740090847015 + }, + { + "split": "train", + "episode_idx": 61, + "seed": 64, + "actual_rows": 1000, + "actual_return": 4435.523238182068 + }, + { + "split": "train", + "episode_idx": 62, + "seed": 65, + "actual_rows": 1000, + "actual_return": 4367.5251343250275 + }, + { + "split": "train", + "episode_idx": 63, + "seed": 67, + "actual_rows": 963, + "actual_return": 4345.349503159523 + }, + { + "split": "train", + "episode_idx": 64, + "seed": 66, + "actual_rows": 1000, + "actual_return": 4513.539445400238 + }, + { + "split": "train", + "episode_idx": 65, + "seed": 73, + "actual_rows": 773, + "actual_return": 3490.61042958498 + }, + { + "split": "train", + "episode_idx": 66, + "seed": 68, + "actual_rows": 1000, + "actual_return": 4431.87359136343 + }, + { + "split": "train", + "episode_idx": 67, + "seed": 69, + "actual_rows": 1000, + "actual_return": 4721.685309708118 + }, + { + "split": "train", + "episode_idx": 68, + "seed": 71, + "actual_rows": 1000, + "actual_return": 4518.990444839001 + }, + { + "split": "train", + "episode_idx": 69, + "seed": 72, + "actual_rows": 1000, + "actual_return": 4513.488130092621 + }, + { + "split": "train", + "episode_idx": 70, + "seed": 76, + "actual_rows": 831, + "actual_return": 3717.469878733158 + }, + { + "split": "train", + "episode_idx": 71, + "seed": 75, + "actual_rows": 1000, + "actual_return": 4569.386827468872 + }, + { + "split": "train", + "episode_idx": 72, + "seed": 77, + "actual_rows": 1000, + "actual_return": 4395.308935403824 + }, + { + "split": "train", + "episode_idx": 73, + "seed": 78, + "actual_rows": 1000, + "actual_return": 4632.641280531883 + }, + { + "split": "train", + "episode_idx": 74, + "seed": 79, + "actual_rows": 1000, + "actual_return": 4591.099643826485 + }, + { + "split": "train", + "episode_idx": 75, + "seed": 80, + "actual_rows": 1000, + "actual_return": 4460.989606976509 + }, + { + "split": "train", + "episode_idx": 76, + "seed": 81, + "actual_rows": 1000, + "actual_return": 4685.384257674217 + }, + { + "split": "train", + "episode_idx": 78, + "seed": 83, + "actual_rows": 1000, + "actual_return": 4467.1219418644905 + }, + { + "split": "train", + "episode_idx": 79, + "seed": 84, + "actual_rows": 1000, + "actual_return": 4601.587870895863 + }, + { + "split": "train", + "episode_idx": 80, + "seed": 85, + "actual_rows": 1000, + "actual_return": 4236.5026268959045 + }, + { + "split": "train", + "episode_idx": 82, + "seed": 88, + "actual_rows": 871, + "actual_return": 3978.1818378567696 + }, + { + "split": "train", + "episode_idx": 83, + "seed": 97, + "actual_rows": 710, + "actual_return": 3069.413112640381 + }, + { + "split": "train", + "episode_idx": 84, + "seed": 89, + "actual_rows": 1000, + "actual_return": 4578.044221043587 + }, + { + "split": "train", + "episode_idx": 85, + "seed": 90, + "actual_rows": 1000, + "actual_return": 4596.4438044428825 + }, + { + "split": "train", + "episode_idx": 86, + "seed": 91, + "actual_rows": 1000, + "actual_return": 4548.513531446457 + }, + { + "split": "train", + "episode_idx": 87, + "seed": 92, + "actual_rows": 1000, + "actual_return": 4794.796605825424 + }, + { + "split": "train", + "episode_idx": 88, + "seed": 93, + "actual_rows": 1000, + "actual_return": 4620.774623036385 + }, + { + "split": "train", + "episode_idx": 89, + "seed": 94, + "actual_rows": 1000, + "actual_return": 4547.734396517277 + }, + { + "split": "train", + "episode_idx": 90, + "seed": 95, + "actual_rows": 1000, + "actual_return": 4530.390145301819 + }, + { + "split": "train", + "episode_idx": 91, + "seed": 96, + "actual_rows": 1000, + "actual_return": 4481.2135281562805 + }, + { + "split": "train", + "episode_idx": 92, + "seed": 98, + "actual_rows": 1000, + "actual_return": 4478.202645242214 + }, + { + "split": "train", + "episode_idx": 93, + "seed": 100, + "actual_rows": 730, + "actual_return": 3204.760389685631 + }, + { + "split": "train", + "episode_idx": 94, + "seed": 99, + "actual_rows": 1000, + "actual_return": 4270.546193242073 + }, + { + "split": "train", + "episode_idx": 95, + "seed": 101, + "actual_rows": 889, + "actual_return": 3995.776729762554 + }, + { + "split": "train", + "episode_idx": 96, + "seed": 106, + "actual_rows": 716, + "actual_return": 3150.100844144821 + }, + { + "split": "train", + "episode_idx": 97, + "seed": 102, + "actual_rows": 1000, + "actual_return": 4725.3805629611015 + }, + { + "split": "train", + "episode_idx": 98, + "seed": 107, + "actual_rows": 852, + "actual_return": 3911.3024450540543 + }, + { + "split": "train", + "episode_idx": 99, + "seed": 104, + "actual_rows": 1000, + "actual_return": 4545.101168990135 + }, + { + "split": "val", + "episode_idx": 1, + "seed": 0, + "actual_rows": 766, + "actual_return": 3383.400418162346 + }, + { + "split": "val", + "episode_idx": 3, + "seed": 5, + "actual_rows": 912, + "actual_return": 4112.48240852356 + }, + { + "split": "val", + "episode_idx": 7, + "seed": 6, + "actual_rows": 1000, + "actual_return": 4603.241932570934 + }, + { + "split": "val", + "episode_idx": 17, + "seed": 17, + "actual_rows": 1000, + "actual_return": 4604.767876505852 + }, + { + "split": "val", + "episode_idx": 25, + "seed": 25, + "actual_rows": 1000, + "actual_return": 4709.377348542213 + }, + { + "split": "val", + "episode_idx": 29, + "seed": 29, + "actual_rows": 1000, + "actual_return": 4720.668054103851 + }, + { + "split": "val", + "episode_idx": 47, + "seed": 49, + "actual_rows": 1000, + "actual_return": 4667.626155257225 + }, + { + "split": "val", + "episode_idx": 58, + "seed": 60, + "actual_rows": 1000, + "actual_return": 4442.80164206028 + }, + { + "split": "val", + "episode_idx": 77, + "seed": 82, + "actual_rows": 1000, + "actual_return": 4616.909422039986 + }, + { + "split": "val", + "episode_idx": 81, + "seed": 87, + "actual_rows": 1000, + "actual_return": 4557.967975914478 + } + ], + "total_episodes": 100, + "total_rows": 96730, + "pass_gate_count": 100, + "mean_return": 4397.493643630147, + "minimum_return": 3069.413112640381, + "probe_train_rows": 87052, + "probe_val_rows": 9678, + "actual_train_rows": 87052, + "actual_val_rows": 9678, + "per_episode_seed_split_length_match": true, + "per_episode_probe_replay_match": false, + "exact_return_matches": 99, + "return_discrepancies": [ + { + "episode_idx": 99, + "seed": 104, + "probe_return": 4545.101155161858, + "actual_return": 4545.101168990135, + "delta": 1.3828277587890625e-05 + } + ] + }, + "code_provenance": { + "root_revision": "86df5ecef6735d5d824944e7acdee5bf204e1545", + "task_code_archive_sha256": "b0a0ca72b5c6eb65c16e5d5d70f59cad540260a27cb3a71109910326b940e135", + "iid_patch_sha256": "ceae6068634e4df2a553d8aee2cb70f7ef0640265330c65f75b9462b449010d0", + "sf_teacher_runtime_sha256": "2e4ba33a1c406bf3fa9b122f1cc66c13f1b0d7a96c89ec142e0428f184b4646d", + "sample_factory_revision": "4b7277842b17804fb928097a9689d889bd2f5cdc", + "starvla_revision": "f364fdf080434aea14bb2a19931e017f0af32d87", + "note": "Task code extracted from pinned root plus tracked IID runtime fix; archive runtime git metadata is unavailable." + }, + "gate_revision": { + "original_threshold": 3000, + "effective_threshold": 3000, + "changed": false, + "reason": "12/12 bounded original exporter attempts pass source threshold", + "probe": { + "status": "diagnostic_budget_reached", + "attempts": 12, + "completed_reset_seeds": [ + 9, + 0, + 2, + 5, + 1, + 3, + 4, + 6, + 7, + 8, + 10, + 11 + ], + "accepted_original_gate": 12, + "min_return": 3208.9899393320084, + "max_return": 4738.329069197178, + "mean_return": 4283.766990204652, + "gate": "episode_raw_return > 3000", + "original_dataset_target": 100, + "checkpoint": "/mnt/results/latency-sensitive-bench/profile-latency-new/walker2d/20260912T092949Z-walker2d/small_models/walker2d_profile_20260912T092949Z/checkpoint_p0/best_000013520_6922240_reward_4348.151.pth", + "deterministic": true, + "env_fps": 10, + "obs_fps": 10, + "latency_semantics": "action-delay IID exporter, no worker occupancy replay", + "note": "Intentional 12-attempt diagnostic cap; no dataset materialized; not exporter failure" + } + }, + "normalization_source": "new Walker2d converted dataset and actual VLA training dataset_statistics.json", + "data_acceptance_decision": { + "decision": "accept_existing_materialized_dataset", + "basis": "All100strict>3000, exactseed/split/length,99exactreturns; one explicitly recorded numerical discrepancy. No gate or data change, no generic tolerance.", + "discrepancies": [ + { + "episode_idx": 99, + "seed": 104, + "probe_return": 4545.101155161858, + "actual_return": 4545.101168990135, + "delta": 1.3828277587890625e-05 + } + ], + "diagnostic": "batch_precision_diagnostic.json", + "causal_limit": "Batch-sensitive float32 forward verified; sole causal origin of return delta not proven.", + "initial_audit": "materialized_data_validation.initial_failed.json", + "original_exact_audit_script": "validate_materialized_data.py preserved unchanged" + }, + "batch_precision_diagnostic": { + "status": "bounded_forward_diagnostic", + "observation_source": "materialized ep99 seed104 saved float32 state", + "checkpoint": "/mnt/results/latency-sensitive-bench/profile-latency-new/walker2d/20260912T092949Z-walker2d/small_models/walker2d_profile_20260912T092949Z/checkpoint_p0/best_000013520_6922240_reward_4348.151.pth", + "dtype": "float32", + "same_observation_repeated_batch_sizes": [ + 1, + 8, + 12 + ], + "results": [ + { + "step": 0, + "outputs": { + "1": [ + 4.379815101623535, + 1.601994514465332, + 1.099876880645752, + 1.424455165863037, + 1.4305202960968018, + 3.825317621231079 + ], + "8": [ + 4.379815101623535, + 1.601994514465332, + 1.0998765230178833, + 1.4244552850723267, + 1.4305202960968018, + 3.8253173828125 + ], + "12": [ + 4.379815101623535, + 1.601994514465332, + 1.0998765230178833, + 1.4244552850723267, + 1.4305202960968018, + 3.8253173828125 + ] + }, + "max_abs_1_vs_8": 3.5762786865234375e-07, + "max_abs_1_vs_12": 3.5762786865234375e-07 + }, + { + "step": 100, + "outputs": { + "1": [ + 2.1918554306030273, + 1.4689583778381348, + 1.5206661224365234, + 2.4722344875335693, + 1.8111798763275146, + 1.4706746339797974 + ], + "8": [ + 2.1918554306030273, + 1.4689583778381348, + 1.520666241645813, + 2.4722342491149902, + 1.8111798763275146, + 1.4706745147705078 + ], + "12": [ + 2.1918554306030273, + 1.4689583778381348, + 1.520666241645813, + 2.4722342491149902, + 1.8111798763275146, + 1.4706745147705078 + ] + }, + "max_abs_1_vs_8": 2.384185791015625e-07, + "max_abs_1_vs_12": 2.384185791015625e-07 + }, + { + "step": 500, + "outputs": { + "1": [ + 1.8575940132141113, + 1.4386999607086182, + 0.056306030601263046, + 1.852372407913208, + 0.3193018138408661, + -2.692965269088745 + ], + "8": [ + 1.8575941324234009, + 1.4387001991271973, + 0.05630632862448692, + 1.852372407913208, + 0.3193017840385437, + -2.692965030670166 + ], + "12": [ + 1.8575941324234009, + 1.4387001991271973, + 0.05630632862448692, + 1.852372407913208, + 0.3193017840385437, + -2.692965030670166 + ] + }, + "max_abs_1_vs_8": 2.980232238769531e-07, + "max_abs_1_vs_12": 2.980232238769531e-07 + }, + { + "step": 900, + "outputs": { + "1": [ + 1.5531995296478271, + 2.0711185932159424, + 0.9206967353820801, + 1.7364460229873657, + 0.7038085460662842, + -1.3887172937393188 + ], + "8": [ + 1.5531994104385376, + 2.0711183547973633, + 0.9206965565681458, + 1.7364461421966553, + 0.7038084268569946, + -1.3887172937393188 + ], + "12": [ + 1.5531994104385376, + 2.0711183547973633, + 0.9206965565681458, + 1.7364461421966553, + 0.7038084268569946, + -1.3887172937393188 + ] + }, + "max_abs_1_vs_8": 2.384185791015625e-07, + "max_abs_1_vs_12": 2.384185791015625e-07 + }, + { + "step": 999, + "outputs": { + "1": [ + 2.3437061309814453, + 1.84669828414917, + 0.36980900168418884, + 2.2871274948120117, + 2.5438475608825684, + 2.275155782699585 + ], + "8": [ + 2.3437061309814453, + 1.846698522567749, + 0.3698088228702545, + 2.2871272563934326, + 2.5438473224639893, + 2.275155782699585 + ], + "12": [ + 2.3437061309814453, + 1.846698522567749, + 0.3698088228702545, + 2.2871272563934326, + 2.5438473224639893, + 2.275155782699585 + ] + }, + "max_abs_1_vs_8": 2.384185791015625e-07, + "max_abs_1_vs_12": 2.384185791015625e-07 + } + ], + "interpretation_limit": "Demonstrates batch-size-sensitive forward arithmetic; does not reconstruct original probe actions or prove exact sole origin of return difference." + } +} diff --git a/latency-aware/walker2d/vla/starvla-qwenoft-h1/walker2d_rgb_state_profile_20260912T092949Z_openvla_native_continuous_projector_sft_5k/task_contract.json b/latency-aware/walker2d/vla/starvla-qwenoft-h1/walker2d_rgb_state_profile_20260912T092949Z_openvla_native_continuous_projector_sft_5k/task_contract.json new file mode 100644 index 0000000000000000000000000000000000000000..ea8394829420a6bf88a4a33390aa64124c51b104 --- /dev/null +++ b/latency-aware/walker2d/vla/starvla-qwenoft-h1/walker2d_rgb_state_profile_20260912T092949Z_openvla_native_continuous_projector_sft_5k/task_contract.json @@ -0,0 +1,80 @@ +{ + "action_space": { + "dtype": "float32", + "high": [ + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0 + ], + "labels": [ + "right_thigh_torque", + "right_leg_torque", + "right_foot_torque", + "left_thigh_torque", + "left_leg_torque", + "left_foot_torque" + ], + "low": [ + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0 + ], + "type": "box" + }, + "base_prompt": "Move the Walker2d robot forward while keeping its torso upright. Predict six continuous torques in [-1, 1] ordered as right thigh, right leg, right foot, left thigh, left leg, and left foot.", + "env_fps": 10.0, + "env_id": "LatencyBench/Walker2dContinuous-v0", + "frame_stack": 1, + "make_kwargs": { + "base_env_id": "Walker2d-v4", + "base_make_kwargs": { + "ctrl_cost_weight": 0.001, + "exclude_current_positions_from_observation": true, + "forward_reward_weight": 1.0, + "healthy_reward": 1.0, + "reset_noise_scale": 0.005, + "terminate_when_unhealthy": true + }, + "render_mode": "rgb_array" + }, + "noop_action": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ], + "obs_fps": 10.0, + "registration_imports": [ + "latency_bench.envs.gymnasium_walker2d" + ], + "state_space": { + "labels": [ + "torso_height", + "torso_angle", + "right_thigh_angle", + "right_leg_angle", + "right_foot_angle", + "left_thigh_angle", + "left_leg_angle", + "left_foot_angle", + "torso_x_velocity", + "torso_z_velocity", + "torso_angular_velocity", + "right_thigh_angular_velocity", + "right_leg_angular_velocity", + "right_foot_angular_velocity", + "left_thigh_angular_velocity", + "left_leg_angular_velocity", + "left_foot_angular_velocity" + ] + }, + "task_name": "walker2d_rgb_state" +} diff --git a/latency-aware/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_profile_h1_5fps_g128_20260922_m32/README.md b/latency-aware/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_profile_h1_5fps_g128_20260922_m32/README.md new file mode 100644 index 0000000000000000000000000000000000000000..b670c7929b9e7c5a18c43841b4dfeec836e8fb0c --- /dev/null +++ b/latency-aware/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_profile_h1_5fps_g128_20260922_m32/README.md @@ -0,0 +1,31 @@ +# walker2d / qwenpi_v3 + +Training condition: `latency-aware`. Run: `walker2d_pi05_profile_h1_5fps_g128_20260922_m32`. + +Copied from [Standard-Pipeline](https://huggingface.co/latency-sensitive-bench/Standard-Pipeline/tree/2208875f92b21e3f8ec2363a30a6705dfe56cc04/walker2d/profile_latency/Pi05) at `2208875f92b21e3f8ec2363a30a6705dfe56cc04`. + +- Checkpoint: `checkpoints/model.pt` +- Selection: only published VLA checkpoint +- Checkpoint SHA256: `248d8959f1c4b4045b64fd0c4f7a8e086606eea5827f263b04ef703e26b4cd84` +- Configuration, FPS, actions, normalization and latency semantics retain their source values. +- This import does not run evaluations or change training/evaluation/acceptance status. + +Action horizon is 1. Use `config.yaml` and `dataset_statistics.json` with the matching +prepared backbone. `task_contract.json` is extracted from the original training configuration. +Original machine paths in historical configurations are provenance: bind checkpoint, +backbone, manifest and profile paths to downloaded local files when running inference. + +See [the original model notes](source/README.md) for experiment results and limitations. + +[Original provenance](source/provenance.json) retains dataset references and recorded experiment evidence. + +## Download + +```python +from huggingface_hub import snapshot_download +prefix = "latency-aware/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_profile_h1_5fps_g128_20260922_m32" +snapshot_download("latency-sensitive-bench/benchmark-models", revision="", + allow_patterns=[prefix + "/**"]) +``` + +Use the verified import commit listed in the repository's import report. diff --git a/latency-aware/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_profile_h1_5fps_g128_20260922_m32/checkpoints/model.pt b/latency-aware/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_profile_h1_5fps_g128_20260922_m32/checkpoints/model.pt new file mode 100644 index 0000000000000000000000000000000000000000..9e1a72b37abfb7e72949dc5942aaaaa4a37d275a --- /dev/null +++ b/latency-aware/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_profile_h1_5fps_g128_20260922_m32/checkpoints/model.pt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:248d8959f1c4b4045b64fd0c4f7a8e086606eea5827f263b04ef703e26b4cd84 +size 10922653853 diff --git a/latency-aware/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_profile_h1_5fps_g128_20260922_m32/config.full.yaml b/latency-aware/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_profile_h1_5fps_g128_20260922_m32/config.full.yaml new file mode 100644 index 0000000000000000000000000000000000000000..041d1322b89fdb1e62693245103e0903999a5e30 --- /dev/null +++ b/latency-aware/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_profile_h1_5fps_g128_20260922_m32/config.full.yaml @@ -0,0 +1,266 @@ +framework: + name: QwenPI_v3 + qwenvl: + base_vlm: ${oc.env:PI05_BACKBONE_DIR} + attn_implementation: flash_attention_2 + vl_hidden_dim: 2560 + num_vl_layers: 36 + enable_gradient_checkpointing: false + action_model: + action_model_type: LayerwiseFM + action_dim: 6 + state_dim: 17 + state_encoding: continuous_projector + task_objective: null + action_horizon: 1 + repeated_diffusion_steps: 2 + num_inference_timesteps: 4 + add_pos_embed: true + max_seq_len: 1024 + num_target_vision_tokens: 32 + noise_beta_alpha: 1.5 + noise_beta_beta: 1.0 + noise_s: 0.999 + num_timestep_buckets: 1000 + diffusion_model_cfg: + action_dit_hidden_dim: 1024 + dropout: 0.2 + final_dropout: true + interleave_self_attention: true + norm_type: ada_norm + positional_embeddings: null + attention_head_dim: 64 + num_layers: 36 + input_embedding_dim: 1024 + cross_attention_dim: 1024 + output_dim: 1024 + num_attention_heads: 16 + action_env_dim: 6 + future_action_window_size: 0 + past_action_window_size: 0 +datasets: + vla_data: + dataset_py: lerobot_datasets + include_state: true + data_root_dir: / + data_mix: train + eval_data_mix: validation + custom_mixtures_path: ${PI05_RUN_DIR}/profile_latency/vla/mixture.json + action_type: continuous + sequential_step_sampling: true + eval_sequential_step_sampling: null + num_workers: 4 + eval_num_workers: 0 + prefetch_factor: 2 + persistent_workers: true + pin_memory: true + shuffle: true + action_balance: + enabled: false + strategy: balanced_epoch + action_key: action_id + target_flap_fraction: 0.3 + noop_id: 0 + flap_id: 1 + latency_curriculum: + enabled: false + strategy: exclusive + latencies: null + phase_steps: null + phase_distributions: null + new_latency_passes: 1.0 + replay_passes: 0.25 + target_total_passes: 2.0 + final_equalization: true + step_budget_mode: auto + eval_at_phase_end: false + save_at_phase_end: false + computed_plan: null + per_device_batch_size: 32 + load_all_data_for_training: true + num_obs_frames: 1 + image_mode: single + prompt_mode: raw + stitch_grid: + - 2 + - 2 + obs_image_size: + - 224 + - 224 + video_backend: pyav + lerobot_version: v2.0 + gymnasium_task_contract: + action_space: + dtype: float32 + high: + - 1.0 + - 1.0 + - 1.0 + - 1.0 + - 1.0 + - 1.0 + labels: + - right_thigh_torque + - right_leg_torque + - right_foot_torque + - left_thigh_torque + - left_leg_torque + - left_foot_torque + low: + - -1.0 + - -1.0 + - -1.0 + - -1.0 + - -1.0 + - -1.0 + type: box + base_prompt: Move the Walker2d robot forward while keeping its torso upright. + Predict six continuous torques in [-1, 1] ordered as right thigh, right leg, + right foot, left thigh, left leg, and left foot. + env_fps: 5.0 + env_id: LatencyBench/Walker2dContinuous-v0 + frame_stack: 1 + make_kwargs: + base_env_id: Walker2d-v4 + base_make_kwargs: + ctrl_cost_weight: 0.001 + exclude_current_positions_from_observation: true + forward_reward_weight: 1.0 + healthy_reward: 1.0 + reset_noise_scale: 0.005 + terminate_when_unhealthy: true + render_mode: rgb_array + noop_action: + - 0.0 + - 0.0 + - 0.0 + - 0.0 + - 0.0 + - 0.0 + obs_fps: 5.0 + registration_imports: + - latency_bench.envs.gymnasium_walker2d + state_space: + labels: + - torso_height + - torso_angle + - right_thigh_angle + - right_leg_angle + - right_foot_angle + - left_thigh_angle + - left_leg_angle + - left_foot_angle + - torso_x_velocity + - torso_z_velocity + - torso_angular_velocity + - right_thigh_angular_velocity + - right_leg_angular_velocity + - right_foot_angular_velocity + - left_thigh_angular_velocity + - left_leg_angular_velocity + - left_foot_angular_velocity + task_name: walker2d + active_action_dim: 6 +dataset: + source_hf: '' + config_name: null + source_subdir: null + converted_name: flappy_train + single_source_hf: '' + mixed_source_hf: '' + single_converted_name: flappy_train + mixed_converted_name: flappy_mixed_latency_train + single_latency_filter: null + mixed_latency_filter: null + force_download: false + setup_force: false + skip_verification: false + target_latency_unit: observation_steps + verify_rows: 200 + max_episodes: null + episodes_per_latency: null + latency_filter: null + debug_subset: + enabled: false + max_episodes: 5 + suffix: debug +base_model: + repo_id: Qwen/Qwen3-VL-4B-Instruct +initialization: + checkpoint_local_dir: null + checkpoint_hf_repo_id: null + checkpoint_filename: null +trainer: + max_train_steps: 5000 + num_warmup_steps: 100 + save_interval: 500 + eval_interval: 500 + eval_num_batches: 200 + per_latency_eval_num_batches: null + eval_action_classification: false + eval_action_classification_interval: null + cc_f1_tolerance: 1 + learning_rate: + base: 2.0e-05 + qwen_vl_interface: 1.0e-05 + action_model: 0.0001 + lr_scheduler_type: cosine_with_min_lr + scheduler_specific_kwargs: + min_lr: 1.0e-06 + freeze_modules: '' + freeze_vit: false + freeze_tied_embedding: false + freeze_llm_layers: [] + loss_scale: + vla: 1.0 + vlm: 0.1 + max_grad_norm: 1.0 + weight_decay: 0.0 + logging_frequency: 1 + profile_timing: + enabled: false + log_interval: 10 + gradient_clipping: 1.0 + gradient_accumulation_steps: 2 + distributed_backend: deepspeed + is_resume: false + pretrained_checkpoint: null + resume_step: 0 + reload_modules: null + optimizer: + name: AdamW + betas: + - 0.9 + - 0.95 + eps: 1.0e-08 + weight_decay: 1.0e-08 + fused: true + save_format: pt +run_id: walker2d_pi05_profile_h1_5fps_g128_20260922_m32 +run_root_dir: ${PI05_RUN_DIR}/profile_latency/vla/training +seed: 42 +is_debug: false +version_id: '0.21' +wandb_project: latency-sensitive-bench +wandb_entity: dongqianyu99-zhejiang-university +wandb_name: walker2d_pi05_profile_h1_5fps_g128_20260922_m32 +wandb_group: pi05-seven-env +wandb_tags: +- walker2d +- profile_latency +- Pi05 +- h1 +training_latency_condition: profile_latency +backbone: + repo_id: Qwen/Qwen3-VL-4B-Instruct + revision: ebb281ec70b05090aa6165b016eac8ec08e71b17 +checkpoint: + save_best_model: false + save_final_model: true + save_pt_file: true + save_training_state: true + local: + keep_last_n: 1 +global_batch: 128 +config_yaml: ${PI05_RUN_DIR}/profile_latency/train-m32.yaml +output_dir: ${PI05_RUN_DIR}/profile_latency/vla/training/walker2d_pi05_profile_h1_5fps_g128_20260922_m32 diff --git a/latency-aware/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_profile_h1_5fps_g128_20260922_m32/config.yaml b/latency-aware/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_profile_h1_5fps_g128_20260922_m32/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..e693865f0de9791008d4311e13a9c1badc44cef3 --- /dev/null +++ b/latency-aware/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_profile_h1_5fps_g128_20260922_m32/config.yaml @@ -0,0 +1,142 @@ +framework: + name: QwenPI_v3 + qwenvl: + base_vlm: ${oc.env:PI05_BACKBONE_DIR} + attn_implementation: flash_attention_2 + vl_hidden_dim: 2560 + num_vl_layers: 36 + enable_gradient_checkpointing: false + action_model: + action_model_type: LayerwiseFM + action_dim: 6 + state_dim: 17 + state_encoding: continuous_projector + task_objective: null + action_horizon: 1 + repeated_diffusion_steps: 2 + num_inference_timesteps: 4 + add_pos_embed: true + max_seq_len: 1024 + num_target_vision_tokens: 32 + noise_beta_alpha: 1.5 + noise_beta_beta: 1.0 + noise_s: 0.999 + num_timestep_buckets: 1000 + diffusion_model_cfg: + action_dit_hidden_dim: 1024 + dropout: 0.2 + final_dropout: true + interleave_self_attention: true + norm_type: ada_norm + positional_embeddings: null + attention_head_dim: 64 + num_layers: 36 + input_embedding_dim: 1024 + cross_attention_dim: 1024 + output_dim: 1024 + num_attention_heads: 16 + action_env_dim: 6 + future_action_window_size: 0 + past_action_window_size: 0 +version_id: '0.21' +datasets: + vla_data: + include_state: true + obs_image_size: + - 224 + - 224 + image_mode: single + stitch_grid: + - 2 + - 2 + num_obs_frames: 1 + gymnasium_task_contract: &id001 + action_space: + dtype: float32 + high: + - 1.0 + - 1.0 + - 1.0 + - 1.0 + - 1.0 + - 1.0 + labels: + - right_thigh_torque + - right_leg_torque + - right_foot_torque + - left_thigh_torque + - left_leg_torque + - left_foot_torque + low: + - -1.0 + - -1.0 + - -1.0 + - -1.0 + - -1.0 + - -1.0 + type: box + base_prompt: Move the Walker2d robot forward while keeping its torso upright. + Predict six continuous torques in [-1, 1] ordered as right thigh, right leg, + right foot, left thigh, left leg, and left foot. + env_fps: 5.0 + env_id: LatencyBench/Walker2dContinuous-v0 + frame_stack: 1 + make_kwargs: + base_env_id: Walker2d-v4 + base_make_kwargs: + ctrl_cost_weight: 0.001 + exclude_current_positions_from_observation: true + forward_reward_weight: 1.0 + healthy_reward: 1.0 + reset_noise_scale: 0.005 + terminate_when_unhealthy: true + render_mode: rgb_array + noop_action: + - 0.0 + - 0.0 + - 0.0 + - 0.0 + - 0.0 + - 0.0 + obs_fps: 5.0 + registration_imports: + - latency_bench.envs.gymnasium_walker2d + state_space: + labels: + - torso_height + - torso_angle + - right_thigh_angle + - right_leg_angle + - right_foot_angle + - left_thigh_angle + - left_leg_angle + - left_foot_angle + - torso_x_velocity + - torso_z_velocity + - torso_angular_velocity + - right_thigh_angular_velocity + - right_leg_angular_velocity + - right_foot_angular_velocity + - left_thigh_angular_velocity + - left_leg_angular_velocity + - left_foot_angular_velocity + task_name: walker2d +trainer: + pretrained_checkpoint: null + profile_timing: + enabled: false + log_interval: 10 +backbone: + repo_id: Qwen/Qwen3-VL-4B-Instruct + revision: ebb281ec70b05090aa6165b016eac8ec08e71b17 +rl_games: + model_alias: pi05 + task: gymnasium + action_carrier: native + gymnasium: + task_contract: *id001 + env_eval: + image_size: 224 + frameskip: 1 + image_transform: raw_rgb + prompt_mode: raw diff --git a/latency-aware/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_profile_h1_5fps_g128_20260922_m32/dataset_statistics.json b/latency-aware/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_profile_h1_5fps_g128_20260922_m32/dataset_statistics.json new file mode 100644 index 0000000000000000000000000000000000000000..1dc43f7c4be63aef75ecda73562bf5428e64c769 --- /dev/null +++ b/latency-aware/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_profile_h1_5fps_g128_20260922_m32/dataset_statistics.json @@ -0,0 +1,180 @@ +{ + "new_embodiment": { + "action": { + "mean": [ + 0.4979535639286041, + 0.46174800395965576, + -0.29137447476387024, + 0.9985677003860474, + 0.7979836463928223, + 0.38867390155792236 + ], + "std": [ + 0.581741213798523, + 0.4915325939655304, + 0.7721808552742004, + 0.035041093826293945, + 0.5022090673446655, + 0.6507065296173096 + ], + "max": [ + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0 + ], + "min": [ + -1.0, + -1.0, + -1.0, + 0.0, + -1.0, + -1.0 + ], + "q01": [ + -1.0, + -1.0, + -1.0, + 1.0, + -1.0, + -1.0 + ], + "q99": [ + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0 + ], + "mask": [ + true, + true, + true, + true, + true, + true + ] + }, + "state": { + "mean": [ + 0.23815855383872986, + -0.1610857993364334, + 0.5441378951072693, + 0.672118604183197, + -0.11596863716840744, + 0.6917569041252136, + 0.637482762336731, + 0.4345299303531647, + -0.18043601512908936, + 0.1474768966436386, + -0.007371156010776758, + -0.01192600466310978, + -0.0426424965262413, + 0.011722072958946228, + -0.002667834050953388, + -0.02254018746316433, + -0.048787184059619904 + ], + "std": [ + 0.37967589497566223, + 0.21708180010318756, + 0.318278968334198, + 0.24248674511909485, + 0.5238001942634583, + 0.1172376424074173, + 0.2069370299577713, + 0.458295077085495, + 0.2840738296508789, + 0.3721705675125122, + 0.3089592158794403, + 0.28156015276908875, + 0.4521273672580719, + 0.5419289469718933, + 0.1749182939529419, + 0.4066046178340912, + 0.5292556285858154 + ], + "max": [ + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0 + ], + "min": [ + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0 + ], + "q01": [ + -0.40312816202640533, + -0.7265578722953796, + -0.3082904410362243, + -0.060336984992027276, + -0.7253159177303314, + 0.18186532378196718, + -0.03887396693229675, + -0.7812737548351287, + -0.8779666405916214, + -0.5537199997901916, + -0.578927686214447, + -0.5143137556314469, + -1.0, + -1.0, + -0.3972254878282547, + -1.0, + -1.0 + ], + "q99": [ + 0.8957907927036282, + 0.27771593093872016, + 0.946374636888504, + 0.9333751380443571, + 0.8444677841663356, + 0.859323478937149, + 0.8583487355709075, + 0.9217252075672145, + 0.6287772941589355, + 0.9185322749614713, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0 + ] + }, + "num_transitions": 88708, + "num_trajectories": 90 + } +} \ No newline at end of file diff --git a/latency-aware/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_profile_h1_5fps_g128_20260922_m32/latency_prompt_map.json b/latency-aware/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_profile_h1_5fps_g128_20260922_m32/latency_prompt_map.json new file mode 100644 index 0000000000000000000000000000000000000000..280753c5aad60b23d1820879ab749bc68a9f8788 --- /dev/null +++ b/latency-aware/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_profile_h1_5fps_g128_20260922_m32/latency_prompt_map.json @@ -0,0 +1,17 @@ +{ + "1": { + "prompt": "Move the Walker2d robot forward while keeping its torso upright. Predict six continuous torques in [-1, 1] ordered as right thigh, right leg, right foot, left thigh, left leg, and left foot. Current action latency is 1 raw frames (200.00 ms). The environment runs at 5 FPS and observations are emitted at 5 FPS. Choose the best next action.", + "latency_raw_frames": 1, + "latency_ms": 200.0 + }, + "2": { + "prompt": "Move the Walker2d robot forward while keeping its torso upright. Predict six continuous torques in [-1, 1] ordered as right thigh, right leg, right foot, left thigh, left leg, and left foot. Current action latency is 2 raw frames (400.00 ms). The environment runs at 5 FPS and observations are emitted at 5 FPS. Choose the best next action.", + "latency_raw_frames": 2, + "latency_ms": 400.0 + }, + "3": { + "prompt": "Move the Walker2d robot forward while keeping its torso upright. Predict six continuous torques in [-1, 1] ordered as right thigh, right leg, right foot, left thigh, left leg, and left foot. Current action latency is 3 raw frames (600.00 ms). The environment runs at 5 FPS and observations are emitted at 5 FPS. Choose the best next action.", + "latency_raw_frames": 3, + "latency_ms": 600.0 + } +} \ No newline at end of file diff --git a/latency-aware/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_profile_h1_5fps_g128_20260922_m32/manifest.json b/latency-aware/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_profile_h1_5fps_g128_20260922_m32/manifest.json new file mode 100644 index 0000000000000000000000000000000000000000..0accecb9d0bc35de149aa1c22576692081ef6e9e --- /dev/null +++ b/latency-aware/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_profile_h1_5fps_g128_20260922_m32/manifest.json @@ -0,0 +1,186 @@ +{ + "dataset_name": "walker2d_h1_5fps_profile", + "env_name": "walker2d", + "episodes": 90, + "frames": 88708, + "task_prompts": [ + "Move the Walker2d robot forward while keeping its torso upright. Predict six continuous torques in [-1, 1] ordered as right thigh, right leg, right foot, left thigh, left leg, and left foot. Current action latency is 1 raw frames (200.00 ms). The environment runs at 5 FPS and observations are emitted at 5 FPS. Choose the best next action.", + "Move the Walker2d robot forward while keeping its torso upright. Predict six continuous torques in [-1, 1] ordered as right thigh, right leg, right foot, left thigh, left leg, and left foot. Current action latency is 2 raw frames (400.00 ms). The environment runs at 5 FPS and observations are emitted at 5 FPS. Choose the best next action.", + "Move the Walker2d robot forward while keeping its torso upright. Predict six continuous torques in [-1, 1] ordered as right thigh, right leg, right foot, left thigh, left leg, and left foot. Current action latency is 3 raw frames (600.00 ms). The environment runs at 5 FPS and observations are emitted at 5 FPS. Choose the best next action." + ], + "format": "lerobot_v2_image_parquet", + "source": "${PI05_RUN_DIR}/profile_latency/data/raw_profile", + "integration_name": "gymnasium", + "task_name": "walker2d", + "action_layout": "gymnasium_continuous_v1", + "action_labels": [ + "right_thigh_torque", + "right_leg_torque", + "right_foot_torque", + "left_thigh_torque", + "left_leg_torque", + "left_foot_torque" + ], + "carrier_action_labels": [ + "right_thigh_torque", + "right_leg_torque", + "right_foot_torque", + "left_thigh_torque", + "left_leg_torque", + "left_foot_torque" + ], + "action_dim": 6, + "active_action_dim": 6, + "action_carrier": "native", + "rows_unit": "decision_step", + "obs_fps": 5.0, + "obs_stride_raw_frames": 1, + "uses_state": true, + "state_dim": 17, + "state_labels": [ + "torso_height", + "torso_angle", + "right_thigh_angle", + "right_leg_angle", + "right_foot_angle", + "left_thigh_angle", + "left_leg_angle", + "left_foot_angle", + "torso_x_velocity", + "torso_z_velocity", + "torso_angular_velocity", + "right_thigh_angular_velocity", + "right_leg_angular_velocity", + "right_foot_angular_velocity", + "left_thigh_angular_velocity", + "left_leg_angular_velocity", + "left_foot_angular_velocity" + ], + "state_normalization": { + "type": "min_max", + "min": [ + 0.801384687423706, + -0.3846876919269562, + -1.3016910552978516, + -2.7476444244384766, + -1.5346914529800415, + -0.8290755152702332, + -2.1717329025268555, + -1.4086319208145142, + -0.415105402469635, + -4.531341552734375, + -10.0, + -10.0, + -10.0, + -10.0, + -10.0, + -10.0, + -10.0 + ], + "max": [ + 1.7503005266189575, + 0.9993473291397095, + 0.15303508937358856, + 0.31470051407814026, + 1.3454551696777344, + 0.16739727556705475, + 0.3809681832790375, + 1.2573986053466797, + 9.562085151672363, + 3.402139902114868, + 10.0, + 10.0, + 10.0, + 10.0, + 10.0, + 10.0, + 10.0 + ] + }, + "robot_type": "rl_games_gymnasium", + "latency_prompt_map_path": "${PI05_RUN_DIR}/profile_latency/data/lerobot/walker2d_h1_5fps_profile/latency_prompt_map.json", + "custom_mixtures_path": "${PI05_RUN_DIR}/profile_latency/data/lerobot/_generated_mixtures/walker2d_h1_5fps_profile.json", + "gymnasium_task": { + "action_space": { + "dtype": "float32", + "high": [ + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0 + ], + "labels": [ + "right_thigh_torque", + "right_leg_torque", + "right_foot_torque", + "left_thigh_torque", + "left_leg_torque", + "left_foot_torque" + ], + "low": [ + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0 + ], + "type": "box" + }, + "base_prompt": "Move the Walker2d robot forward while keeping its torso upright. Predict six continuous torques in [-1, 1] ordered as right thigh, right leg, right foot, left thigh, left leg, and left foot.", + "env_fps": 5.0, + "env_id": "LatencyBench/Walker2dContinuous-v0", + "frame_stack": 1, + "make_kwargs": { + "base_env_id": "Walker2d-v4", + "base_make_kwargs": { + "ctrl_cost_weight": 0.001, + "exclude_current_positions_from_observation": true, + "forward_reward_weight": 1.0, + "healthy_reward": 1.0, + "reset_noise_scale": 0.005, + "terminate_when_unhealthy": true + }, + "render_mode": "rgb_array" + }, + "noop_action": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ], + "obs_fps": 5.0, + "registration_imports": [ + "latency_bench.envs.gymnasium_walker2d" + ], + "state_space": { + "labels": [ + "torso_height", + "torso_angle", + "right_thigh_angle", + "right_leg_angle", + "right_foot_angle", + "left_thigh_angle", + "left_leg_angle", + "left_foot_angle", + "torso_x_velocity", + "torso_z_velocity", + "torso_angular_velocity", + "right_thigh_angular_velocity", + "right_leg_angular_velocity", + "right_foot_angular_velocity", + "left_thigh_angular_velocity", + "left_leg_angular_velocity", + "left_foot_angular_velocity" + ] + }, + "task_name": "walker2d" + }, + "validation_dataset_name": "walker2d_h1_5fps_profile__val", + "validation_episodes": 10, + "validation_frames": 10000 +} \ No newline at end of file diff --git a/latency-aware/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_profile_h1_5fps_g128_20260922_m32/provenance.json b/latency-aware/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_profile_h1_5fps_g128_20260922_m32/provenance.json new file mode 100644 index 0000000000000000000000000000000000000000..51fe619b830ef26b878ddae43fec25bc62f21917 --- /dev/null +++ b/latency-aware/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_profile_h1_5fps_g128_20260922_m32/provenance.json @@ -0,0 +1,34 @@ +{ + "task": "walker2d", + "model": "qwenpi_v3", + "training_condition": "latency-aware", + "training_run_id": "walker2d_pi05_profile_h1_5fps_g128_20260922_m32", + "source": { + "repo_id": "latency-sensitive-bench/Standard-Pipeline", + "revision": "2208875f92b21e3f8ec2363a30a6705dfe56cc04", + "path": "walker2d/profile_latency/Pi05", + "url": "https://huggingface.co/latency-sensitive-bench/Standard-Pipeline/tree/2208875f92b21e3f8ec2363a30a6705dfe56cc04/walker2d/profile_latency/Pi05" + }, + "hf_repo_id": "latency-sensitive-bench/benchmark-models", + "hf_path": "latency-aware/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_profile_h1_5fps_g128_20260922_m32", + "checkpoint": { + "source_file": "walker2d/profile_latency/Pi05/checkpoints/model.pt", + "source_sha256": "248d8959f1c4b4045b64fd0c4f7a8e086606eea5827f263b04ef703e26b4cd84", + "source_bytes": 10922653853, + "file": "checkpoints/model.pt", + "selection_rule": "only published VLA checkpoint", + "method": "copy", + "sha256": "248d8959f1c4b4045b64fd0c4f7a8e086606eea5827f263b04ef703e26b4cd84", + "bytes": 10922653853 + }, + "config_source": "walker2d/profile_latency/Pi05/config.yaml", + "task_contract_file": "task_contract.json", + "missing_source_files": [], + "migration": { + "operation": "copy_selected_inference_bundle", + "source_modified": false, + "evaluation_performed": false, + "experiment_status": "preserved_in_source_evidence" + }, + "action_horizon": 1 +} diff --git a/latency-aware/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_profile_h1_5fps_g128_20260922_m32/source/README.md b/latency-aware/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_profile_h1_5fps_g128_20260922_m32/source/README.md new file mode 100644 index 0000000000000000000000000000000000000000..79bf51d3a3b48b57ccd8f3b4efc69889cba9669a --- /dev/null +++ b/latency-aware/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_profile_h1_5fps_g128_20260922_m32/source/README.md @@ -0,0 +1,5 @@ +# walker2d Pi0.5 profile-latency VLA + +Fresh QwenPI_v3,5000 updates,seed42,global128 (micro32 × accumulation2 (explicitly approved after original micro64 OOM) ×2 GPUs),GC off,ZeRO2,H1,RGB224,5/5FPS,native state/action and train-only state min-max once. Dataset retained the >3000 gate: 100 accepted episodes in 118 total attempts including the original probe. + +Set PI05_BACKBONE_DIR to Qwen3-VL-4B-Instruct revision ebb281ec70b05090aa6165b016eac8ec08e71b17. Use config.yaml/checkpoints/model.pt and latency_prompt_map.json. Weights are byte-identical to final training export. CPU tensor-finiteness/hash validation passed; actual realtime F evaluation is a separate stage. No optimizer, training state, W&B cache or credentials are included. diff --git a/latency-aware/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_profile_h1_5fps_g128_20260922_m32/source/provenance.json b/latency-aware/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_profile_h1_5fps_g128_20260922_m32/source/provenance.json new file mode 100644 index 0000000000000000000000000000000000000000..8a55acebd85e170fc0a15e08348e5828d0c5384e --- /dev/null +++ b/latency-aware/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_profile_h1_5fps_g128_20260922_m32/source/provenance.json @@ -0,0 +1,96 @@ +{ + "task": "walker2d", + "backbone": { + "repo_id": "Qwen/Qwen3-VL-4B-Instruct", + "revision": "ebb281ec70b05090aa6165b016eac8ec08e71b17" + }, + "checkpoint": "steps_5000_pytorch_model.pt", + "training_steps": 5000, + "seed": 42, + "global_batch": 128, + "training_run_id": "walker2d_pi05_profile_h1_5fps_g128_20260922_m32", + "condition": "profile_latency", + "source": { + "task": "walker2d", + "teacher_verification_sha256": "20a4dde06a7ab677d8cb59116b3cb69d75204aa0fc463e8c48511a69288920e2", + "profile_sha256": "b5409f8578bec62e2c5232c3916b3ddd2e0e9494031259265540ba6aef882e5e", + "profile_publication_revision": "e1a9bdd566b96d5da4703f769d40eab656af491e", + "selected_checkpoint": "${PI05_RUN_DIR}/profile_latency/teacher/checkpoints/pi05_walker2d_profile_appo_h1_5fps_g128_20260921/checkpoint_p0/best_000018928_9691136_reward_4148.139.pth", + "selected_checkpoint_sha256": "3d51959c7d65d8e9d1ec34cfc4a8dc7741abf9726439512bc7a0442fd265044e", + "selected_checkpoint_env_steps": 9691136, + "probe12_state_sha256": "266f2ec22ca31030bcc2e2161543314b579d178a1f45ff428b3278aa0bdead58", + "full_probe_state_sha256": "369efebe9c5abd237b02f5d368c4d1f21b0533fea1c3d9f59e501484d386f67f", + "data_attempts": 118, + "data_accepted": 100, + "data_rejected": 18, + "splits": { + "train": { + "rows": 88708, + "episodes": 90, + "return_min": 3049.0915127396584, + "return_max": 4926.397267758846, + "length_min": 707, + "length_max": 1000 + }, + "val": { + "rows": 10000, + "episodes": 10, + "return_min": 4344.816579580307, + "return_max": 4727.460265636444, + "length_min": 1000, + "length_max": 1000 + } + }, + "state_dim": 17, + "native_action_dim": 6, + "action_carrier": "native", + "state_normalization": { + "type": "min_max", + "min": [ + 0.801384687423706, + -0.3846876919269562, + -1.3016910552978516, + -2.7476444244384766, + -1.5346914529800415, + -0.8290755152702332, + -2.1717329025268555, + -1.4086319208145142, + -0.415105402469635, + -4.531341552734375, + -10.0, + -10.0, + -10.0, + -10.0, + -10.0, + -10.0, + -10.0 + ], + "max": [ + 1.7503005266189575, + 0.9993473291397095, + 0.15303508937358856, + 0.31470051407814026, + 1.3454551696777344, + 0.16739727556705475, + 0.3809681832790375, + 1.2573986053466797, + 9.562085151672363, + 3.402139902114868, + 10.0, + 10.0, + 10.0, + 10.0, + 10.0, + 10.0, + 10.0 + ] + }, + "raw_metadata_sha256": "f01614b1df2395b2a4ae5f7bcaf7c215c42a029543ee7d31758ce49170fdbea2", + "raw_train_parquet_sha256": "755e77d3f27a1e73ae2e01b720288b8ea261cbaaffb537b93794f3f8b2bae595", + "raw_val_parquet_sha256": "eaff1190ab7ecb98aa95051e06e9e5efc8dcdcb2054e3464b73da2cc286e824e", + "converted_manifest_sha256": "e850ecfe50be72061722c0a57d95efb797f705e5c0aa1049c56f9d0bec54fbdb", + "latency_prompt_map_sha256": "ea71f353aa39bd39ca32d0fdda58cac6a0d0ae158f948241d8d494d261e1c3bc" + }, + "training_config_sha256": "385d1a431384f36a79d4d6531ea26da7e4751a6924aac343ed685d268440f83a", + "dataset_manifest_sha256": "e850ecfe50be72061722c0a57d95efb797f705e5c0aa1049c56f9d0bec54fbdb" +} diff --git a/latency-aware/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_profile_h1_5fps_g128_20260922_m32/task_contract.json b/latency-aware/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_profile_h1_5fps_g128_20260922_m32/task_contract.json new file mode 100644 index 0000000000000000000000000000000000000000..83e3865ecb5e5aad835cb8de453f17e41e7eb216 --- /dev/null +++ b/latency-aware/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_profile_h1_5fps_g128_20260922_m32/task_contract.json @@ -0,0 +1,80 @@ +{ + "action_space": { + "dtype": "float32", + "high": [ + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0 + ], + "labels": [ + "right_thigh_torque", + "right_leg_torque", + "right_foot_torque", + "left_thigh_torque", + "left_leg_torque", + "left_foot_torque" + ], + "low": [ + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0 + ], + "type": "box" + }, + "base_prompt": "Move the Walker2d robot forward while keeping its torso upright. Predict six continuous torques in [-1, 1] ordered as right thigh, right leg, right foot, left thigh, left leg, and left foot.", + "env_fps": 5.0, + "env_id": "LatencyBench/Walker2dContinuous-v0", + "frame_stack": 1, + "make_kwargs": { + "base_env_id": "Walker2d-v4", + "base_make_kwargs": { + "ctrl_cost_weight": 0.001, + "exclude_current_positions_from_observation": true, + "forward_reward_weight": 1.0, + "healthy_reward": 1.0, + "reset_noise_scale": 0.005, + "terminate_when_unhealthy": true + }, + "render_mode": "rgb_array" + }, + "noop_action": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ], + "obs_fps": 5.0, + "registration_imports": [ + "latency_bench.envs.gymnasium_walker2d" + ], + "state_space": { + "labels": [ + "torso_height", + "torso_angle", + "right_thigh_angle", + "right_leg_angle", + "right_foot_angle", + "left_thigh_angle", + "left_leg_angle", + "left_foot_angle", + "torso_x_velocity", + "torso_z_velocity", + "torso_angular_velocity", + "right_thigh_angular_velocity", + "right_leg_angular_velocity", + "right_foot_angular_velocity", + "left_thigh_angular_velocity", + "left_leg_angular_velocity", + "left_foot_angular_velocity" + ] + }, + "task_name": "walker2d" +} diff --git a/latency-aware/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_profile_h1_5fps_g128_20260922_m32/validation.json b/latency-aware/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_profile_h1_5fps_g128_20260922_m32/validation.json new file mode 100644 index 0000000000000000000000000000000000000000..0fc56c96caf6d78d6d06a60dbbff4cbc3ae3b351 --- /dev/null +++ b/latency-aware/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_profile_h1_5fps_g128_20260922_m32/validation.json @@ -0,0 +1,10 @@ +{ + "state": "CPU_STATE_DICT_AND_BUNDLE_HASH_VERIFIED", + "checkpoint_sha256": "248d8959f1c4b4045b64fd0c4f7a8e086606eea5827f263b04ef703e26b4cd84", + "tensor_count": 1386, + "all_floating_tensors_finite": true, + "training_steps": 5000, + "global_batch": 128, + "gpu_forward": "NOT_RUN_AT_PUBLICATION", + "verified_at": "2026-09-22T01:32:01.661335+00:00" +} diff --git a/zero-latency/walker2d/small-policy/sample-factory-v1/README.md b/zero-latency/walker2d/small-policy/sample-factory-v1/README.md new file mode 100644 index 0000000000000000000000000000000000000000..0ae4a302099cf74191544aa8ae6042473e91e235 --- /dev/null +++ b/zero-latency/walker2d/small-policy/sample-factory-v1/README.md @@ -0,0 +1,26 @@ +# walker2d / sample-factory-appo + +Training condition: `zero-latency`. Run: `gymnasium_walker2d_continuous_zero_latency_sf_official_10m`. + +Copied from [Standard-Pipeline](https://huggingface.co/latency-sensitive-bench/Standard-Pipeline/tree/2208875f92b21e3f8ec2363a30a6705dfe56cc04/walker2d/zero_latency/small_model) at `2208875f92b21e3f8ec2363a30a6705dfe56cc04`. + +- Checkpoint: `checkpoint.pth` +- Selection: existing zero-latency training best +- Checkpoint SHA256: `db194fa59b1b9f0ec182b06e0fe4463ab77c8b9e8169fd616bcd5c0b4c41d482` +- Configuration, FPS, actions, normalization and latency semantics retain their source values. +- This import does not run evaluations or change training/evaluation/acceptance status. + +Sample Factory APPO inference bundle: `config.json` and the explicit `checkpoint.pth`. +The checkpoint contains `model`, `train_step`, and `env_steps`; every exported model +tensor was reloaded and compared bit-for-bit with the selected source checkpoint. + +## Download + +```python +from huggingface_hub import snapshot_download +prefix = "zero-latency/walker2d/small-policy/sample-factory-v1" +snapshot_download("latency-sensitive-bench/benchmark-models", revision="", + allow_patterns=[prefix + "/**"]) +``` + +Use the verified import commit listed in the repository's import report. diff --git a/zero-latency/walker2d/small-policy/sample-factory-v1/checkpoint.pth b/zero-latency/walker2d/small-policy/sample-factory-v1/checkpoint.pth new file mode 100644 index 0000000000000000000000000000000000000000..cc12678f132afdcf19ef60791e7f8230d548d6a2 --- /dev/null +++ b/zero-latency/walker2d/small-policy/sample-factory-v1/checkpoint.pth @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:db194fa59b1b9f0ec182b06e0fe4463ab77c8b9e8169fd616bcd5c0b4c41d482 +size 29877 diff --git a/zero-latency/walker2d/small-policy/sample-factory-v1/config.json b/zero-latency/walker2d/small-policy/sample-factory-v1/config.json new file mode 100644 index 0000000000000000000000000000000000000000..cd67a0e248450d8ab6199b661d57fc83c394dc92 --- /dev/null +++ b/zero-latency/walker2d/small-policy/sample-factory-v1/config.json @@ -0,0 +1,230 @@ +{ + "help": false, + "algo": "APPO", + "env": "latency_gymnasium", + "experiment": "gymnasium_walker2d_continuous_zero_latency_sf_official_10m", + "train_dir": "/mnt/checkpoints/latency-sensitive-bench/walker2d", + "restart_behavior": "overwrite", + "device": "gpu", + "seed": 0, + "num_policies": 1, + "async_rl": false, + "serial_mode": false, + "batched_sampling": false, + "num_batches_to_accumulate": 2, + "worker_num_splits": 2, + "policy_workers_per_policy": 1, + "max_policy_lag": 300, + "num_workers": 8, + "num_envs_per_worker": 8, + "batch_size": 1024, + "num_batches_per_epoch": 4, + "num_epochs": 2, + "rollout": 64, + "recurrence": 1, + "shuffle_minibatches": false, + "gamma": 0.99, + "reward_scale": 1.0, + "reward_clip": 1000.0, + "value_bootstrap": true, + "normalize_returns": true, + "exploration_loss_coeff": 0.0, + "value_loss_coeff": 1.3, + "kl_loss_coeff": 0.1, + "exploration_loss": "entropy", + "gae_lambda": 0.95, + "ppo_clip_ratio": 0.2, + "ppo_clip_value": 0.2, + "with_vtrace": false, + "vtrace_rho": 1.0, + "vtrace_c": 1.0, + "optimizer": "adam", + "adam_eps": 1e-06, + "adam_beta1": 0.9, + "adam_beta2": 0.999, + "max_grad_norm": 3.5, + "learning_rate": 0.00295, + "lr_schedule": "linear_decay", + "lr_schedule_kl_threshold": 0.008, + "lr_adaptive_min": 1e-06, + "lr_adaptive_max": 0.01, + "obs_subtract_mean": 0.0, + "obs_scale": 1.0, + "normalize_input": true, + "normalize_input_keys": null, + "decorrelate_experience_max_seconds": 0, + "decorrelate_envs_on_one_worker": true, + "actor_worker_gpus": [], + "set_workers_cpu_affinity": true, + "force_envs_single_thread": false, + "default_niceness": 0, + "log_to_file": true, + "experiment_summaries_interval": 10, + "flush_summaries_interval": 30, + "stats_avg": 100, + "summaries_use_frameskip": true, + "heartbeat_interval": 20, + "heartbeat_reporting_interval": 180, + "train_for_env_steps": 10000000, + "train_for_seconds": 10000000000, + "save_every_sec": 600, + "keep_checkpoints": 5, + "load_checkpoint_kind": "latest", + "save_milestones_sec": -1, + "save_best_every_sec": 5, + "save_best_metric": "reward", + "save_best_after": 100000, + "benchmark": false, + "encoder_mlp_layers": [ + 64, + 64 + ], + "encoder_conv_architecture": "convnet_simple", + "encoder_conv_mlp_layers": [ + 512 + ], + "use_rnn": false, + "rnn_size": 512, + "rnn_type": "gru", + "rnn_num_layers": 1, + "decoder_mlp_layers": [], + "nonlinearity": "tanh", + "policy_initialization": "torch_default", + "policy_init_gain": 1.0, + "actor_critic_share_weights": true, + "adaptive_stddev": false, + "continuous_tanh_scale": 0.0, + "initial_stddev": 1.0, + "use_env_info_cache": false, + "env_gpu_actions": false, + "env_gpu_observations": true, + "env_frameskip": 1, + "env_framestack": 1, + "pixel_format": "CHW", + "use_record_episode_statistics": false, + "with_wandb": false, + "wandb_user": null, + "wandb_project": "sample_factory", + "wandb_group": null, + "wandb_job_type": "SF", + "wandb_tags": [], + "with_pbt": false, + "pbt_mix_policies_in_one_env": true, + "pbt_period_env_steps": 5000000, + "pbt_start_mutation": 20000000, + "pbt_replace_fraction": 0.3, + "pbt_mutation_rate": 0.15, + "pbt_replace_reward_gap": 0.1, + "pbt_replace_reward_gap_absolute": 1e-06, + "pbt_optimize_gamma": false, + "pbt_target_objective": "true_objective", + "pbt_perturb_min": 1.1, + "pbt_perturb_max": 1.5, + "gym_task_name": "walker2d_continuous_torque", + "gym_env_id": "LatencyBench/Walker2dContinuous-v0", + "gym_make_kwargs_json": "{\"base_env_id\": \"Walker2d-v4\", \"render_mode\": \"rgb_array\", \"base_make_kwargs\": {\"forward_reward_weight\": 1.0, \"ctrl_cost_weight\": 0.001, \"healthy_reward\": 1.0, \"terminate_when_unhealthy\": true, \"reset_noise_scale\": 0.005, \"exclude_current_positions_from_observation\": true}}", + "gym_registration_imports_json": "[\"latency_bench.envs.gymnasium_walker2d\"]", + "gym_action_space_json": "{\"type\": \"box\", \"labels\": [\"right_thigh_torque\", \"right_leg_torque\", \"right_foot_torque\", \"left_thigh_torque\", \"left_leg_torque\", \"left_foot_torque\"], \"low\": [-1.0, -1.0, -1.0, -1.0, -1.0, -1.0], \"high\": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0], \"dtype\": \"float32\", \"openvla_carrier_dim\": 7, \"openvla_padding\": [0.0]}", + "gym_action_labels_json": null, + "gym_action_values_json": null, + "gym_noop_action_id": null, + "gym_noop_action_json": "[0.0, 0.0, 0.0, 0.0, 0.0, 0.0]", + "gym_base_prompt": "Move the Walker2d robot forward while keeping its torso upright. Predict six continuous torques in [-1, 1] ordered as right thigh, right leg, right foot, left thigh, left leg, and left foot.", + "env_fps": 125.0, + "obs_fps": 125.0, + "frame_stack": 1, + "export_env_raw_rgb_frames": false, + "mode": "train", + "latency_type": "zero", + "fixed_latency_ms": null, + "mean_latency_ms": null, + "std_latency_ms": null, + "min_latency_ms": null, + "max_latency_ms": null, + "latency_seed": null, + "add_latency_info": false, + "max_pending_actions": null, + "hold_policy": "hold", + "ordering_policy": "issue_order_fifo", + "eval_episodes": 5, + "eval_parallel_envs": 1, + "eval_latency_raw_frame_values": null, + "eval_max_steps": 1000, + "eval_deterministic": true, + "eval_raw_reward": false, + "episode_metrics_path": "/mnt/checkpoints/latency-sensitive-bench/walker2d/gymnasium_walker2d_continuous_zero_latency_sf_official_10m/episode_metrics.jsonl", + "command_line": "--mode train --algo APPO --env latency_gymnasium --experiment gymnasium_walker2d_continuous_zero_latency_sf_official_10m --train_dir /mnt/checkpoints/latency-sensitive-bench/walker2d --restart_behavior overwrite --device gpu --seed 0 --episode_metrics_path /mnt/checkpoints/latency-sensitive-bench/walker2d/gymnasium_walker2d_continuous_zero_latency_sf_official_10m/episode_metrics.jsonl --train_for_env_steps 10000000 --num_workers 8 --num_envs_per_worker 8 --num_policies 1 --batch_size 1024 --rollout 64 --recurrence 1 --num_epochs 2 --num_batches_per_epoch 4 --worker_num_splits 2 --max_policy_lag 300 --learning_rate 0.00295 --kl_loss_coeff 0.1 --lr_schedule linear_decay --nonlinearity tanh --policy_initialization torch_default --initial_stddev 1.0 --gamma 0.99 --gae_lambda 0.95 --ppo_clip_ratio 0.2 --ppo_clip_value 0.2 --exploration_loss_coeff 0.0 --value_loss_coeff 1.3 --max_grad_norm 3.5 --save_every_sec 600 --keep_checkpoints 5 --async_rl False --batched_sampling False --serial_mode False --use_rnn False --normalize_returns True --normalize_input True --adaptive_stddev False --shuffle_minibatches False --value_bootstrap True --encoder_mlp_layers 64 64 --latency-type zero --add-latency-info False --eval-episodes 5 --eval-parallel-envs 1 --eval-max-steps 1000 --eval-deterministic True --gym-task-name walker2d_continuous_torque --gym-env-id LatencyBench/Walker2dContinuous-v0 --gym-make-kwargs-json {\"base_env_id\": \"Walker2d-v4\", \"render_mode\": \"rgb_array\", \"base_make_kwargs\": {\"forward_reward_weight\": 1.0, \"ctrl_cost_weight\": 0.001, \"healthy_reward\": 1.0, \"terminate_when_unhealthy\": true, \"reset_noise_scale\": 0.005, \"exclude_current_positions_from_observation\": true}} --gym-registration-imports-json [\"latency_bench.envs.gymnasium_walker2d\"] --gym-action-space-json {\"type\": \"box\", \"labels\": [\"right_thigh_torque\", \"right_leg_torque\", \"right_foot_torque\", \"left_thigh_torque\", \"left_leg_torque\", \"left_foot_torque\"], \"low\": [-1.0, -1.0, -1.0, -1.0, -1.0, -1.0], \"high\": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0], \"dtype\": \"float32\", \"openvla_carrier_dim\": 7, \"openvla_padding\": [0.0]} --gym-noop-action-json [0.0, 0.0, 0.0, 0.0, 0.0, 0.0] --gym-base-prompt Move the Walker2d robot forward while keeping its torso upright. Predict six continuous torques in [-1, 1] ordered as right thigh, right leg, right foot, left thigh, left leg, and left foot. --env-fps 125.0 --obs-fps 125.0 --frame-stack 1 --hold-policy hold --ordering-policy issue_order_fifo --use_env_info_cache False", + "cli_args": { + "algo": "APPO", + "env": "latency_gymnasium", + "experiment": "gymnasium_walker2d_continuous_zero_latency_sf_official_10m", + "train_dir": "/mnt/checkpoints/latency-sensitive-bench/walker2d", + "restart_behavior": "overwrite", + "device": "gpu", + "seed": 0, + "num_policies": 1, + "async_rl": false, + "serial_mode": false, + "batched_sampling": false, + "worker_num_splits": 2, + "max_policy_lag": 300, + "num_workers": 8, + "num_envs_per_worker": 8, + "batch_size": 1024, + "num_batches_per_epoch": 4, + "num_epochs": 2, + "rollout": 64, + "recurrence": 1, + "shuffle_minibatches": false, + "gamma": 0.99, + "value_bootstrap": true, + "normalize_returns": true, + "exploration_loss_coeff": 0.0, + "value_loss_coeff": 1.3, + "kl_loss_coeff": 0.1, + "gae_lambda": 0.95, + "ppo_clip_ratio": 0.2, + "ppo_clip_value": 0.2, + "max_grad_norm": 3.5, + "learning_rate": 0.00295, + "lr_schedule": "linear_decay", + "normalize_input": true, + "train_for_env_steps": 10000000, + "save_every_sec": 600, + "keep_checkpoints": 5, + "encoder_mlp_layers": [ + 64, + 64 + ], + "use_rnn": false, + "nonlinearity": "tanh", + "policy_initialization": "torch_default", + "adaptive_stddev": false, + "initial_stddev": 1.0, + "use_env_info_cache": false, + "gym_task_name": "walker2d_continuous_torque", + "gym_env_id": "LatencyBench/Walker2dContinuous-v0", + "gym_make_kwargs_json": "{\"base_env_id\": \"Walker2d-v4\", \"render_mode\": \"rgb_array\", \"base_make_kwargs\": {\"forward_reward_weight\": 1.0, \"ctrl_cost_weight\": 0.001, \"healthy_reward\": 1.0, \"terminate_when_unhealthy\": true, \"reset_noise_scale\": 0.005, \"exclude_current_positions_from_observation\": true}}", + "gym_registration_imports_json": "[\"latency_bench.envs.gymnasium_walker2d\"]", + "gym_action_space_json": "{\"type\": \"box\", \"labels\": [\"right_thigh_torque\", \"right_leg_torque\", \"right_foot_torque\", \"left_thigh_torque\", \"left_leg_torque\", \"left_foot_torque\"], \"low\": [-1.0, -1.0, -1.0, -1.0, -1.0, -1.0], \"high\": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0], \"dtype\": \"float32\", \"openvla_carrier_dim\": 7, \"openvla_padding\": [0.0]}", + "gym_noop_action_json": "[0.0, 0.0, 0.0, 0.0, 0.0, 0.0]", + "gym_base_prompt": "Move the Walker2d robot forward while keeping its torso upright. Predict six continuous torques in [-1, 1] ordered as right thigh, right leg, right foot, left thigh, left leg, and left foot.", + "env_fps": 125.0, + "obs_fps": 125.0, + "frame_stack": 1, + "mode": "train", + "latency_type": "zero", + "add_latency_info": false, + "hold_policy": "hold", + "ordering_policy": "issue_order_fifo", + "eval_episodes": 5, + "eval_parallel_envs": 1, + "eval_max_steps": 1000, + "eval_deterministic": true, + "episode_metrics_path": "/mnt/checkpoints/latency-sensitive-bench/walker2d/gymnasium_walker2d_continuous_zero_latency_sf_official_10m/episode_metrics.jsonl" + }, + "git_hash": "6f691b7d5e27102654adf6b72f069f959b496cb6", + "git_repo_name": "https://github.com/ZihanWang314/latency-sensitive-bench.git", + "eval_env_frameskip": 1, + "output_dir": "/mnt/data/latency-sensitive-bench/walker2d/continuous_sf_official_10m_train" +} \ No newline at end of file diff --git a/zero-latency/walker2d/small-policy/sample-factory-v1/provenance.json b/zero-latency/walker2d/small-policy/sample-factory-v1/provenance.json new file mode 100644 index 0000000000000000000000000000000000000000..fe2f6304e360670f055331fa6e92b04581a35db6 --- /dev/null +++ b/zero-latency/walker2d/small-policy/sample-factory-v1/provenance.json @@ -0,0 +1,44 @@ +{ + "task": "walker2d", + "model": "sample-factory-appo", + "training_condition": "zero-latency", + "training_run_id": "gymnasium_walker2d_continuous_zero_latency_sf_official_10m", + "source": { + "repo_id": "latency-sensitive-bench/Standard-Pipeline", + "revision": "2208875f92b21e3f8ec2363a30a6705dfe56cc04", + "path": "walker2d/zero_latency/small_model", + "url": "https://huggingface.co/latency-sensitive-bench/Standard-Pipeline/tree/2208875f92b21e3f8ec2363a30a6705dfe56cc04/walker2d/zero_latency/small_model" + }, + "hf_repo_id": "latency-sensitive-bench/benchmark-models", + "hf_path": "zero-latency/walker2d/small-policy/sample-factory-v1", + "checkpoint": { + "source_file": "walker2d/zero_latency/small_model/checkpoint_p0/best_000013056_6684672_reward_4212.219.pth", + "source_sha256": "31ce2cf679ab25fd8693c93626daf42342c84d35f0568f9371a1675bd0cddbfe", + "source_bytes": 83973, + "file": "checkpoint.pth", + "selection_rule": "existing zero-latency training best", + "method": "inference_export", + "sha256": "db194fa59b1b9f0ec182b06e0fe4463ab77c8b9e8169fd616bcd5c0b4c41d482", + "bytes": 29877, + "train_step": 13056, + "env_steps": 6684672, + "tensor_count": 15, + "tensor_equality_verified": true, + "removed_fields": [ + "best_performance", + "curr_lr", + "optimizer" + ] + }, + "config_source": "walker2d/zero_latency/small_model/config.json", + "task_contract_file": null, + "missing_source_files": [], + "migration": { + "operation": "copy_selected_inference_bundle", + "source_modified": false, + "evaluation_performed": false, + "experiment_status": "preserved_in_source_evidence" + }, + "algorithm": "APPO", + "latency_model": null +} diff --git a/zero-latency/walker2d/vla/starvla-qwengr00t-h1/walker2d_l0_gr00t_h1_5fps_2h200_20260917/README.md b/zero-latency/walker2d/vla/starvla-qwengr00t-h1/walker2d_l0_gr00t_h1_5fps_2h200_20260917/README.md new file mode 100644 index 0000000000000000000000000000000000000000..a35ce8bf10ed15665f9f5e45603498f278f92e94 --- /dev/null +++ b/zero-latency/walker2d/vla/starvla-qwengr00t-h1/walker2d_l0_gr00t_h1_5fps_2h200_20260917/README.md @@ -0,0 +1,31 @@ +# walker2d / qwengr00t + +Training condition: `zero-latency`. Run: `walker2d_l0_gr00t_h1_5fps_2h200_20260917`. + +Copied from [Standard-Pipeline](https://huggingface.co/latency-sensitive-bench/Standard-Pipeline/tree/2208875f92b21e3f8ec2363a30a6705dfe56cc04/walker2d/zero_latency/GR00T) at `2208875f92b21e3f8ec2363a30a6705dfe56cc04`. + +- Checkpoint: `checkpoints/model.pt` +- Selection: only published VLA checkpoint +- Checkpoint SHA256: `5df7bf29e8cc0bc984cb21ad4f8db673f16b17ceb438f4f4c91be65ee52c4de1` +- Configuration, FPS, actions, normalization and latency semantics retain their source values. +- This import does not run evaluations or change training/evaluation/acceptance status. + +Action horizon is 1. Use `config.yaml` and `dataset_statistics.json` with the matching +prepared backbone. `task_contract.json` is extracted from the original training configuration. +Original machine paths in historical configurations are provenance: bind checkpoint, +backbone, manifest and profile paths to downloaded local files when running inference. + +See [the original model notes](source/README.md) for experiment results and limitations. + +[Original provenance](source/provenance.json) retains dataset references and recorded experiment evidence. + +## Download + +```python +from huggingface_hub import snapshot_download +prefix = "zero-latency/walker2d/vla/starvla-qwengr00t-h1/walker2d_l0_gr00t_h1_5fps_2h200_20260917" +snapshot_download("latency-sensitive-bench/benchmark-models", revision="", + allow_patterns=[prefix + "/**"]) +``` + +Use the verified import commit listed in the repository's import report. diff --git a/zero-latency/walker2d/vla/starvla-qwengr00t-h1/walker2d_l0_gr00t_h1_5fps_2h200_20260917/checkpoints/model.pt b/zero-latency/walker2d/vla/starvla-qwengr00t-h1/walker2d_l0_gr00t_h1_5fps_2h200_20260917/checkpoints/model.pt new file mode 100644 index 0000000000000000000000000000000000000000..0ef20241e07e3f5c8681d63e0c544fa3b9adb63d --- /dev/null +++ b/zero-latency/walker2d/vla/starvla-qwengr00t-h1/walker2d_l0_gr00t_h1_5fps_2h200_20260917/checkpoints/model.pt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5df7bf29e8cc0bc984cb21ad4f8db673f16b17ceb438f4f4c91be65ee52c4de1 +size 9976860963 diff --git a/zero-latency/walker2d/vla/starvla-qwengr00t-h1/walker2d_l0_gr00t_h1_5fps_2h200_20260917/config.full.yaml b/zero-latency/walker2d/vla/starvla-qwengr00t-h1/walker2d_l0_gr00t_h1_5fps_2h200_20260917/config.full.yaml new file mode 100644 index 0000000000000000000000000000000000000000..3f211f346f0058c288528b5d535d5440083d25b1 --- /dev/null +++ b/zero-latency/walker2d/vla/starvla-qwengr00t-h1/walker2d_l0_gr00t_h1_5fps_2h200_20260917/config.full.yaml @@ -0,0 +1,261 @@ +framework: + name: QwenGR00T + qwenvl: + base_vlm: ${oc.env:GR00T_BACKBONE_DIR} + attn_implementation: flash_attention_2 + vl_hidden_dim: 2048 + enable_gradient_checkpointing: true + action_model: + action_model_type: DiT-B + action_hidden_dim: 1024 + hidden_size: 1024 + add_pos_embed: true + max_seq_len: 1024 + action_dim: 6 + state_dim: 17 + task_objective: null + action_horizon: 1 + repeated_diffusion_steps: 8 + noise_beta_alpha: 1.5 + noise_beta_beta: 1.0 + noise_s: 0.999 + num_timestep_buckets: 1000 + num_inference_timesteps: 4 + num_target_vision_tokens: 32 + diffusion_model_cfg: + cross_attention_dim: 2560 + dropout: 0.2 + final_dropout: true + interleave_self_attention: true + norm_type: ada_norm + num_layers: 16 + output_dim: 1024 + positional_embeddings: null + action_env_dim: 6 + future_action_window_size: 0 + past_action_window_size: 0 +datasets: + vla_data: + dataset_py: lerobot_datasets + include_state: true + data_root_dir: / + data_mix: train + eval_data_mix: validation + custom_mixtures_path: /mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/zero_latency/vla/mixture.json + action_type: continuous + sequential_step_sampling: true + eval_sequential_step_sampling: null + num_workers: 4 + eval_num_workers: 0 + prefetch_factor: 2 + persistent_workers: true + pin_memory: true + shuffle: true + action_balance: + enabled: false + strategy: balanced_epoch + action_key: action_id + target_flap_fraction: 0.3 + noop_id: 0 + flap_id: 1 + latency_curriculum: + enabled: false + strategy: exclusive + latencies: null + phase_steps: null + phase_distributions: null + new_latency_passes: 1.0 + replay_passes: 0.25 + target_total_passes: 2.0 + final_equalization: true + step_budget_mode: auto + eval_at_phase_end: false + save_at_phase_end: false + computed_plan: null + per_device_batch_size: 32 + load_all_data_for_training: true + num_obs_frames: 1 + image_mode: single + prompt_mode: raw + stitch_grid: + - 2 + - 2 + obs_image_size: + - 224 + - 224 + video_backend: pyav + lerobot_version: v2.0 + gymnasium_task_contract: + action_space: + dtype: float32 + high: + - 1.0 + - 1.0 + - 1.0 + - 1.0 + - 1.0 + - 1.0 + labels: + - right_thigh_torque + - right_leg_torque + - right_foot_torque + - left_thigh_torque + - left_leg_torque + - left_foot_torque + low: + - -1.0 + - -1.0 + - -1.0 + - -1.0 + - -1.0 + - -1.0 + type: box + base_prompt: Move the Walker2d robot forward while keeping its torso upright. + Predict six continuous torques in [-1, 1] ordered as right thigh, right leg, + right foot, left thigh, left leg, and left foot. + env_fps: 5 + env_id: LatencyBench/Walker2dContinuous-v0 + frame_stack: 1 + make_kwargs: + base_env_id: Walker2d-v4 + base_make_kwargs: + ctrl_cost_weight: 0.001 + exclude_current_positions_from_observation: true + forward_reward_weight: 1.0 + healthy_reward: 1.0 + reset_noise_scale: 0.005 + terminate_when_unhealthy: true + render_mode: rgb_array + noop_action: + - 0.0 + - 0.0 + - 0.0 + - 0.0 + - 0.0 + - 0.0 + obs_fps: 5 + registration_imports: + - latency_bench.envs.gymnasium_walker2d + state_space: + labels: + - torso_height + - torso_angle + - right_thigh_angle + - right_leg_angle + - right_foot_angle + - left_thigh_angle + - left_leg_angle + - left_foot_angle + - torso_x_velocity + - torso_z_velocity + - torso_angular_velocity + - right_thigh_angular_velocity + - right_leg_angular_velocity + - right_foot_angular_velocity + - left_thigh_angular_velocity + - left_leg_angular_velocity + - left_foot_angular_velocity + task_name: walker2d_rgb_state + active_action_dim: 6 +dataset: + source_hf: '' + config_name: null + source_subdir: null + converted_name: flappy_train + single_source_hf: '' + mixed_source_hf: '' + single_converted_name: flappy_train + mixed_converted_name: flappy_mixed_latency_train + single_latency_filter: null + mixed_latency_filter: null + force_download: false + setup_force: false + skip_verification: false + target_latency_unit: observation_steps + verify_rows: 200 + max_episodes: null + episodes_per_latency: null + latency_filter: null + debug_subset: + enabled: false + max_episodes: 5 + suffix: debug +base_model: + repo_id: Qwen/Qwen3-VL-4B-Instruct +initialization: + checkpoint_local_dir: null + checkpoint_hf_repo_id: null + checkpoint_filename: null +trainer: + max_train_steps: 5000 + num_warmup_steps: 100 + save_interval: 500 + eval_interval: 500 + eval_num_batches: 200 + per_latency_eval_num_batches: null + eval_action_classification: false + eval_action_classification_interval: null + cc_f1_tolerance: 1 + learning_rate: + base: 2.0e-05 + qwen_vl_interface: 1.0e-05 + action_model: 0.0001 + lr_scheduler_type: cosine_with_min_lr + scheduler_specific_kwargs: + min_lr: 1.0e-06 + freeze_modules: '' + freeze_vit: false + freeze_tied_embedding: false + freeze_llm_layers: [] + loss_scale: + vla: 1.0 + vlm: 0.1 + max_grad_norm: 1.0 + weight_decay: 0.0 + logging_frequency: 1 + profile_timing: + enabled: false + log_interval: 10 + gradient_clipping: 1.0 + gradient_accumulation_steps: 1 + distributed_backend: deepspeed + is_resume: false + pretrained_checkpoint: null + resume_step: 0 + reload_modules: null + optimizer: + name: AdamW + betas: + - 0.9 + - 0.95 + eps: 1.0e-08 + weight_decay: 1.0e-08 + fused: true + save_format: pt +run_id: walker2d_l0_gr00t_h1_5fps_2h200_20260917 +run_root_dir: /mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/zero_latency/vla/training +seed: 42 +is_debug: false +version_id: '0.21' +wandb_project: latency-sensitive-bench +wandb_entity: dongqianyu99-zhejiang-university +wandb_name: walker2d_l0_gr00t_h1_5fps_2h200_20260917 +wandb_group: gr00t-six-env-h1-5fps +wandb_tags: +- walker2d +- zero_latency +- GR00T +- h1 +training_latency_condition: zero_latency +backbone: + repo_id: Qwen/Qwen3-VL-4B-Instruct + revision: ebb281ec70b05090aa6165b016eac8ec08e71b17 +checkpoint: + save_best_model: false + save_final_model: true + save_pt_file: true + save_training_state: true + local: + keep_last_n: 1 +config_yaml: /mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/zero_latency/train.yaml +output_dir: /mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/zero_latency/vla/training/walker2d_l0_gr00t_h1_5fps_2h200_20260917 diff --git a/zero-latency/walker2d/vla/starvla-qwengr00t-h1/walker2d_l0_gr00t_h1_5fps_2h200_20260917/config.yaml b/zero-latency/walker2d/vla/starvla-qwengr00t-h1/walker2d_l0_gr00t_h1_5fps_2h200_20260917/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..d3e418e22db02a97577e775f225a3d7c0f9a1525 --- /dev/null +++ b/zero-latency/walker2d/vla/starvla-qwengr00t-h1/walker2d_l0_gr00t_h1_5fps_2h200_20260917/config.yaml @@ -0,0 +1,138 @@ +framework: + name: QwenGR00T + qwenvl: + base_vlm: ${oc.env:GR00T_BACKBONE_DIR} + attn_implementation: flash_attention_2 + vl_hidden_dim: 2048 + enable_gradient_checkpointing: true + action_model: + action_model_type: DiT-B + action_hidden_dim: 1024 + hidden_size: 1024 + add_pos_embed: true + max_seq_len: 1024 + action_dim: 6 + state_dim: 17 + task_objective: null + action_horizon: 1 + repeated_diffusion_steps: 8 + noise_beta_alpha: 1.5 + noise_beta_beta: 1.0 + noise_s: 0.999 + num_timestep_buckets: 1000 + num_inference_timesteps: 4 + num_target_vision_tokens: 32 + diffusion_model_cfg: + cross_attention_dim: 2560 + dropout: 0.2 + final_dropout: true + interleave_self_attention: true + norm_type: ada_norm + num_layers: 16 + output_dim: 1024 + positional_embeddings: null + action_env_dim: 6 + future_action_window_size: 0 + past_action_window_size: 0 +version_id: '0.21' +datasets: + vla_data: + include_state: true + obs_image_size: + - 224 + - 224 + image_mode: single + stitch_grid: + - 2 + - 2 + num_obs_frames: 1 + gymnasium_task_contract: &id001 + action_space: + dtype: float32 + high: + - 1.0 + - 1.0 + - 1.0 + - 1.0 + - 1.0 + - 1.0 + labels: + - right_thigh_torque + - right_leg_torque + - right_foot_torque + - left_thigh_torque + - left_leg_torque + - left_foot_torque + low: + - -1.0 + - -1.0 + - -1.0 + - -1.0 + - -1.0 + - -1.0 + type: box + base_prompt: Move the Walker2d robot forward while keeping its torso upright. + Predict six continuous torques in [-1, 1] ordered as right thigh, right leg, + right foot, left thigh, left leg, and left foot. + env_fps: 5 + env_id: LatencyBench/Walker2dContinuous-v0 + frame_stack: 1 + make_kwargs: + base_env_id: Walker2d-v4 + base_make_kwargs: + ctrl_cost_weight: 0.001 + exclude_current_positions_from_observation: true + forward_reward_weight: 1.0 + healthy_reward: 1.0 + reset_noise_scale: 0.005 + terminate_when_unhealthy: true + render_mode: rgb_array + noop_action: + - 0.0 + - 0.0 + - 0.0 + - 0.0 + - 0.0 + - 0.0 + obs_fps: 5 + registration_imports: + - latency_bench.envs.gymnasium_walker2d + state_space: + labels: + - torso_height + - torso_angle + - right_thigh_angle + - right_leg_angle + - right_foot_angle + - left_thigh_angle + - left_leg_angle + - left_foot_angle + - torso_x_velocity + - torso_z_velocity + - torso_angular_velocity + - right_thigh_angular_velocity + - right_leg_angular_velocity + - right_foot_angular_velocity + - left_thigh_angular_velocity + - left_leg_angular_velocity + - left_foot_angular_velocity + task_name: walker2d_rgb_state +trainer: + pretrained_checkpoint: null + profile_timing: + enabled: false + log_interval: 10 +backbone: + repo_id: Qwen/Qwen3-VL-4B-Instruct + revision: ebb281ec70b05090aa6165b016eac8ec08e71b17 +rl_games: + model_alias: gr00t + task: gymnasium + action_carrier: native + gymnasium: + task_contract: *id001 + env_eval: + image_size: 224 + frameskip: 1 + image_transform: raw_rgb + prompt_mode: raw diff --git a/zero-latency/walker2d/vla/starvla-qwengr00t-h1/walker2d_l0_gr00t_h1_5fps_2h200_20260917/dataset_statistics.json b/zero-latency/walker2d/vla/starvla-qwengr00t-h1/walker2d_l0_gr00t_h1_5fps_2h200_20260917/dataset_statistics.json new file mode 100644 index 0000000000000000000000000000000000000000..bfb9e72f511b15f1aba80c6eb1faa007fa88cee2 --- /dev/null +++ b/zero-latency/walker2d/vla/starvla-qwengr00t-h1/walker2d_l0_gr00t_h1_5fps_2h200_20260917/dataset_statistics.json @@ -0,0 +1,180 @@ +{ + "new_embodiment": { + "action": { + "mean": [ + 0.8840610980987549, + 0.34449484944343567, + 0.11654752492904663, + 0.8480036854743958, + 0.32758569717407227, + 0.9875396490097046 + ], + "std": [ + 0.32312333583831787, + 0.7096099853515625, + 0.9048961400985718, + 0.343753844499588, + 0.7393367886543274, + 0.11781969666481018 + ], + "max": [ + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0 + ], + "min": [ + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0 + ], + "q01": [ + -0.5243898153305053, + -1.0, + -1.0, + -0.5588397586345673, + -1.0, + 0.5681399703025818 + ], + "q99": [ + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0 + ], + "mask": [ + true, + true, + true, + true, + true, + true + ] + }, + "state": { + "mean": [ + 0.38908126950263977, + -0.06854692101478577, + 0.5603042840957642, + 0.295283704996109, + 0.20391421020030975, + 0.7933022975921631, + 0.32835325598716736, + 0.7283809781074524, + -0.11274679005146027, + 0.36438047885894775, + -0.041566070169210434, + -0.006785172037780285, + -0.21467362344264984, + -0.2302139699459076, + -0.008182121440768242, + -0.19842907786369324, + -0.0016649218741804361 + ], + "std": [ + 0.15210415422916412, + 0.28636232018470764, + 0.08957907557487488, + 0.40512198209762573, + 0.5324621796607971, + 0.09768694639205933, + 0.4175085127353668, + 0.07296281307935677, + 0.3539130389690399, + 0.21250014007091522, + 0.5710971355438232, + 0.19414709508419037, + 0.797458827495575, + 0.7378174662590027, + 0.32647520303726196, + 0.7747778296470642, + 0.10512100160121918 + ], + "max": [ + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0 + ], + "min": [ + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0 + ], + "q01": [ + -0.01641636550426483, + -0.5725254046916962, + 0.1497414314746857, + -0.5082594507932663, + -0.704446622133255, + 0.4100250256061554, + -0.7682463544607162, + 0.7055095672607422, + -0.8300059294700622, + -0.09350170254707335, + -1.0, + -0.7184711527824402, + -1.0, + -1.0, + -0.9244865983724594, + -1.0, + -0.37168214678764344 + ], + "q99": [ + 0.6826163899898535, + 0.5332732021808625, + 0.7265087735652924, + 0.9429451179504394, + 0.963300715684891, + 0.9189580070972443, + 0.8997541296482087, + 0.7992216396331789, + 0.7250751173496247, + 0.9080468678474427, + 1.0, + 0.8595866250991845, + 1.0, + 1.0, + 1.0, + 1.0, + 0.04421793460845973 + ] + }, + "num_transitions": 85160, + "num_trajectories": 90 + } +} \ No newline at end of file diff --git a/zero-latency/walker2d/vla/starvla-qwengr00t-h1/walker2d_l0_gr00t_h1_5fps_2h200_20260917/evaluation/l0_sim20/config.yaml b/zero-latency/walker2d/vla/starvla-qwengr00t-h1/walker2d_l0_gr00t_h1_5fps_2h200_20260917/evaluation/l0_sim20/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..6a76b63f577ea32bb5e3a4106782c3ada18759a8 --- /dev/null +++ b/zero-latency/walker2d/vla/starvla-qwengr00t-h1/walker2d_l0_gr00t_h1_5fps_2h200_20260917/evaluation/l0_sim20/config.yaml @@ -0,0 +1,161 @@ +experiment: + name: gr00t_walker2d_zero + seed: 42 +backend: + type: sample_factory + algo: APPO + device: cuda + train_dir: /mnt/checkpoints/latency-sensitive-bench/small_models/walker2d + restart_behavior: overwrite + run_mode: eval +executor: + mode: simulated + inference_devices: + - cuda:0 + inference_batch_size: 1 + simulated_inference_pool: true + simulated_worker_capacity: 1 +env: + name: gymnasium + task_name: walker2d + env_id: LatencyBench/Walker2dContinuous-v0 + registration_imports: + - latency_bench.envs.gymnasium_walker2d + make_kwargs: + base_env_id: Walker2d-v4 + base_make_kwargs: + ctrl_cost_weight: 0.001 + exclude_current_positions_from_observation: true + forward_reward_weight: 1.0 + healthy_reward: 1.0 + reset_noise_scale: 0.005 + terminate_when_unhealthy: true + render_mode: rgb_array + env_fps: 5 + obs_fps: 5 + frame_stack: 1 + action_space: + dtype: float32 + high: + - 1.0 + - 1.0 + - 1.0 + - 1.0 + - 1.0 + - 1.0 + labels: + - right_thigh_torque + - right_leg_torque + - right_foot_torque + - left_thigh_torque + - left_leg_torque + - left_foot_torque + low: + - -1.0 + - -1.0 + - -1.0 + - -1.0 + - -1.0 + - -1.0 + type: box + noop_action: + - 0.0 + - 0.0 + - 0.0 + - 0.0 + - 0.0 + - 0.0 + base_prompt: Move the Walker2d robot forward while keeping its torso upright. Predict + six continuous torques in [-1, 1] ordered as right thigh, right leg, right foot, + left thigh, left leg, and left foot. + state_space: + labels: + - torso_height + - torso_angle + - right_thigh_angle + - right_leg_angle + - right_foot_angle + - left_thigh_angle + - left_leg_angle + - left_foot_angle + - torso_x_velocity + - torso_z_velocity + - torso_angular_velocity + - right_thigh_angular_velocity + - right_leg_angular_velocity + - right_foot_angular_velocity + - left_thigh_angular_velocity + - left_leg_angular_velocity + - left_foot_angular_velocity + obs_resize: + - 224 + - 224 +latency: + method: zero + sync_cuda: false +scheduler: + hold_policy: hold + ordering_policy: issue_order_fifo +policy: + type: starvla + checkpoint_path: /mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/zero_latency/bundle/checkpoints/model.pt + model_config_path: /mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/zero_latency/bundle/config.yaml + backbone_path: /mnt/local/lzj/latency-sensitive-bench/models/Qwen3-VL-4B-Instruct + device: cuda:0 + task_manifest_path: /mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/zero_latency/bundle/manifest.json + latency_prompt_map_path: /mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/zero_latency/bundle/latency_prompt_map.json + latency_prompt_key: 0 +training: + train_for_env_steps: 10000000 + num_workers: 8 + num_envs_per_worker: 8 + worker_num_splits: 2 + num_policies: 1 + batch_size: 1024 + rollout: 64 + recurrence: 1 + num_epochs: 2 + num_batches_per_epoch: 4 + max_policy_lag: 300 + learning_rate: 0.00295 + gamma: 0.99 + gae_lambda: 0.95 + ppo_clip_ratio: 0.2 + ppo_clip_value: 0.2 + value_loss_coeff: 1.3 + max_grad_norm: 3.5 + exploration_loss_coeff: 0.0 + async_rl: false + batched_sampling: false + use_rnn: false + encoder_mlp_layers: + - 64 + - 64 + normalize_input: true + normalize_returns: true + save_every_sec: 600 + keep_checkpoints: 5 + lr_schedule: linear_decay + kl_loss_coeff: 0.1 + serial_mode: false + nonlinearity: tanh + adaptive_stddev: false + policy_initialization: torch_default + initial_stddev: 1.0 + shuffle_minibatches: false + value_bootstrap: true +evaluation: + eval_episodes: 20 + eval_parallel_envs: 1 + eval_max_steps: 1000 + eval_deterministic: true + eval_latency_values: null +logging: + output_dir: /mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/zero_latency/l0_sim20 + video: + enabled: false + num_bins: 1 + save_step_records: true + save_action_records: true + save_latency_records: true + realtime_pipeline_profile: false diff --git a/zero-latency/walker2d/vla/starvla-qwengr00t-h1/walker2d_l0_gr00t_h1_5fps_2h200_20260917/evaluation/l0_sim20/episode_metrics.jsonl b/zero-latency/walker2d/vla/starvla-qwengr00t-h1/walker2d_l0_gr00t_h1_5fps_2h200_20260917/evaluation/l0_sim20/episode_metrics.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..321a65c0faf5701a60533e322fa9a0071b625b46 --- /dev/null +++ b/zero-latency/walker2d/vla/starvla-qwengr00t-h1/walker2d_l0_gr00t_h1_5fps_2h200_20260917/evaluation/l0_sim20/episode_metrics.jsonl @@ -0,0 +1,20 @@ +{"episode_id": 0, "episode_return": 3808.226836737725, "episode_return_env": 3808.226836737725, "game_score": null, "mean_latency_ms": 0.0, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/zero_latency/bundle/checkpoints/model.pt", "config_name": "gr00t_walker2d_zero", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Walker2dContinuous-v0", "episode_seed": 42, "frame_ms": 200.0, "gpu_class": null, "idle_worker_count": 0, "in_flight_count": 1, "inference_worker_count": 1, "instance_id": null, "latency_type": "zero", "mode": "simulated", "model_id": null, "obs_fps": 5.0, "output_dir": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/zero_latency/l0_sim20", "policy_id": "starvla", "profile_ref": null, "run_name": "gr00t_walker2d_zero", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_run_id": null, "submitted_observation_frames": 1000, "workload_id": null}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 0.0, "p99_latency_ms": 0.0, "return_raw": null, "survival_steps": 1000} +{"episode_id": 1, "episode_return": 3853.1032959767176, "episode_return_env": 3853.1032959767176, "game_score": null, "mean_latency_ms": 0.0, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/zero_latency/bundle/checkpoints/model.pt", "config_name": "gr00t_walker2d_zero", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Walker2dContinuous-v0", "episode_seed": 43, "frame_ms": 200.0, "gpu_class": null, "idle_worker_count": 0, "in_flight_count": 1, "inference_worker_count": 1, "instance_id": null, "latency_type": "zero", "mode": "simulated", "model_id": null, "obs_fps": 5.0, "output_dir": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/zero_latency/l0_sim20", "policy_id": "starvla", "profile_ref": null, "run_name": "gr00t_walker2d_zero", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_run_id": null, "submitted_observation_frames": 1000, "workload_id": null}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 0.0, "p99_latency_ms": 0.0, "return_raw": null, "survival_steps": 1000} +{"episode_id": 2, "episode_return": 1723.999349832787, "episode_return_env": 1723.999349832787, "game_score": null, "mean_latency_ms": 0.0, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/zero_latency/bundle/checkpoints/model.pt", "config_name": "gr00t_walker2d_zero", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Walker2dContinuous-v0", "episode_seed": 44, "frame_ms": 200.0, "gpu_class": null, "idle_worker_count": 0, "in_flight_count": 1, "inference_worker_count": 1, "instance_id": null, "latency_type": "zero", "mode": "simulated", "model_id": null, "obs_fps": 5.0, "output_dir": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/zero_latency/l0_sim20", "policy_id": "starvla", "profile_ref": null, "run_name": "gr00t_walker2d_zero", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_run_id": null, "submitted_observation_frames": 514, "workload_id": null}, "num_actions": 514, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 0.0, "p99_latency_ms": 0.0, "return_raw": null, "survival_steps": 514} +{"episode_id": 3, "episode_return": 4140.340620286889, "episode_return_env": 4140.340620286889, "game_score": null, "mean_latency_ms": 0.0, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/zero_latency/bundle/checkpoints/model.pt", "config_name": "gr00t_walker2d_zero", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Walker2dContinuous-v0", "episode_seed": 45, "frame_ms": 200.0, "gpu_class": null, "idle_worker_count": 0, "in_flight_count": 1, "inference_worker_count": 1, "instance_id": null, "latency_type": "zero", "mode": "simulated", "model_id": null, "obs_fps": 5.0, "output_dir": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/zero_latency/l0_sim20", "policy_id": "starvla", "profile_ref": null, "run_name": "gr00t_walker2d_zero", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_run_id": null, "submitted_observation_frames": 1000, "workload_id": null}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 0.0, "p99_latency_ms": 0.0, "return_raw": null, "survival_steps": 1000} +{"episode_id": 4, "episode_return": 4096.512537306539, "episode_return_env": 4096.512537306539, "game_score": null, "mean_latency_ms": 0.0, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/zero_latency/bundle/checkpoints/model.pt", "config_name": "gr00t_walker2d_zero", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Walker2dContinuous-v0", "episode_seed": 46, "frame_ms": 200.0, "gpu_class": null, "idle_worker_count": 0, "in_flight_count": 1, "inference_worker_count": 1, "instance_id": null, "latency_type": "zero", "mode": "simulated", "model_id": null, "obs_fps": 5.0, "output_dir": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/zero_latency/l0_sim20", "policy_id": "starvla", "profile_ref": null, "run_name": "gr00t_walker2d_zero", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_run_id": null, "submitted_observation_frames": 972, "workload_id": null}, "num_actions": 972, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 0.0, "p99_latency_ms": 0.0, "return_raw": null, "survival_steps": 972} +{"episode_id": 5, "episode_return": 3788.9854100645875, "episode_return_env": 3788.9854100645875, "game_score": null, "mean_latency_ms": 0.0, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/zero_latency/bundle/checkpoints/model.pt", "config_name": "gr00t_walker2d_zero", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Walker2dContinuous-v0", "episode_seed": 47, "frame_ms": 200.0, "gpu_class": null, "idle_worker_count": 0, "in_flight_count": 1, "inference_worker_count": 1, "instance_id": null, "latency_type": "zero", "mode": "simulated", "model_id": null, "obs_fps": 5.0, "output_dir": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/zero_latency/l0_sim20", "policy_id": "starvla", "profile_ref": null, "run_name": "gr00t_walker2d_zero", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_run_id": null, "submitted_observation_frames": 890, "workload_id": null}, "num_actions": 890, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 0.0, "p99_latency_ms": 0.0, "return_raw": null, "survival_steps": 890} +{"episode_id": 6, "episode_return": 4156.574688705178, "episode_return_env": 4156.574688705178, "game_score": null, "mean_latency_ms": 0.0, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/zero_latency/bundle/checkpoints/model.pt", "config_name": "gr00t_walker2d_zero", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Walker2dContinuous-v0", "episode_seed": 48, "frame_ms": 200.0, "gpu_class": null, "idle_worker_count": 0, "in_flight_count": 1, "inference_worker_count": 1, "instance_id": null, "latency_type": "zero", "mode": "simulated", "model_id": null, "obs_fps": 5.0, "output_dir": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/zero_latency/l0_sim20", "policy_id": "starvla", "profile_ref": null, "run_name": "gr00t_walker2d_zero", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_run_id": null, "submitted_observation_frames": 1000, "workload_id": null}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 0.0, "p99_latency_ms": 0.0, "return_raw": null, "survival_steps": 1000} +{"episode_id": 7, "episode_return": 3251.442409375431, "episode_return_env": 3251.442409375431, "game_score": null, "mean_latency_ms": 0.0, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/zero_latency/bundle/checkpoints/model.pt", "config_name": "gr00t_walker2d_zero", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Walker2dContinuous-v0", "episode_seed": 49, "frame_ms": 200.0, "gpu_class": null, "idle_worker_count": 0, "in_flight_count": 1, "inference_worker_count": 1, "instance_id": null, "latency_type": "zero", "mode": "simulated", "model_id": null, "obs_fps": 5.0, "output_dir": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/zero_latency/l0_sim20", "policy_id": "starvla", "profile_ref": null, "run_name": "gr00t_walker2d_zero", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_run_id": null, "submitted_observation_frames": 1000, "workload_id": null}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 0.0, "p99_latency_ms": 0.0, "return_raw": null, "survival_steps": 1000} +{"episode_id": 8, "episode_return": 4033.286305383909, "episode_return_env": 4033.286305383909, "game_score": null, "mean_latency_ms": 0.0, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/zero_latency/bundle/checkpoints/model.pt", "config_name": "gr00t_walker2d_zero", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Walker2dContinuous-v0", "episode_seed": 50, "frame_ms": 200.0, "gpu_class": null, "idle_worker_count": 0, "in_flight_count": 1, "inference_worker_count": 1, "instance_id": null, "latency_type": "zero", "mode": "simulated", "model_id": null, "obs_fps": 5.0, "output_dir": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/zero_latency/l0_sim20", "policy_id": "starvla", "profile_ref": null, "run_name": "gr00t_walker2d_zero", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_run_id": null, "submitted_observation_frames": 1000, "workload_id": null}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 0.0, "p99_latency_ms": 0.0, "return_raw": null, "survival_steps": 1000} +{"episode_id": 9, "episode_return": 4025.248397336756, "episode_return_env": 4025.248397336756, "game_score": null, "mean_latency_ms": 0.0, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/zero_latency/bundle/checkpoints/model.pt", "config_name": "gr00t_walker2d_zero", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Walker2dContinuous-v0", "episode_seed": 51, "frame_ms": 200.0, "gpu_class": null, "idle_worker_count": 0, "in_flight_count": 1, "inference_worker_count": 1, "instance_id": null, "latency_type": "zero", "mode": "simulated", "model_id": null, "obs_fps": 5.0, "output_dir": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/zero_latency/l0_sim20", "policy_id": "starvla", "profile_ref": null, "run_name": "gr00t_walker2d_zero", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_run_id": null, "submitted_observation_frames": 1000, "workload_id": null}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 0.0, "p99_latency_ms": 0.0, "return_raw": null, "survival_steps": 1000} +{"episode_id": 10, "episode_return": 4105.286382644159, "episode_return_env": 4105.286382644159, "game_score": null, "mean_latency_ms": 0.0, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/zero_latency/bundle/checkpoints/model.pt", "config_name": "gr00t_walker2d_zero", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Walker2dContinuous-v0", "episode_seed": 52, "frame_ms": 200.0, "gpu_class": null, "idle_worker_count": 0, "in_flight_count": 1, "inference_worker_count": 1, "instance_id": null, "latency_type": "zero", "mode": "simulated", "model_id": null, "obs_fps": 5.0, "output_dir": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/zero_latency/l0_sim20", "policy_id": "starvla", "profile_ref": null, "run_name": "gr00t_walker2d_zero", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_run_id": null, "submitted_observation_frames": 1000, "workload_id": null}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 0.0, "p99_latency_ms": 0.0, "return_raw": null, "survival_steps": 1000} +{"episode_id": 11, "episode_return": 4124.8478277361955, "episode_return_env": 4124.8478277361955, "game_score": null, "mean_latency_ms": 0.0, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/zero_latency/bundle/checkpoints/model.pt", "config_name": "gr00t_walker2d_zero", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Walker2dContinuous-v0", "episode_seed": 53, "frame_ms": 200.0, "gpu_class": null, "idle_worker_count": 0, "in_flight_count": 1, "inference_worker_count": 1, "instance_id": null, "latency_type": "zero", "mode": "simulated", "model_id": null, "obs_fps": 5.0, "output_dir": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/zero_latency/l0_sim20", "policy_id": "starvla", "profile_ref": null, "run_name": "gr00t_walker2d_zero", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_run_id": null, "submitted_observation_frames": 1000, "workload_id": null}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 0.0, "p99_latency_ms": 0.0, "return_raw": null, "survival_steps": 1000} +{"episode_id": 12, "episode_return": 1718.3891440521081, "episode_return_env": 1718.3891440521081, "game_score": null, "mean_latency_ms": 0.0, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/zero_latency/bundle/checkpoints/model.pt", "config_name": "gr00t_walker2d_zero", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Walker2dContinuous-v0", "episode_seed": 54, "frame_ms": 200.0, "gpu_class": null, "idle_worker_count": 0, "in_flight_count": 1, "inference_worker_count": 1, "instance_id": null, "latency_type": "zero", "mode": "simulated", "model_id": null, "obs_fps": 5.0, "output_dir": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/zero_latency/l0_sim20", "policy_id": "starvla", "profile_ref": null, "run_name": "gr00t_walker2d_zero", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_run_id": null, "submitted_observation_frames": 519, "workload_id": null}, "num_actions": 519, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 0.0, "p99_latency_ms": 0.0, "return_raw": null, "survival_steps": 519} +{"episode_id": 13, "episode_return": 773.3146074280513, "episode_return_env": 773.3146074280513, "game_score": null, "mean_latency_ms": 0.0, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/zero_latency/bundle/checkpoints/model.pt", "config_name": "gr00t_walker2d_zero", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Walker2dContinuous-v0", "episode_seed": 55, "frame_ms": 200.0, "gpu_class": null, "idle_worker_count": 0, "in_flight_count": 1, "inference_worker_count": 1, "instance_id": null, "latency_type": "zero", "mode": "simulated", "model_id": null, "obs_fps": 5.0, "output_dir": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/zero_latency/l0_sim20", "policy_id": "starvla", "profile_ref": null, "run_name": "gr00t_walker2d_zero", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_run_id": null, "submitted_observation_frames": 282, "workload_id": null}, "num_actions": 282, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 0.0, "p99_latency_ms": 0.0, "return_raw": null, "survival_steps": 282} +{"episode_id": 14, "episode_return": 3239.1715436786335, "episode_return_env": 3239.1715436786335, "game_score": null, "mean_latency_ms": 0.0, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/zero_latency/bundle/checkpoints/model.pt", "config_name": "gr00t_walker2d_zero", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Walker2dContinuous-v0", "episode_seed": 56, "frame_ms": 200.0, "gpu_class": null, "idle_worker_count": 0, "in_flight_count": 1, "inference_worker_count": 1, "instance_id": null, "latency_type": "zero", "mode": "simulated", "model_id": null, "obs_fps": 5.0, "output_dir": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/zero_latency/l0_sim20", "policy_id": "starvla", "profile_ref": null, "run_name": "gr00t_walker2d_zero", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_run_id": null, "submitted_observation_frames": 1000, "workload_id": null}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 0.0, "p99_latency_ms": 0.0, "return_raw": null, "survival_steps": 1000} +{"episode_id": 15, "episode_return": 3784.661174823245, "episode_return_env": 3784.661174823245, "game_score": null, "mean_latency_ms": 0.0, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/zero_latency/bundle/checkpoints/model.pt", "config_name": "gr00t_walker2d_zero", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Walker2dContinuous-v0", "episode_seed": 57, "frame_ms": 200.0, "gpu_class": null, "idle_worker_count": 0, "in_flight_count": 1, "inference_worker_count": 1, "instance_id": null, "latency_type": "zero", "mode": "simulated", "model_id": null, "obs_fps": 5.0, "output_dir": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/zero_latency/l0_sim20", "policy_id": "starvla", "profile_ref": null, "run_name": "gr00t_walker2d_zero", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_run_id": null, "submitted_observation_frames": 1000, "workload_id": null}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 0.0, "p99_latency_ms": 0.0, "return_raw": null, "survival_steps": 1000} +{"episode_id": 16, "episode_return": 3999.8377210951776, "episode_return_env": 3999.8377210951776, "game_score": null, "mean_latency_ms": 0.0, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/zero_latency/bundle/checkpoints/model.pt", "config_name": "gr00t_walker2d_zero", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Walker2dContinuous-v0", "episode_seed": 58, "frame_ms": 200.0, "gpu_class": null, "idle_worker_count": 0, "in_flight_count": 1, "inference_worker_count": 1, "instance_id": null, "latency_type": "zero", "mode": "simulated", "model_id": null, "obs_fps": 5.0, "output_dir": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/zero_latency/l0_sim20", "policy_id": "starvla", "profile_ref": null, "run_name": "gr00t_walker2d_zero", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_run_id": null, "submitted_observation_frames": 1000, "workload_id": null}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 0.0, "p99_latency_ms": 0.0, "return_raw": null, "survival_steps": 1000} +{"episode_id": 17, "episode_return": 4091.4434992007014, "episode_return_env": 4091.4434992007014, "game_score": null, "mean_latency_ms": 0.0, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/zero_latency/bundle/checkpoints/model.pt", "config_name": "gr00t_walker2d_zero", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Walker2dContinuous-v0", "episode_seed": 59, "frame_ms": 200.0, "gpu_class": null, "idle_worker_count": 0, "in_flight_count": 1, "inference_worker_count": 1, "instance_id": null, "latency_type": "zero", "mode": "simulated", "model_id": null, "obs_fps": 5.0, "output_dir": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/zero_latency/l0_sim20", "policy_id": "starvla", "profile_ref": null, "run_name": "gr00t_walker2d_zero", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_run_id": null, "submitted_observation_frames": 1000, "workload_id": null}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 0.0, "p99_latency_ms": 0.0, "return_raw": null, "survival_steps": 1000} +{"episode_id": 18, "episode_return": 3659.760759472817, "episode_return_env": 3659.760759472817, "game_score": null, "mean_latency_ms": 0.0, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/zero_latency/bundle/checkpoints/model.pt", "config_name": "gr00t_walker2d_zero", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Walker2dContinuous-v0", "episode_seed": 60, "frame_ms": 200.0, "gpu_class": null, "idle_worker_count": 0, "in_flight_count": 1, "inference_worker_count": 1, "instance_id": null, "latency_type": "zero", "mode": "simulated", "model_id": null, "obs_fps": 5.0, "output_dir": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/zero_latency/l0_sim20", "policy_id": "starvla", "profile_ref": null, "run_name": "gr00t_walker2d_zero", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_run_id": null, "submitted_observation_frames": 852, "workload_id": null}, "num_actions": 852, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 0.0, "p99_latency_ms": 0.0, "return_raw": null, "survival_steps": 852} +{"episode_id": 19, "episode_return": 4185.446772066988, "episode_return_env": 4185.446772066988, "game_score": null, "mean_latency_ms": 0.0, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/zero_latency/bundle/checkpoints/model.pt", "config_name": "gr00t_walker2d_zero", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Walker2dContinuous-v0", "episode_seed": 61, "frame_ms": 200.0, "gpu_class": null, "idle_worker_count": 0, "in_flight_count": 1, "inference_worker_count": 1, "instance_id": null, "latency_type": "zero", "mode": "simulated", "model_id": null, "obs_fps": 5.0, "output_dir": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/zero_latency/l0_sim20", "policy_id": "starvla", "profile_ref": null, "run_name": "gr00t_walker2d_zero", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_run_id": null, "submitted_observation_frames": 1000, "workload_id": null}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 0.0, "p99_latency_ms": 0.0, "return_raw": null, "survival_steps": 1000} diff --git a/zero-latency/walker2d/vla/starvla-qwengr00t-h1/walker2d_l0_gr00t_h1_5fps_2h200_20260917/evaluation/l0_sim20/verification.json b/zero-latency/walker2d/vla/starvla-qwengr00t-h1/walker2d_l0_gr00t_h1_5fps_2h200_20260917/evaluation/l0_sim20/verification.json new file mode 100644 index 0000000000000000000000000000000000000000..a6535a1ea57c992b17b8636dc939f4b899db2432 --- /dev/null +++ b/zero-latency/walker2d/vla/starvla-qwengr00t-h1/walker2d_l0_gr00t_h1_5fps_2h200_20260917/evaluation/l0_sim20/verification.json @@ -0,0 +1,95 @@ +{ + "task": "walker2d", + "state": "L0_SIM20_COMPLETED_AND_VERIFIED", + "verified_at": "2026-09-17T15:02:36.902285+00:00", + "evaluation_finished_at": "2026-09-17T14:44:50.630321+00:00", + "episodes": 20, + "seeds": [ + 42, + 43, + 44, + 45, + 46, + 47, + 48, + 49, + 50, + 51, + 52, + 53, + 54, + 55, + 56, + 57, + 58, + 59, + 60, + 61 + ], + "env_fps": 5, + "obs_fps": 5, + "latency": "zero", + "mode": "simulated", + "max_raw_steps": 1000, + "returns": [ + 3808.226836737725, + 3853.1032959767176, + 1723.999349832787, + 4140.340620286889, + 4096.512537306539, + 3788.9854100645875, + 4156.574688705178, + 3251.442409375431, + 4033.286305383909, + 4025.248397336756, + 4105.286382644159, + 4124.8478277361955, + 1718.3891440521081, + 773.3146074280513, + 3239.1715436786335, + 3784.661174823245, + 3999.8377210951776, + 4091.4434992007014, + 3659.760759472817, + 4185.446772066988 + ], + "mean_return": 3527.99396416023, + "population_std_return": 945.2656326298496, + "lengths": [ + 1000, + 1000, + 514, + 1000, + 972, + 890, + 1000, + 1000, + 1000, + 1000, + 1000, + 1000, + 519, + 282, + 1000, + 1000, + 1000, + 1000, + 852, + 1000 + ], + "mean_length": 901.45, + "total_raw_steps": 18029, + "raw_step_rewards_match_episode_returns": true, + "raw_step_indices_contiguous": true, + "dropped_actions": 0, + "dropped_observations": 0, + "invalid_actions": 0, + "action_events": 18029, + "all_action_latencies_zero": true, + "training_steps": 5000, + "global_batch": 64, + "checkpoint_sha256": "5df7bf29e8cc0bc984cb21ad4f8db673f16b17ceb438f4f4c91be65ee52c4de1", + "command_semantics": "Generic adapter applied_action may be preclip; native environment clipping unchanged. No preclip-bound assertion was imposed.", + "remaining": "RTX3090 P/B, actual-profile teacher/data/profile-VLA, F; not yet complete", + "acceptance": "Full pipeline acceptance not established" +} diff --git a/zero-latency/walker2d/vla/starvla-qwengr00t-h1/walker2d_l0_gr00t_h1_5fps_2h200_20260917/latency_prompt_map.json b/zero-latency/walker2d/vla/starvla-qwengr00t-h1/walker2d_l0_gr00t_h1_5fps_2h200_20260917/latency_prompt_map.json new file mode 100644 index 0000000000000000000000000000000000000000..727b2dd7642ac6926aa0451c1859cfa4ed29d397 --- /dev/null +++ b/zero-latency/walker2d/vla/starvla-qwengr00t-h1/walker2d_l0_gr00t_h1_5fps_2h200_20260917/latency_prompt_map.json @@ -0,0 +1,7 @@ +{ + "0": { + "prompt": "Move the Walker2d robot forward while keeping its torso upright. Predict six continuous torques in [-1, 1] ordered as right thigh, right leg, right foot, left thigh, left leg, and left foot. Current action latency is 0 raw frames (0.00 ms). The environment runs at 5 FPS and observations are emitted at 5 FPS. Choose the best next action.", + "latency_raw_frames": 0, + "latency_ms": 0.0 + } +} \ No newline at end of file diff --git a/zero-latency/walker2d/vla/starvla-qwengr00t-h1/walker2d_l0_gr00t_h1_5fps_2h200_20260917/manifest.json b/zero-latency/walker2d/vla/starvla-qwengr00t-h1/walker2d_l0_gr00t_h1_5fps_2h200_20260917/manifest.json new file mode 100644 index 0000000000000000000000000000000000000000..ebb173fb91b2996519c2692aceb8ee656c35b8d1 --- /dev/null +++ b/zero-latency/walker2d/vla/starvla-qwengr00t-h1/walker2d_l0_gr00t_h1_5fps_2h200_20260917/manifest.json @@ -0,0 +1,184 @@ +{ + "dataset_name": "walker2d_h1_5fps_l0", + "env_name": "walker2d_rgb_state", + "episodes": 90, + "frames": 85160, + "task_prompts": [ + "Move the Walker2d robot forward while keeping its torso upright. Predict six continuous torques in [-1, 1] ordered as right thigh, right leg, right foot, left thigh, left leg, and left foot. Current action latency is 0 raw frames (0.00 ms). The environment runs at 5 FPS and observations are emitted at 5 FPS. Choose the best next action." + ], + "format": "lerobot_v2_image_parquet", + "source": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/zero_latency/data/raw_5fps", + "integration_name": "gymnasium", + "task_name": "walker2d_rgb_state", + "action_layout": "gymnasium_continuous_v1", + "action_labels": [ + "right_thigh_torque", + "right_leg_torque", + "right_foot_torque", + "left_thigh_torque", + "left_leg_torque", + "left_foot_torque" + ], + "carrier_action_labels": [ + "right_thigh_torque", + "right_leg_torque", + "right_foot_torque", + "left_thigh_torque", + "left_leg_torque", + "left_foot_torque" + ], + "action_dim": 6, + "active_action_dim": 6, + "action_carrier": "native", + "rows_unit": "decision_step", + "obs_fps": 5, + "obs_stride_raw_frames": 1, + "uses_state": true, + "state_dim": 17, + "state_labels": [ + "torso_height", + "torso_angle", + "right_thigh_angle", + "right_leg_angle", + "right_foot_angle", + "left_thigh_angle", + "left_leg_angle", + "left_foot_angle", + "torso_x_velocity", + "torso_z_velocity", + "torso_angular_velocity", + "right_thigh_angular_velocity", + "right_leg_angular_velocity", + "right_foot_angular_velocity", + "left_thigh_angular_velocity", + "left_leg_angular_velocity", + "left_foot_angular_velocity" + ], + "state_normalization": { + "type": "min_max", + "min": [ + 0.8001501560211182, + -0.9963381290435791, + -1.2083065509796143, + -2.6031124591827393, + -1.4880801439285278, + -1.7395809888839722, + -2.8114991188049316, + -1.4173651933670044, + -1.9014521837234497, + -5.880380630493164, + -10.0, + -10.0, + -10.0, + -10.0, + -10.0, + -10.0, + -10.0 + ], + "max": [ + 1.4312387704849243, + 0.9942846298217773, + 0.35824716091156006, + 0.42332249879837036, + 1.4080666303634644, + 0.19403602182865143, + 0.508411705493927, + 1.1665540933609009, + 9.758368492126465, + 2.7117056846618652, + 10.0, + 10.0, + 10.0, + 10.0, + 10.0, + 10.0, + 10.0 + ] + }, + "robot_type": "rl_games_gymnasium", + "latency_prompt_map_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/zero_latency/data/lerobot/walker2d_h1_5fps_l0/latency_prompt_map.json", + "custom_mixtures_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/zero_latency/data/lerobot/_generated_mixtures/walker2d_h1_5fps_l0.json", + "gymnasium_task": { + "action_space": { + "dtype": "float32", + "high": [ + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0 + ], + "labels": [ + "right_thigh_torque", + "right_leg_torque", + "right_foot_torque", + "left_thigh_torque", + "left_leg_torque", + "left_foot_torque" + ], + "low": [ + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0 + ], + "type": "box" + }, + "base_prompt": "Move the Walker2d robot forward while keeping its torso upright. Predict six continuous torques in [-1, 1] ordered as right thigh, right leg, right foot, left thigh, left leg, and left foot.", + "env_fps": 5, + "env_id": "LatencyBench/Walker2dContinuous-v0", + "frame_stack": 1, + "make_kwargs": { + "base_env_id": "Walker2d-v4", + "base_make_kwargs": { + "ctrl_cost_weight": 0.001, + "exclude_current_positions_from_observation": true, + "forward_reward_weight": 1.0, + "healthy_reward": 1.0, + "reset_noise_scale": 0.005, + "terminate_when_unhealthy": true + }, + "render_mode": "rgb_array" + }, + "noop_action": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ], + "obs_fps": 5, + "registration_imports": [ + "latency_bench.envs.gymnasium_walker2d" + ], + "state_space": { + "labels": [ + "torso_height", + "torso_angle", + "right_thigh_angle", + "right_leg_angle", + "right_foot_angle", + "left_thigh_angle", + "left_leg_angle", + "left_foot_angle", + "torso_x_velocity", + "torso_z_velocity", + "torso_angular_velocity", + "right_thigh_angular_velocity", + "right_leg_angular_velocity", + "right_foot_angular_velocity", + "left_thigh_angular_velocity", + "left_leg_angular_velocity", + "left_foot_angular_velocity" + ] + }, + "task_name": "walker2d_rgb_state" + }, + "validation_dataset_name": "walker2d_h1_5fps_l0__val", + "validation_episodes": 10, + "validation_frames": 9664 +} \ No newline at end of file diff --git a/zero-latency/walker2d/vla/starvla-qwengr00t-h1/walker2d_l0_gr00t_h1_5fps_2h200_20260917/provenance.json b/zero-latency/walker2d/vla/starvla-qwengr00t-h1/walker2d_l0_gr00t_h1_5fps_2h200_20260917/provenance.json new file mode 100644 index 0000000000000000000000000000000000000000..d36ec6d6d14dfb9bcf10b50fccc045c10d9e557d --- /dev/null +++ b/zero-latency/walker2d/vla/starvla-qwengr00t-h1/walker2d_l0_gr00t_h1_5fps_2h200_20260917/provenance.json @@ -0,0 +1,34 @@ +{ + "task": "walker2d", + "model": "qwengr00t", + "training_condition": "zero-latency", + "training_run_id": "walker2d_l0_gr00t_h1_5fps_2h200_20260917", + "source": { + "repo_id": "latency-sensitive-bench/Standard-Pipeline", + "revision": "2208875f92b21e3f8ec2363a30a6705dfe56cc04", + "path": "walker2d/zero_latency/GR00T", + "url": "https://huggingface.co/latency-sensitive-bench/Standard-Pipeline/tree/2208875f92b21e3f8ec2363a30a6705dfe56cc04/walker2d/zero_latency/GR00T" + }, + "hf_repo_id": "latency-sensitive-bench/benchmark-models", + "hf_path": "zero-latency/walker2d/vla/starvla-qwengr00t-h1/walker2d_l0_gr00t_h1_5fps_2h200_20260917", + "checkpoint": { + "source_file": "walker2d/zero_latency/GR00T/checkpoints/model.pt", + "source_sha256": "5df7bf29e8cc0bc984cb21ad4f8db673f16b17ceb438f4f4c91be65ee52c4de1", + "source_bytes": 9976860963, + "file": "checkpoints/model.pt", + "selection_rule": "only published VLA checkpoint", + "method": "copy", + "sha256": "5df7bf29e8cc0bc984cb21ad4f8db673f16b17ceb438f4f4c91be65ee52c4de1", + "bytes": 9976860963 + }, + "config_source": "walker2d/zero_latency/GR00T/config.yaml", + "task_contract_file": "task_contract.json", + "missing_source_files": [], + "migration": { + "operation": "copy_selected_inference_bundle", + "source_modified": false, + "evaluation_performed": false, + "experiment_status": "preserved_in_source_evidence" + }, + "action_horizon": 1 +} diff --git a/zero-latency/walker2d/vla/starvla-qwengr00t-h1/walker2d_l0_gr00t_h1_5fps_2h200_20260917/source/README.md b/zero-latency/walker2d/vla/starvla-qwengr00t-h1/walker2d_l0_gr00t_h1_5fps_2h200_20260917/source/README.md new file mode 100644 index 0000000000000000000000000000000000000000..078612fe671b3bafcf2dda820744bd26857f7c9a --- /dev/null +++ b/zero-latency/walker2d/vla/starvla-qwengr00t-h1/walker2d_l0_gr00t_h1_5fps_2h200_20260917/source/README.md @@ -0,0 +1,7 @@ +# walker2d: GR00T H1, 5 FPS + +Fresh zero-latency training completed5000 updates. Final checkpoint SHA256: `5df7bf29e8cc0bc984cb21ad4f8db673f16b17ceb438f4f4c91be65ee52c4de1`. + +L0 simulated zero-latency evaluation is now complete and audited:20episodes, seeds42–61,5FPS, cap1000rawsteps. Return mean 3527.993964, population SD 945.265633; mean length 901.45. Dropped actions 0, dropped observations 0, invalid actions 0. Raw step rewards and lengths match episode metrics. + +See `evaluation/l0_sim20/verification.json` and `episode_metrics.jsonl` for final evidence. Earlier training/load receipts remain historical snapshots. P/B and subsequent profile pipeline/F are pending; full pipeline acceptance is not established. Set GR00T_BACKBONE_DIR to Qwen/Qwen3-VL-4B-Instruct revision ebb281ec70b05090aa6165b016eac8ec08e71b17 before inference. diff --git a/zero-latency/walker2d/vla/starvla-qwengr00t-h1/walker2d_l0_gr00t_h1_5fps_2h200_20260917/source/provenance.json b/zero-latency/walker2d/vla/starvla-qwengr00t-h1/walker2d_l0_gr00t_h1_5fps_2h200_20260917/source/provenance.json new file mode 100644 index 0000000000000000000000000000000000000000..c6ba98f5de0b2ff924f7a9467cb84f771b8a0b5b --- /dev/null +++ b/zero-latency/walker2d/vla/starvla-qwengr00t-h1/walker2d_l0_gr00t_h1_5fps_2h200_20260917/source/provenance.json @@ -0,0 +1,26 @@ +{ + "task": "walker2d", + "backbone": { + "repo_id": "Qwen/Qwen3-VL-4B-Instruct", + "revision": "ebb281ec70b05090aa6165b016eac8ec08e71b17" + }, + "checkpoint": "steps_5000_pytorch_model.pt", + "training_steps": 5000, + "seed": 42, + "global_batch": 64, + "training_run_id": "walker2d_l0_gr00t_h1_5fps_2h200_20260917", + "condition": "zero_latency", + "source": { + "repo": "latency-sensitive-bench/Standard-Pipeline", + "revision": "571cb5801720488ed6458b9cfe3f4c595b069cea", + "prefix": "walker2d/zero_latency/shared/source_100ep_v1/demonstrations/raw", + "sha256": { + "metadata.json": "c8ea2669493beabd97d9e7ecd06cc443adcb1760d8e71dd03588ea4adea20ae9", + "train.parquet": "6a105846e7def09f10fd2a985413cc9a04b9433f57519d330ce33a1033ba7e65", + "val.parquet": "687c1a5d08b3641b5ae76d218dd6afe483f8873b057aa3ab7a81108a6772f424" + }, + "source_tar_sha256": "9847e8924ba7a6446aaa3ce735f4dc2849046c37e21aeb8867f3154ab163ede0" + }, + "training_config_sha256": "d4d71e6e8f2e38ba3b1eaa0499bf1390c6e98c929e77d0aeaecc91896e2bf664", + "dataset_manifest_sha256": "5e574c5784420866052fc87140692550c11921af07dd1541064e46d94c499908" +} diff --git a/zero-latency/walker2d/vla/starvla-qwengr00t-h1/walker2d_l0_gr00t_h1_5fps_2h200_20260917/task_contract.json b/zero-latency/walker2d/vla/starvla-qwengr00t-h1/walker2d_l0_gr00t_h1_5fps_2h200_20260917/task_contract.json new file mode 100644 index 0000000000000000000000000000000000000000..7567af8871901a6d8a45081d77c7b0ca38924e72 --- /dev/null +++ b/zero-latency/walker2d/vla/starvla-qwengr00t-h1/walker2d_l0_gr00t_h1_5fps_2h200_20260917/task_contract.json @@ -0,0 +1,80 @@ +{ + "action_space": { + "dtype": "float32", + "high": [ + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0 + ], + "labels": [ + "right_thigh_torque", + "right_leg_torque", + "right_foot_torque", + "left_thigh_torque", + "left_leg_torque", + "left_foot_torque" + ], + "low": [ + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0 + ], + "type": "box" + }, + "base_prompt": "Move the Walker2d robot forward while keeping its torso upright. Predict six continuous torques in [-1, 1] ordered as right thigh, right leg, right foot, left thigh, left leg, and left foot.", + "env_fps": 5, + "env_id": "LatencyBench/Walker2dContinuous-v0", + "frame_stack": 1, + "make_kwargs": { + "base_env_id": "Walker2d-v4", + "base_make_kwargs": { + "ctrl_cost_weight": 0.001, + "exclude_current_positions_from_observation": true, + "forward_reward_weight": 1.0, + "healthy_reward": 1.0, + "reset_noise_scale": 0.005, + "terminate_when_unhealthy": true + }, + "render_mode": "rgb_array" + }, + "noop_action": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ], + "obs_fps": 5, + "registration_imports": [ + "latency_bench.envs.gymnasium_walker2d" + ], + "state_space": { + "labels": [ + "torso_height", + "torso_angle", + "right_thigh_angle", + "right_leg_angle", + "right_foot_angle", + "left_thigh_angle", + "left_leg_angle", + "left_foot_angle", + "torso_x_velocity", + "torso_z_velocity", + "torso_angular_velocity", + "right_thigh_angular_velocity", + "right_leg_angular_velocity", + "right_foot_angular_velocity", + "left_thigh_angular_velocity", + "left_leg_angular_velocity", + "left_foot_angular_velocity" + ] + }, + "task_name": "walker2d_rgb_state" +} diff --git a/zero-latency/walker2d/vla/starvla-qwengr00t-h1/walker2d_l0_gr00t_h1_5fps_2h200_20260917/validation/first_completed_reload_episode.json b/zero-latency/walker2d/vla/starvla-qwengr00t-h1/walker2d_l0_gr00t_h1_5fps_2h200_20260917/validation/first_completed_reload_episode.json new file mode 100644 index 0000000000000000000000000000000000000000..19e9bd8366793a2da347ad65fe06f34f3846a6c7 --- /dev/null +++ b/zero-latency/walker2d/vla/starvla-qwengr00t-h1/walker2d_l0_gr00t_h1_5fps_2h200_20260917/validation/first_completed_reload_episode.json @@ -0,0 +1,15 @@ +{ + "evidence": "existingL0sim20log", + "completed_episodes": 7, + "checkpoint": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/zero_latency/bundle/checkpoints/model.pt", + "completed_episode_log_lines": [ + "[bench] sweep 1/1 episode 1/20 done", + "[bench] sweep 1/1 episode 2/20 done", + "[bench] sweep 1/1 episode 3/20 done", + "[bench] sweep 1/1 episode 4/20 done", + "[bench] sweep 1/1 episode 5/20 done", + "[bench] sweep 1/1 episode 6/20 done", + "[bench] sweep 1/1 episode 7/20 done" + ], + "evaluation": "running; bufferedtracefilesnotyetflushed; nofinalscores" +} diff --git a/zero-latency/walker2d/vla/starvla-qwengr00t-h1/walker2d_l0_gr00t_h1_5fps_2h200_20260917/validation/formal.status.json b/zero-latency/walker2d/vla/starvla-qwengr00t-h1/walker2d_l0_gr00t_h1_5fps_2h200_20260917/validation/formal.status.json new file mode 100644 index 0000000000000000000000000000000000000000..ef75c4cd2020e4a82b917fd0ed186fddf0950765 --- /dev/null +++ b/zero-latency/walker2d/vla/starvla-qwengr00t-h1/walker2d_l0_gr00t_h1_5fps_2h200_20260917/validation/formal.status.json @@ -0,0 +1,25 @@ +{ + "task": "walker2d", + "stage": "formal", + "state": "COMPLETED", + "started_at": "2026-09-17T12:32:39.124476+00:00", + "command": [ + "taskset", + "-c", + "28-55,140-167", + "/home/ubuntu/miniconda3/envs/starvla_rl_games_gr00t/bin/python", + "-m", + "accelerate.commands.launch", + "--config_file", + "/mnt/local/lzj/latency-sensitive-bench/code/h1-5fps-20260917-v2/third_party/starVLA/starVLA/config/deepseeds/deepspeed_zero2.yaml", + "--num_processes", + "2", + "--main_process_port", + "29792", + "starVLA/training/train_starvla.py", + "--config_yaml", + "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/zero_latency/train.yaml" + ], + "exit_code": 0, + "ended_at": "2026-09-17T14:19:42.948167+00:00" +} diff --git a/zero-latency/walker2d/vla/starvla-qwengr00t-h1/walker2d_l0_gr00t_h1_5fps_2h200_20260917/validation/l0_sim20.yaml b/zero-latency/walker2d/vla/starvla-qwengr00t-h1/walker2d_l0_gr00t_h1_5fps_2h200_20260917/validation/l0_sim20.yaml new file mode 100644 index 0000000000000000000000000000000000000000..6a76b63f577ea32bb5e3a4106782c3ada18759a8 --- /dev/null +++ b/zero-latency/walker2d/vla/starvla-qwengr00t-h1/walker2d_l0_gr00t_h1_5fps_2h200_20260917/validation/l0_sim20.yaml @@ -0,0 +1,161 @@ +experiment: + name: gr00t_walker2d_zero + seed: 42 +backend: + type: sample_factory + algo: APPO + device: cuda + train_dir: /mnt/checkpoints/latency-sensitive-bench/small_models/walker2d + restart_behavior: overwrite + run_mode: eval +executor: + mode: simulated + inference_devices: + - cuda:0 + inference_batch_size: 1 + simulated_inference_pool: true + simulated_worker_capacity: 1 +env: + name: gymnasium + task_name: walker2d + env_id: LatencyBench/Walker2dContinuous-v0 + registration_imports: + - latency_bench.envs.gymnasium_walker2d + make_kwargs: + base_env_id: Walker2d-v4 + base_make_kwargs: + ctrl_cost_weight: 0.001 + exclude_current_positions_from_observation: true + forward_reward_weight: 1.0 + healthy_reward: 1.0 + reset_noise_scale: 0.005 + terminate_when_unhealthy: true + render_mode: rgb_array + env_fps: 5 + obs_fps: 5 + frame_stack: 1 + action_space: + dtype: float32 + high: + - 1.0 + - 1.0 + - 1.0 + - 1.0 + - 1.0 + - 1.0 + labels: + - right_thigh_torque + - right_leg_torque + - right_foot_torque + - left_thigh_torque + - left_leg_torque + - left_foot_torque + low: + - -1.0 + - -1.0 + - -1.0 + - -1.0 + - -1.0 + - -1.0 + type: box + noop_action: + - 0.0 + - 0.0 + - 0.0 + - 0.0 + - 0.0 + - 0.0 + base_prompt: Move the Walker2d robot forward while keeping its torso upright. Predict + six continuous torques in [-1, 1] ordered as right thigh, right leg, right foot, + left thigh, left leg, and left foot. + state_space: + labels: + - torso_height + - torso_angle + - right_thigh_angle + - right_leg_angle + - right_foot_angle + - left_thigh_angle + - left_leg_angle + - left_foot_angle + - torso_x_velocity + - torso_z_velocity + - torso_angular_velocity + - right_thigh_angular_velocity + - right_leg_angular_velocity + - right_foot_angular_velocity + - left_thigh_angular_velocity + - left_leg_angular_velocity + - left_foot_angular_velocity + obs_resize: + - 224 + - 224 +latency: + method: zero + sync_cuda: false +scheduler: + hold_policy: hold + ordering_policy: issue_order_fifo +policy: + type: starvla + checkpoint_path: /mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/zero_latency/bundle/checkpoints/model.pt + model_config_path: /mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/zero_latency/bundle/config.yaml + backbone_path: /mnt/local/lzj/latency-sensitive-bench/models/Qwen3-VL-4B-Instruct + device: cuda:0 + task_manifest_path: /mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/zero_latency/bundle/manifest.json + latency_prompt_map_path: /mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/zero_latency/bundle/latency_prompt_map.json + latency_prompt_key: 0 +training: + train_for_env_steps: 10000000 + num_workers: 8 + num_envs_per_worker: 8 + worker_num_splits: 2 + num_policies: 1 + batch_size: 1024 + rollout: 64 + recurrence: 1 + num_epochs: 2 + num_batches_per_epoch: 4 + max_policy_lag: 300 + learning_rate: 0.00295 + gamma: 0.99 + gae_lambda: 0.95 + ppo_clip_ratio: 0.2 + ppo_clip_value: 0.2 + value_loss_coeff: 1.3 + max_grad_norm: 3.5 + exploration_loss_coeff: 0.0 + async_rl: false + batched_sampling: false + use_rnn: false + encoder_mlp_layers: + - 64 + - 64 + normalize_input: true + normalize_returns: true + save_every_sec: 600 + keep_checkpoints: 5 + lr_schedule: linear_decay + kl_loss_coeff: 0.1 + serial_mode: false + nonlinearity: tanh + adaptive_stddev: false + policy_initialization: torch_default + initial_stddev: 1.0 + shuffle_minibatches: false + value_bootstrap: true +evaluation: + eval_episodes: 20 + eval_parallel_envs: 1 + eval_max_steps: 1000 + eval_deterministic: true + eval_latency_values: null +logging: + output_dir: /mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/zero_latency/l0_sim20 + video: + enabled: false + num_bins: 1 + save_step_records: true + save_action_records: true + save_latency_records: true + realtime_pipeline_profile: false diff --git a/zero-latency/walker2d/vla/starvla-qwengr00t-h1/walker2d_l0_gr00t_h1_5fps_2h200_20260917/validation/loader_check.json b/zero-latency/walker2d/vla/starvla-qwengr00t-h1/walker2d_l0_gr00t_h1_5fps_2h200_20260917/validation/loader_check.json new file mode 100644 index 0000000000000000000000000000000000000000..a7654eac51342e16210ff75ad7537f0cf61f5a5c --- /dev/null +++ b/zero-latency/walker2d/vla/starvla-qwengr00t-h1/walker2d_l0_gr00t_h1_5fps_2h200_20260917/validation/loader_check.json @@ -0,0 +1,27 @@ +{ + "loader_action_shape": [ + 1, + 6 + ], + "loader_state_shape": [ + 1, + 17 + ], + "native_action_roundtrip": true, + "state_minmax_once": true, + "image_pixels_identical": true, + "robot_type": "rl_games_gymnasium", + "sample_action": [ + [ + 0.8583984375, + -1.0, + 1.0, + 1.0, + -1.0, + 1.0 + ] + ], + "native_decoder": "box identity slice; unnorm_key unused", + "action_dtype": "float16", + "loader_float16_rounding_max_abs": 0.00012290477752685547 +} diff --git a/zero-latency/walker2d/vla/starvla-qwengr00t-h1/walker2d_l0_gr00t_h1_5fps_2h200_20260917/validation/preparation.json b/zero-latency/walker2d/vla/starvla-qwengr00t-h1/walker2d_l0_gr00t_h1_5fps_2h200_20260917/validation/preparation.json new file mode 100644 index 0000000000000000000000000000000000000000..b7359d6a26051030fe60af6ea06832397fc6b9c7 --- /dev/null +++ b/zero-latency/walker2d/vla/starvla-qwengr00t-h1/walker2d_l0_gr00t_h1_5fps_2h200_20260917/validation/preparation.json @@ -0,0 +1,42 @@ +{ + "repo": "latency-sensitive-bench/Standard-Pipeline", + "revision": "571cb5801720488ed6458b9cfe3f4c595b069cea", + "prefix": "walker2d/zero_latency/shared/source_100ep_v1/demonstrations/raw", + "sha256": { + "metadata.json": "c8ea2669493beabd97d9e7ecd06cc443adcb1760d8e71dd03588ea4adea20ae9", + "train.parquet": "6a105846e7def09f10fd2a985413cc9a04b9433f57519d330ce33a1033ba7e65", + "val.parquet": "687c1a5d08b3641b5ae76d218dd6afe483f8873b057aa3ab7a81108a6772f424" + }, + "source_tar_sha256": "9847e8924ba7a6446aaa3ce735f4dc2849046c37e21aeb8867f3154ab163ede0", + "frames": { + "train": 85160, + "val": 9664 + }, + "episodes": { + "train": 90, + "val": 10 + }, + "return_ranges": { + "train": [ + 3019.899699331727, + 4524.623699762858 + ], + "val": [ + 3668.9183055646718, + 4558.868211328983 + ] + }, + "image_size": [ + 224, + 224 + ], + "state_dim": 17, + "action_dim": 6, + "action_horizon": 1, + "fps": 5, + "all_source_returns_gt3000": true, + "all_nonprompt_columns_unchanged": true, + "normalization": "train-only min_max; same transform in validation", + "train_config": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/zero_latency/train.yaml", + "dataset": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/zero_latency/data/lerobot/walker2d_h1_5fps_l0" +} diff --git a/zero-latency/walker2d/vla/starvla-qwengr00t-h1/walker2d_l0_gr00t_h1_5fps_2h200_20260917/validation/publication_state.json b/zero-latency/walker2d/vla/starvla-qwengr00t-h1/walker2d_l0_gr00t_h1_5fps_2h200_20260917/validation/publication_state.json new file mode 100644 index 0000000000000000000000000000000000000000..a485a2ff422efd4f051f1445cf2199ab43077a5e --- /dev/null +++ b/zero-latency/walker2d/vla/starvla-qwengr00t-h1/walker2d_l0_gr00t_h1_5fps_2h200_20260917/validation/publication_state.json @@ -0,0 +1,32 @@ +{ + "task": "walker2d", + "verified_at": "2026-09-17T14:30:05.544543+00:00", + "training": "COMPLETED5000exit0", + "l0_eval_status_snapshot": { + "task": "walker2d", + "stage": "l0_sim20", + "state": "RUNNING", + "started_at": "2026-09-17T14:19:53.991430+00:00", + "command": [ + "/home/ubuntu/miniconda3/envs/starvla_rl_games_gr00t/bin/python", + "-m", + "latency_bench.run", + "--config", + "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/walker2d/zero_latency/l0_sim20.yaml" + ] + }, + "l0_episodes_completed_at_snapshot": 7, + "l0_final_scores": "notreportedwhile20episodeevaluationpending", + "reload_forward_evidence": "firstcompletedepisodefromexistingfinalbundleL0evaluation;noextraevaluation", + "downstream": "P/B/Fpending;parentmanaged3090queue", + "bundle_files": 8, + "bundle_sha256": { + "checkpoints/model.pt": "5df7bf29e8cc0bc984cb21ad4f8db673f16b17ceb438f4f4c91be65ee52c4de1", + "config.full.yaml": "23a1f179abba3d28587f14b89f99cfbaca0634aa36ab7c78539ee0ed95e0d58a", + "config.yaml": "453c0e1175ed33e703476ffbc07b5d3a17e0d89f65d3038d53c8bc6436b537a7", + "dataset_statistics.json": "1956e498026db81002e20cb62543ee44c55a82b4bf1a10dce3517483de0ecd83", + "latency_prompt_map.json": "03688d51cea9f0ad31aee9666aeb02a58be4099aee9f083271eceaa9139d2871", + "manifest.json": "5e574c5784420866052fc87140692550c11921af07dd1541064e46d94c499908", + "provenance.json": "3e6427330cd062bd82f60e385c0bb099dba1fa339abaa0c108ae349ea936115c" + } +} diff --git a/zero-latency/walker2d/vla/starvla-qwenoft-h1/walker2d_rgb_state_l0_return_gt3000_100ep_openvla_native_continuous_projector_sft_5k/README.md b/zero-latency/walker2d/vla/starvla-qwenoft-h1/walker2d_rgb_state_l0_return_gt3000_100ep_openvla_native_continuous_projector_sft_5k/README.md new file mode 100644 index 0000000000000000000000000000000000000000..847a773a3e96c7548dbd3ed16faf47dfc6934544 --- /dev/null +++ b/zero-latency/walker2d/vla/starvla-qwenoft-h1/walker2d_rgb_state_l0_return_gt3000_100ep_openvla_native_continuous_projector_sft_5k/README.md @@ -0,0 +1,30 @@ +# walker2d / qwenoft + +Training condition: `zero-latency`. Run: `walker2d_rgb_state_l0_return_gt3000_100ep_openvla_native_continuous_projector_sft_5k`. + +Copied from [Standard-Pipeline](https://huggingface.co/latency-sensitive-bench/Standard-Pipeline/tree/2208875f92b21e3f8ec2363a30a6705dfe56cc04/walker2d/zero_latency/OpenVLA) at `2208875f92b21e3f8ec2363a30a6705dfe56cc04`. + +- Checkpoint: `checkpoints/model.pt` +- Selection: only published VLA checkpoint +- Checkpoint SHA256: `8287406d754a46cf89c798a807bff000c9dc0bf93a6b7c49c51e74928599f259` +- Configuration, FPS, actions, normalization and latency semantics retain their source values. +- This import does not run evaluations or change training/evaluation/acceptance status. + +Action horizon is 1. Use `config.yaml` and `dataset_statistics.json` with the matching +prepared backbone. `task_contract.json` is extracted from the original training configuration. +Original machine paths in historical configurations are provenance: bind checkpoint, +backbone, manifest and profile paths to downloaded local files when running inference. + +Missing in this source model bundle: `manifest.json`, `latency_prompt_map.json`. +These files were not substituted with files from another training condition. + +## Download + +```python +from huggingface_hub import snapshot_download +prefix = "zero-latency/walker2d/vla/starvla-qwenoft-h1/walker2d_rgb_state_l0_return_gt3000_100ep_openvla_native_continuous_projector_sft_5k" +snapshot_download("latency-sensitive-bench/benchmark-models", revision="", + allow_patterns=[prefix + "/**"]) +``` + +Use the verified import commit listed in the repository's import report. diff --git a/zero-latency/walker2d/vla/starvla-qwenoft-h1/walker2d_rgb_state_l0_return_gt3000_100ep_openvla_native_continuous_projector_sft_5k/checkpoints/model.pt b/zero-latency/walker2d/vla/starvla-qwenoft-h1/walker2d_rgb_state_l0_return_gt3000_100ep_openvla_native_continuous_projector_sft_5k/checkpoints/model.pt new file mode 100644 index 0000000000000000000000000000000000000000..32104598f6dc545ff38d346d1addb8b27f75d75b --- /dev/null +++ b/zero-latency/walker2d/vla/starvla-qwenoft-h1/walker2d_rgb_state_l0_return_gt3000_100ep_openvla_native_continuous_projector_sft_5k/checkpoints/model.pt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8287406d754a46cf89c798a807bff000c9dc0bf93a6b7c49c51e74928599f259 +size 9785142689 diff --git a/zero-latency/walker2d/vla/starvla-qwenoft-h1/walker2d_rgb_state_l0_return_gt3000_100ep_openvla_native_continuous_projector_sft_5k/config.full.yaml b/zero-latency/walker2d/vla/starvla-qwenoft-h1/walker2d_rgb_state_l0_return_gt3000_100ep_openvla_native_continuous_projector_sft_5k/config.full.yaml new file mode 100644 index 0000000000000000000000000000000000000000..a329004ac43ab9a027bbaadd3a73ed490564da2d --- /dev/null +++ b/zero-latency/walker2d/vla/starvla-qwenoft-h1/walker2d_rgb_state_l0_return_gt3000_100ep_openvla_native_continuous_projector_sft_5k/config.full.yaml @@ -0,0 +1,394 @@ +framework: + name: QwenOFT + qwenvl: + base_vlm: /mnt/checkpoints/latency-sensitive-bench/openvla/Qwen3-VL-4B-Instruct + attn_implementation: flash_attention_2 + enable_gradient_checkpointing: true + action_model: + action_model_type: MLP + action_dim: 6 + action_hidden_dim: 2560 + future_action_window_size: 0 + past_action_window_size: 0 + loss_type: l1 + state_encoding: continuous_projector + state_dim: 17 + action_horizon: 1 + action_env_dim: 6 + kv_memory: + enabled: false + window: 4 + rollout_len: 8 + packed_train: false + rebased_sink: true +datasets: + vla_data: + dataset_py: lerobot_datasets + include_state: true + data_root_dir: /mnt/data/latency-sensitive-bench/vla_datasets/rl_games + data_mix: walker2d_rgb_state_l0_return_gt3000_100ep + eval_data_mix: walker2d_rgb_state_l0_return_gt3000_100ep__val + custom_mixtures_path: /mnt/data/latency-sensitive-bench/vla_datasets/rl_games/_generated_mixtures/walker2d_rgb_state_l0_return_gt3000_100ep.json + action_type: discrete + sequential_step_sampling: false + eval_sequential_step_sampling: null + num_workers: 8 + eval_num_workers: 8 + prefetch_factor: 4 + persistent_workers: true + pin_memory: true + shuffle: true + action_balance: + enabled: false + strategy: balanced_epoch + action_key: action_id + target_flap_fraction: 0.3 + noop_id: 0 + flap_id: 1 + latency_curriculum: + enabled: false + strategy: exclusive + latencies: null + phase_steps: null + phase_distributions: null + new_latency_passes: 1.0 + replay_passes: 0.25 + target_total_passes: 2.0 + final_equalization: true + step_budget_mode: auto + eval_at_phase_end: false + save_at_phase_end: false + computed_plan: null + per_device_batch_size: 16 + load_all_data_for_training: true + num_obs_frames: 1 + image_mode: single + prompt_mode: raw + stitch_grid: + - 2 + - 2 + obs_image_size: null + video_backend: torchvision_av + gymnasium_task_contract: + action_space: + dtype: float32 + high: + - 1.0 + - 1.0 + - 1.0 + - 1.0 + - 1.0 + - 1.0 + labels: + - right_thigh_torque + - right_leg_torque + - right_foot_torque + - left_thigh_torque + - left_leg_torque + - left_foot_torque + low: + - -1.0 + - -1.0 + - -1.0 + - -1.0 + - -1.0 + - -1.0 + type: box + base_prompt: Move the Walker2d robot forward while keeping its torso upright. + Predict six continuous torques in [-1, 1] ordered as right thigh, right leg, + right foot, left thigh, left leg, and left foot. + env_fps: 125.0 + env_id: LatencyBench/Walker2dContinuous-v0 + frame_stack: 1 + make_kwargs: + base_env_id: Walker2d-v4 + base_make_kwargs: + ctrl_cost_weight: 0.001 + exclude_current_positions_from_observation: true + forward_reward_weight: 1.0 + healthy_reward: 1.0 + reset_noise_scale: 0.005 + terminate_when_unhealthy: true + render_mode: rgb_array + noop_action: + - 0.0 + - 0.0 + - 0.0 + - 0.0 + - 0.0 + - 0.0 + obs_fps: 125.0 + registration_imports: + - latency_bench.envs.gymnasium_walker2d + state_space: + labels: + - torso_height + - torso_angle + - right_thigh_angle + - right_leg_angle + - right_foot_angle + - left_thigh_angle + - left_leg_angle + - left_foot_angle + - torso_x_velocity + - torso_z_velocity + - torso_angular_velocity + - right_thigh_angular_velocity + - right_leg_angular_velocity + - right_foot_angular_velocity + - left_thigh_angular_velocity + - left_leg_angular_velocity + - left_foot_angular_velocity + task_name: walker2d_rgb_state + active_action_dim: 6 +dataset: + source_hf: '' + config_name: null + source_subdir: null + converted_name: walker2d_rgb_state_l0_return_gt3000_100ep + single_source_hf: '' + mixed_source_hf: '' + single_converted_name: walker2d_rgb_state_l0_return_gt3000_100ep + mixed_converted_name: walker2d_rgb_state_l0_return_gt3000_100ep + single_latency_filter: null + mixed_latency_filter: null + force_download: false + setup_force: false + skip_verification: false + target_latency_unit: observation_steps + verify_rows: 200 + max_episodes: null + episodes_per_latency: null + latency_filter: null + debug_subset: + enabled: false + max_episodes: 5 + suffix: debug +base_model: + repo_id: Qwen/Qwen3-VL-4B-Instruct +initialization: + checkpoint_local_dir: null + checkpoint_hf_repo_id: null + checkpoint_filename: null +trainer: + max_train_steps: 5000 + num_warmup_steps: 100 + save_interval: 500 + eval_interval: 500 + eval_num_batches: 200 + per_latency_eval_num_batches: null + eval_action_classification: false + eval_action_classification_interval: null + cc_f1_tolerance: 1 + learning_rate: + base: 2.0e-05 + qwen_vl_interface: 1.0e-05 + action_model: 0.0001 + lr_scheduler_type: cosine_with_min_lr + scheduler_specific_kwargs: + min_lr: 1.0e-06 + freeze_modules: '' + freeze_vit: false + freeze_tied_embedding: false + freeze_llm_layers: [] + loss_scale: + vla: 1.0 + vlm: 0.1 + max_grad_norm: 1.0 + weight_decay: 0.0 + logging_frequency: 1 + profile_timing: + enabled: true + log_interval: 10 + gradient_clipping: 1.0 + gradient_accumulation_steps: 1 + distributed_backend: deepspeed + is_resume: false + pretrained_checkpoint: null + resume_step: 0 + reload_modules: null + optimizer: + name: AdamW + betas: + - 0.9 + - 0.95 + eps: 1.0e-08 + weight_decay: 1.0e-08 + fused: true + save_format: pt +workspace_dir: /home/lzj/code/latency-sensitive-bench +run_root_dir: /mnt/checkpoints/latency-sensitive-bench/openvla_runs +seed: 42 +wandb_entity: dongqianyu99-zhejiang-university +wandb_project: latency-sensitive-bench +auth: + env_file: null + hf_token_env: HF_TOKEN + wandb_api_key_env: WANDB_API_KEY +paths: + run_root_dir: /mnt/checkpoints/latency-sensitive-bench/openvla_runs + dataset_local_dir: /mnt/data/latency-sensitive-bench/vla_datasets/rl_games + dataset_cache_dir: null + base_model_dir: /mnt/checkpoints/latency-sensitive-bench/openvla/Qwen3-VL-4B-Instruct + accelerate_config: starVLA/config/deepseeds/deepspeed_zero2_cpu_offload.yaml +launch: + use_accelerate: true + gpus: null + num_processes: 1 + dry_run: false +conda: + enabled: true + env_name: null +rl_games: + model_alias: openvla + env_eval: + image_size: 224 + frameskip: 1 + image_transform: raw_rgb + prompt_mode: raw + ghost_trail: + history_frames: 5 + gamma: 1.3 + min_alpha: 35 + scroll_px_per_step: 4.0 + ground_fraction: 0.22 + seed: 42 + fixed_episode_seeds: true + latency_seed_stride: 0 + task_seed_stride: 0 + task_description: '' + eval_parallel_envs: 5 + action_chunk_execution: + enabled: false + chunk_size: null + enabled: false + gymnasium: + make_kwargs: {} + mid_train: + enabled: false + interval_steps: 100 + latencies: + - 0 + num_episodes: 5 + max_steps_per_episode: 3600 + post_train: + enabled: false + latencies: + - 0 + - 1 + - 2 + - 3 + - 4 + num_episodes: 5 + max_steps_per_episode: 3600 + eval_backend: latency_bench + distributed_mode: none + vectorized: + enabled: false + batch_size: 1 + latency: + prompt_map_path: null + mode: single + values: + - 0 + task: gymnasium + gymnasium: + task_contract: + action_space: + dtype: float32 + high: + - 1.0 + - 1.0 + - 1.0 + - 1.0 + - 1.0 + - 1.0 + labels: + - right_thigh_torque + - right_leg_torque + - right_foot_torque + - left_thigh_torque + - left_leg_torque + - left_foot_torque + low: + - -1.0 + - -1.0 + - -1.0 + - -1.0 + - -1.0 + - -1.0 + type: box + base_prompt: Move the Walker2d robot forward while keeping its torso upright. + Predict six continuous torques in [-1, 1] ordered as right thigh, right leg, + right foot, left thigh, left leg, and left foot. + env_fps: 125.0 + env_id: LatencyBench/Walker2dContinuous-v0 + frame_stack: 1 + make_kwargs: + base_env_id: Walker2d-v4 + base_make_kwargs: + ctrl_cost_weight: 0.001 + exclude_current_positions_from_observation: true + forward_reward_weight: 1.0 + healthy_reward: 1.0 + reset_noise_scale: 0.005 + terminate_when_unhealthy: true + render_mode: rgb_array + noop_action: + - 0.0 + - 0.0 + - 0.0 + - 0.0 + - 0.0 + - 0.0 + obs_fps: 125.0 + registration_imports: + - latency_bench.envs.gymnasium_walker2d + state_space: + labels: + - torso_height + - torso_angle + - right_thigh_angle + - right_leg_angle + - right_foot_angle + - left_thigh_angle + - left_leg_angle + - left_foot_angle + - torso_x_velocity + - torso_z_velocity + - torso_angular_velocity + - right_thigh_angular_velocity + - right_leg_angular_velocity + - right_foot_angular_velocity + - left_thigh_angular_velocity + - left_leg_angular_velocity + - left_foot_angular_velocity + task_name: walker2d_rgb_state + task_name: walker2d_rgb_state + initialization_mode: scratch + action_carrier: native +model: openvla +env: gymnasium +init: scratch +mode: single +checkpoint: + load: none + hf_repo_id: null + save_best_model: false + save_final_model: true + save_pt_file: true + save_training_state: false + save_safetensors_file: false + local: + keep_last_n: 1 + sync: + enabled: false + repo_id: null + keep_last_n: 0 + sync_every_n_checkpoints: 1 + resume_policy: local_latest +run_id: walker2d_rgb_state_l0_return_gt3000_100ep_openvla_native_continuous_projector_sft_5k +output_dir: /mnt/checkpoints/latency-sensitive-bench/openvla_runs/walker2d_rgb_state_l0_return_gt3000_100ep_openvla_native_continuous_projector_sft_5k +config_yaml: null +is_debug: false +version_id: '0.21' diff --git a/zero-latency/walker2d/vla/starvla-qwenoft-h1/walker2d_rgb_state_l0_return_gt3000_100ep_openvla_native_continuous_projector_sft_5k/config.yaml b/zero-latency/walker2d/vla/starvla-qwenoft-h1/walker2d_rgb_state_l0_return_gt3000_100ep_openvla_native_continuous_projector_sft_5k/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..a329004ac43ab9a027bbaadd3a73ed490564da2d --- /dev/null +++ b/zero-latency/walker2d/vla/starvla-qwenoft-h1/walker2d_rgb_state_l0_return_gt3000_100ep_openvla_native_continuous_projector_sft_5k/config.yaml @@ -0,0 +1,394 @@ +framework: + name: QwenOFT + qwenvl: + base_vlm: /mnt/checkpoints/latency-sensitive-bench/openvla/Qwen3-VL-4B-Instruct + attn_implementation: flash_attention_2 + enable_gradient_checkpointing: true + action_model: + action_model_type: MLP + action_dim: 6 + action_hidden_dim: 2560 + future_action_window_size: 0 + past_action_window_size: 0 + loss_type: l1 + state_encoding: continuous_projector + state_dim: 17 + action_horizon: 1 + action_env_dim: 6 + kv_memory: + enabled: false + window: 4 + rollout_len: 8 + packed_train: false + rebased_sink: true +datasets: + vla_data: + dataset_py: lerobot_datasets + include_state: true + data_root_dir: /mnt/data/latency-sensitive-bench/vla_datasets/rl_games + data_mix: walker2d_rgb_state_l0_return_gt3000_100ep + eval_data_mix: walker2d_rgb_state_l0_return_gt3000_100ep__val + custom_mixtures_path: /mnt/data/latency-sensitive-bench/vla_datasets/rl_games/_generated_mixtures/walker2d_rgb_state_l0_return_gt3000_100ep.json + action_type: discrete + sequential_step_sampling: false + eval_sequential_step_sampling: null + num_workers: 8 + eval_num_workers: 8 + prefetch_factor: 4 + persistent_workers: true + pin_memory: true + shuffle: true + action_balance: + enabled: false + strategy: balanced_epoch + action_key: action_id + target_flap_fraction: 0.3 + noop_id: 0 + flap_id: 1 + latency_curriculum: + enabled: false + strategy: exclusive + latencies: null + phase_steps: null + phase_distributions: null + new_latency_passes: 1.0 + replay_passes: 0.25 + target_total_passes: 2.0 + final_equalization: true + step_budget_mode: auto + eval_at_phase_end: false + save_at_phase_end: false + computed_plan: null + per_device_batch_size: 16 + load_all_data_for_training: true + num_obs_frames: 1 + image_mode: single + prompt_mode: raw + stitch_grid: + - 2 + - 2 + obs_image_size: null + video_backend: torchvision_av + gymnasium_task_contract: + action_space: + dtype: float32 + high: + - 1.0 + - 1.0 + - 1.0 + - 1.0 + - 1.0 + - 1.0 + labels: + - right_thigh_torque + - right_leg_torque + - right_foot_torque + - left_thigh_torque + - left_leg_torque + - left_foot_torque + low: + - -1.0 + - -1.0 + - -1.0 + - -1.0 + - -1.0 + - -1.0 + type: box + base_prompt: Move the Walker2d robot forward while keeping its torso upright. + Predict six continuous torques in [-1, 1] ordered as right thigh, right leg, + right foot, left thigh, left leg, and left foot. + env_fps: 125.0 + env_id: LatencyBench/Walker2dContinuous-v0 + frame_stack: 1 + make_kwargs: + base_env_id: Walker2d-v4 + base_make_kwargs: + ctrl_cost_weight: 0.001 + exclude_current_positions_from_observation: true + forward_reward_weight: 1.0 + healthy_reward: 1.0 + reset_noise_scale: 0.005 + terminate_when_unhealthy: true + render_mode: rgb_array + noop_action: + - 0.0 + - 0.0 + - 0.0 + - 0.0 + - 0.0 + - 0.0 + obs_fps: 125.0 + registration_imports: + - latency_bench.envs.gymnasium_walker2d + state_space: + labels: + - torso_height + - torso_angle + - right_thigh_angle + - right_leg_angle + - right_foot_angle + - left_thigh_angle + - left_leg_angle + - left_foot_angle + - torso_x_velocity + - torso_z_velocity + - torso_angular_velocity + - right_thigh_angular_velocity + - right_leg_angular_velocity + - right_foot_angular_velocity + - left_thigh_angular_velocity + - left_leg_angular_velocity + - left_foot_angular_velocity + task_name: walker2d_rgb_state + active_action_dim: 6 +dataset: + source_hf: '' + config_name: null + source_subdir: null + converted_name: walker2d_rgb_state_l0_return_gt3000_100ep + single_source_hf: '' + mixed_source_hf: '' + single_converted_name: walker2d_rgb_state_l0_return_gt3000_100ep + mixed_converted_name: walker2d_rgb_state_l0_return_gt3000_100ep + single_latency_filter: null + mixed_latency_filter: null + force_download: false + setup_force: false + skip_verification: false + target_latency_unit: observation_steps + verify_rows: 200 + max_episodes: null + episodes_per_latency: null + latency_filter: null + debug_subset: + enabled: false + max_episodes: 5 + suffix: debug +base_model: + repo_id: Qwen/Qwen3-VL-4B-Instruct +initialization: + checkpoint_local_dir: null + checkpoint_hf_repo_id: null + checkpoint_filename: null +trainer: + max_train_steps: 5000 + num_warmup_steps: 100 + save_interval: 500 + eval_interval: 500 + eval_num_batches: 200 + per_latency_eval_num_batches: null + eval_action_classification: false + eval_action_classification_interval: null + cc_f1_tolerance: 1 + learning_rate: + base: 2.0e-05 + qwen_vl_interface: 1.0e-05 + action_model: 0.0001 + lr_scheduler_type: cosine_with_min_lr + scheduler_specific_kwargs: + min_lr: 1.0e-06 + freeze_modules: '' + freeze_vit: false + freeze_tied_embedding: false + freeze_llm_layers: [] + loss_scale: + vla: 1.0 + vlm: 0.1 + max_grad_norm: 1.0 + weight_decay: 0.0 + logging_frequency: 1 + profile_timing: + enabled: true + log_interval: 10 + gradient_clipping: 1.0 + gradient_accumulation_steps: 1 + distributed_backend: deepspeed + is_resume: false + pretrained_checkpoint: null + resume_step: 0 + reload_modules: null + optimizer: + name: AdamW + betas: + - 0.9 + - 0.95 + eps: 1.0e-08 + weight_decay: 1.0e-08 + fused: true + save_format: pt +workspace_dir: /home/lzj/code/latency-sensitive-bench +run_root_dir: /mnt/checkpoints/latency-sensitive-bench/openvla_runs +seed: 42 +wandb_entity: dongqianyu99-zhejiang-university +wandb_project: latency-sensitive-bench +auth: + env_file: null + hf_token_env: HF_TOKEN + wandb_api_key_env: WANDB_API_KEY +paths: + run_root_dir: /mnt/checkpoints/latency-sensitive-bench/openvla_runs + dataset_local_dir: /mnt/data/latency-sensitive-bench/vla_datasets/rl_games + dataset_cache_dir: null + base_model_dir: /mnt/checkpoints/latency-sensitive-bench/openvla/Qwen3-VL-4B-Instruct + accelerate_config: starVLA/config/deepseeds/deepspeed_zero2_cpu_offload.yaml +launch: + use_accelerate: true + gpus: null + num_processes: 1 + dry_run: false +conda: + enabled: true + env_name: null +rl_games: + model_alias: openvla + env_eval: + image_size: 224 + frameskip: 1 + image_transform: raw_rgb + prompt_mode: raw + ghost_trail: + history_frames: 5 + gamma: 1.3 + min_alpha: 35 + scroll_px_per_step: 4.0 + ground_fraction: 0.22 + seed: 42 + fixed_episode_seeds: true + latency_seed_stride: 0 + task_seed_stride: 0 + task_description: '' + eval_parallel_envs: 5 + action_chunk_execution: + enabled: false + chunk_size: null + enabled: false + gymnasium: + make_kwargs: {} + mid_train: + enabled: false + interval_steps: 100 + latencies: + - 0 + num_episodes: 5 + max_steps_per_episode: 3600 + post_train: + enabled: false + latencies: + - 0 + - 1 + - 2 + - 3 + - 4 + num_episodes: 5 + max_steps_per_episode: 3600 + eval_backend: latency_bench + distributed_mode: none + vectorized: + enabled: false + batch_size: 1 + latency: + prompt_map_path: null + mode: single + values: + - 0 + task: gymnasium + gymnasium: + task_contract: + action_space: + dtype: float32 + high: + - 1.0 + - 1.0 + - 1.0 + - 1.0 + - 1.0 + - 1.0 + labels: + - right_thigh_torque + - right_leg_torque + - right_foot_torque + - left_thigh_torque + - left_leg_torque + - left_foot_torque + low: + - -1.0 + - -1.0 + - -1.0 + - -1.0 + - -1.0 + - -1.0 + type: box + base_prompt: Move the Walker2d robot forward while keeping its torso upright. + Predict six continuous torques in [-1, 1] ordered as right thigh, right leg, + right foot, left thigh, left leg, and left foot. + env_fps: 125.0 + env_id: LatencyBench/Walker2dContinuous-v0 + frame_stack: 1 + make_kwargs: + base_env_id: Walker2d-v4 + base_make_kwargs: + ctrl_cost_weight: 0.001 + exclude_current_positions_from_observation: true + forward_reward_weight: 1.0 + healthy_reward: 1.0 + reset_noise_scale: 0.005 + terminate_when_unhealthy: true + render_mode: rgb_array + noop_action: + - 0.0 + - 0.0 + - 0.0 + - 0.0 + - 0.0 + - 0.0 + obs_fps: 125.0 + registration_imports: + - latency_bench.envs.gymnasium_walker2d + state_space: + labels: + - torso_height + - torso_angle + - right_thigh_angle + - right_leg_angle + - right_foot_angle + - left_thigh_angle + - left_leg_angle + - left_foot_angle + - torso_x_velocity + - torso_z_velocity + - torso_angular_velocity + - right_thigh_angular_velocity + - right_leg_angular_velocity + - right_foot_angular_velocity + - left_thigh_angular_velocity + - left_leg_angular_velocity + - left_foot_angular_velocity + task_name: walker2d_rgb_state + task_name: walker2d_rgb_state + initialization_mode: scratch + action_carrier: native +model: openvla +env: gymnasium +init: scratch +mode: single +checkpoint: + load: none + hf_repo_id: null + save_best_model: false + save_final_model: true + save_pt_file: true + save_training_state: false + save_safetensors_file: false + local: + keep_last_n: 1 + sync: + enabled: false + repo_id: null + keep_last_n: 0 + sync_every_n_checkpoints: 1 + resume_policy: local_latest +run_id: walker2d_rgb_state_l0_return_gt3000_100ep_openvla_native_continuous_projector_sft_5k +output_dir: /mnt/checkpoints/latency-sensitive-bench/openvla_runs/walker2d_rgb_state_l0_return_gt3000_100ep_openvla_native_continuous_projector_sft_5k +config_yaml: null +is_debug: false +version_id: '0.21' diff --git a/zero-latency/walker2d/vla/starvla-qwenoft-h1/walker2d_rgb_state_l0_return_gt3000_100ep_openvla_native_continuous_projector_sft_5k/dataset_statistics.json b/zero-latency/walker2d/vla/starvla-qwenoft-h1/walker2d_rgb_state_l0_return_gt3000_100ep_openvla_native_continuous_projector_sft_5k/dataset_statistics.json new file mode 100644 index 0000000000000000000000000000000000000000..bfb9e72f511b15f1aba80c6eb1faa007fa88cee2 --- /dev/null +++ b/zero-latency/walker2d/vla/starvla-qwenoft-h1/walker2d_rgb_state_l0_return_gt3000_100ep_openvla_native_continuous_projector_sft_5k/dataset_statistics.json @@ -0,0 +1,180 @@ +{ + "new_embodiment": { + "action": { + "mean": [ + 0.8840610980987549, + 0.34449484944343567, + 0.11654752492904663, + 0.8480036854743958, + 0.32758569717407227, + 0.9875396490097046 + ], + "std": [ + 0.32312333583831787, + 0.7096099853515625, + 0.9048961400985718, + 0.343753844499588, + 0.7393367886543274, + 0.11781969666481018 + ], + "max": [ + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0 + ], + "min": [ + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0 + ], + "q01": [ + -0.5243898153305053, + -1.0, + -1.0, + -0.5588397586345673, + -1.0, + 0.5681399703025818 + ], + "q99": [ + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0 + ], + "mask": [ + true, + true, + true, + true, + true, + true + ] + }, + "state": { + "mean": [ + 0.38908126950263977, + -0.06854692101478577, + 0.5603042840957642, + 0.295283704996109, + 0.20391421020030975, + 0.7933022975921631, + 0.32835325598716736, + 0.7283809781074524, + -0.11274679005146027, + 0.36438047885894775, + -0.041566070169210434, + -0.006785172037780285, + -0.21467362344264984, + -0.2302139699459076, + -0.008182121440768242, + -0.19842907786369324, + -0.0016649218741804361 + ], + "std": [ + 0.15210415422916412, + 0.28636232018470764, + 0.08957907557487488, + 0.40512198209762573, + 0.5324621796607971, + 0.09768694639205933, + 0.4175085127353668, + 0.07296281307935677, + 0.3539130389690399, + 0.21250014007091522, + 0.5710971355438232, + 0.19414709508419037, + 0.797458827495575, + 0.7378174662590027, + 0.32647520303726196, + 0.7747778296470642, + 0.10512100160121918 + ], + "max": [ + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0 + ], + "min": [ + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0 + ], + "q01": [ + -0.01641636550426483, + -0.5725254046916962, + 0.1497414314746857, + -0.5082594507932663, + -0.704446622133255, + 0.4100250256061554, + -0.7682463544607162, + 0.7055095672607422, + -0.8300059294700622, + -0.09350170254707335, + -1.0, + -0.7184711527824402, + -1.0, + -1.0, + -0.9244865983724594, + -1.0, + -0.37168214678764344 + ], + "q99": [ + 0.6826163899898535, + 0.5332732021808625, + 0.7265087735652924, + 0.9429451179504394, + 0.963300715684891, + 0.9189580070972443, + 0.8997541296482087, + 0.7992216396331789, + 0.7250751173496247, + 0.9080468678474427, + 1.0, + 0.8595866250991845, + 1.0, + 1.0, + 1.0, + 1.0, + 0.04421793460845973 + ] + }, + "num_transitions": 85160, + "num_trajectories": 90 + } +} \ No newline at end of file diff --git a/zero-latency/walker2d/vla/starvla-qwenoft-h1/walker2d_rgb_state_l0_return_gt3000_100ep_openvla_native_continuous_projector_sft_5k/dataset_statistics_eval.json b/zero-latency/walker2d/vla/starvla-qwenoft-h1/walker2d_rgb_state_l0_return_gt3000_100ep_openvla_native_continuous_projector_sft_5k/dataset_statistics_eval.json new file mode 100644 index 0000000000000000000000000000000000000000..c2678c26ec551e184f0488da1e3ad5e067f3fef9 --- /dev/null +++ b/zero-latency/walker2d/vla/starvla-qwenoft-h1/walker2d_rgb_state_l0_return_gt3000_100ep_openvla_native_continuous_projector_sft_5k/dataset_statistics_eval.json @@ -0,0 +1,180 @@ +{ + "new_embodiment": { + "action": { + "mean": [ + 0.8849527835845947, + 0.35430335998535156, + 0.12333488464355469, + 0.8605653643608093, + 0.32735443115234375, + 0.9852222204208374 + ], + "std": [ + 0.32089999318122864, + 0.7054017186164856, + 0.907700777053833, + 0.32899200916290283, + 0.7426145076751709, + 0.1215449869632721 + ], + "max": [ + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0 + ], + "min": [ + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0 + ], + "q01": [ + -0.5130668139457703, + -1.0, + -1.0, + -0.5488155257701873, + -1.0, + 0.3306524598598479 + ], + "q99": [ + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0 + ], + "mask": [ + true, + true, + true, + true, + true, + true + ] + }, + "state": { + "mean": [ + 0.3954218029975891, + -0.0746895968914032, + 0.5602800250053406, + 0.3015284240245819, + 0.2082919180393219, + 0.7967728972434998, + 0.3465990126132965, + 0.7285168170928955, + -0.09946807473897934, + 0.36348482966423035, + -0.04188957437872887, + -0.006726500112563372, + -0.21870803833007812, + -0.23826205730438232, + -0.008270090445876122, + -0.2051077038049698, + -0.0017182693118229508 + ], + "std": [ + 0.15569402277469635, + 0.2810911238193512, + 0.09028489887714386, + 0.4023565948009491, + 0.5337862968444824, + 0.08760403096675873, + 0.40798017382621765, + 0.0689404234290119, + 0.3508071303367615, + 0.2143576443195343, + 0.5698263645172119, + 0.18698696792125702, + 0.800293505191803, + 0.7367855906486511, + 0.32409295439720154, + 0.768949031829834, + 0.11262209713459015 + ], + "max": [ + 0.8644862174987793, + 0.7157082557678223, + 0.9386438131332397, + 0.9675707817077637, + 0.9948835372924805, + 0.9897574186325073, + 0.987257719039917, + 0.9977864027023315, + 0.9504057168960571, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0 + ], + "min": [ + -0.9982708692550659, + -0.8229403495788574, + -0.6960538625717163, + -0.9217309951782227, + -0.877301812171936, + 0.2215665578842163, + -0.9920427799224854, + -0.9526035785675049, + -0.9228333234786987, + -0.8903961777687073, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0 + ], + "q01": [ + -0.017054073810577396, + -0.5626240783929825, + 0.15242856860160828, + -0.44991630494594576, + -0.6997699916362763, + 0.4606837844848633, + -0.7162600827217103, + 0.625782380104065, + -0.8147157776355743, + -0.07589246273040771, + -1.0, + -0.6913001000881195, + -1.0, + -1.0, + -0.9122443580627442, + -1.0, + -0.3875079840421677 + ], + "q99": [ + 0.6713557291030892, + 0.49713534116745084, + 0.7179951703548436, + 0.9427642738819123, + 0.9646597146987915, + 0.9173498451709751, + 0.8999581170082093, + 0.8004228246212006, + 0.7161488473415385, + 0.910248271226883, + 1.0, + 0.7649935555458134, + 1.0, + 1.0, + 1.0, + 1.0, + 0.10979756355286273 + ] + }, + "num_transitions": 9664, + "num_trajectories": 10 + } +} \ No newline at end of file diff --git a/zero-latency/walker2d/vla/starvla-qwenoft-h1/walker2d_rgb_state_l0_return_gt3000_100ep_openvla_native_continuous_projector_sft_5k/provenance.json b/zero-latency/walker2d/vla/starvla-qwenoft-h1/walker2d_rgb_state_l0_return_gt3000_100ep_openvla_native_continuous_projector_sft_5k/provenance.json new file mode 100644 index 0000000000000000000000000000000000000000..ec4735a201dce35ee1fd4215ed5f44d923e75845 --- /dev/null +++ b/zero-latency/walker2d/vla/starvla-qwenoft-h1/walker2d_rgb_state_l0_return_gt3000_100ep_openvla_native_continuous_projector_sft_5k/provenance.json @@ -0,0 +1,37 @@ +{ + "task": "walker2d", + "model": "qwenoft", + "training_condition": "zero-latency", + "training_run_id": "walker2d_rgb_state_l0_return_gt3000_100ep_openvla_native_continuous_projector_sft_5k", + "source": { + "repo_id": "latency-sensitive-bench/Standard-Pipeline", + "revision": "2208875f92b21e3f8ec2363a30a6705dfe56cc04", + "path": "walker2d/zero_latency/OpenVLA", + "url": "https://huggingface.co/latency-sensitive-bench/Standard-Pipeline/tree/2208875f92b21e3f8ec2363a30a6705dfe56cc04/walker2d/zero_latency/OpenVLA" + }, + "hf_repo_id": "latency-sensitive-bench/benchmark-models", + "hf_path": "zero-latency/walker2d/vla/starvla-qwenoft-h1/walker2d_rgb_state_l0_return_gt3000_100ep_openvla_native_continuous_projector_sft_5k", + "checkpoint": { + "source_file": "walker2d/zero_latency/OpenVLA/checkpoints/steps_5000_pytorch_model.pt", + "source_sha256": "8287406d754a46cf89c798a807bff000c9dc0bf93a6b7c49c51e74928599f259", + "source_bytes": 9785142689, + "file": "checkpoints/model.pt", + "selection_rule": "only published VLA checkpoint", + "method": "copy", + "sha256": "8287406d754a46cf89c798a807bff000c9dc0bf93a6b7c49c51e74928599f259", + "bytes": 9785142689 + }, + "config_source": "walker2d/zero_latency/OpenVLA/config.full.yaml", + "task_contract_file": "task_contract.json", + "missing_source_files": [ + "manifest.json", + "latency_prompt_map.json" + ], + "migration": { + "operation": "copy_selected_inference_bundle", + "source_modified": false, + "evaluation_performed": false, + "experiment_status": "preserved_in_source_evidence" + }, + "action_horizon": 1 +} diff --git a/zero-latency/walker2d/vla/starvla-qwenoft-h1/walker2d_rgb_state_l0_return_gt3000_100ep_openvla_native_continuous_projector_sft_5k/source/config.yaml b/zero-latency/walker2d/vla/starvla-qwenoft-h1/walker2d_rgb_state_l0_return_gt3000_100ep_openvla_native_continuous_projector_sft_5k/source/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..b9b94b2135195d605ba888ef8344adf7c7c0dd42 --- /dev/null +++ b/zero-latency/walker2d/vla/starvla-qwenoft-h1/walker2d_rgb_state_l0_return_gt3000_100ep_openvla_native_continuous_projector_sft_5k/source/config.yaml @@ -0,0 +1,86 @@ +checkpoint: + local: + keep_last_n: 1 + save_best_model: false + save_final_model: true + save_pt_file: true + save_safetensors_file: false + save_training_state: false + sync: + enabled: false + keep_last_n: 0 + repo_id: null +datasets: + vla_data: + data_mix: walker2d_rgb_state_l0_return_gt3000_100ep + dataset_py: lerobot_datasets + eval_data_mix: walker2d_rgb_state_l0_return_gt3000_100ep__val + latency_curriculum: + enabled: false + per_device_batch_size: 16 +framework: + action_model: + action_dim: 6 + action_env_dim: 6 + action_hidden_dim: 2560 + action_horizon: 1 + action_model_type: MLP + loss_type: l1 + state_dim: 17 + state_encoding: continuous_projector + kv_memory: + enabled: false + packed_train: false + rebased_sink: true + rollout_len: 8 + window: 4 + name: QwenOFT + qwenvl: + attn_implementation: flash_attention_2 + base_vlm: /mnt/checkpoints/latency-sensitive-bench/openvla/Qwen3-VL-4B-Instruct + enable_gradient_checkpointing: true +output_dir: /mnt/checkpoints/latency-sensitive-bench/openvla_runs/walker2d_rgb_state_l0_return_gt3000_100ep_openvla_native_continuous_projector_sft_5k +rl_games: + env_eval: + enabled: false + task: gymnasium +run_id: walker2d_rgb_state_l0_return_gt3000_100ep_openvla_native_continuous_projector_sft_5k +run_root_dir: /mnt/checkpoints/latency-sensitive-bench/openvla_runs +seed: 42 +trainer: + distributed_backend: deepspeed + eval_action_classification: false + eval_action_classification_interval: null + eval_interval: 500 + eval_num_batches: 200 + freeze_llm_layers: [] + freeze_modules: '' + freeze_tied_embedding: false + freeze_vit: false + gradient_accumulation_steps: 1 + is_resume: false + learning_rate: + action_model: 0.0001 + base: 2.0e-05 + qwen_vl_interface: 1.0e-05 + logging_frequency: 1 + lr_scheduler_type: cosine_with_min_lr + max_train_steps: 5000 + num_warmup_steps: 100 + optimizer: + betas: + - 0.9 + - 0.95 + eps: 1.0e-08 + fused: true + weight_decay: 1.0e-08 + per_latency_eval_num_batches: null + pretrained_checkpoint: null + profile_timing: + enabled: true + log_interval: 10 + save_interval: 500 + scheduler_specific_kwargs: + min_lr: 1.0e-06 +wandb_entity: dongqianyu99-zhejiang-university +wandb_project: latency-sensitive-bench diff --git a/zero-latency/walker2d/vla/starvla-qwenoft-h1/walker2d_rgb_state_l0_return_gt3000_100ep_openvla_native_continuous_projector_sft_5k/task_contract.json b/zero-latency/walker2d/vla/starvla-qwenoft-h1/walker2d_rgb_state_l0_return_gt3000_100ep_openvla_native_continuous_projector_sft_5k/task_contract.json new file mode 100644 index 0000000000000000000000000000000000000000..6c0a68cd29b247488218a7381facb1a7e9de3aa2 --- /dev/null +++ b/zero-latency/walker2d/vla/starvla-qwenoft-h1/walker2d_rgb_state_l0_return_gt3000_100ep_openvla_native_continuous_projector_sft_5k/task_contract.json @@ -0,0 +1,80 @@ +{ + "action_space": { + "dtype": "float32", + "high": [ + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0 + ], + "labels": [ + "right_thigh_torque", + "right_leg_torque", + "right_foot_torque", + "left_thigh_torque", + "left_leg_torque", + "left_foot_torque" + ], + "low": [ + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0 + ], + "type": "box" + }, + "base_prompt": "Move the Walker2d robot forward while keeping its torso upright. Predict six continuous torques in [-1, 1] ordered as right thigh, right leg, right foot, left thigh, left leg, and left foot.", + "env_fps": 125.0, + "env_id": "LatencyBench/Walker2dContinuous-v0", + "frame_stack": 1, + "make_kwargs": { + "base_env_id": "Walker2d-v4", + "base_make_kwargs": { + "ctrl_cost_weight": 0.001, + "exclude_current_positions_from_observation": true, + "forward_reward_weight": 1.0, + "healthy_reward": 1.0, + "reset_noise_scale": 0.005, + "terminate_when_unhealthy": true + }, + "render_mode": "rgb_array" + }, + "noop_action": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ], + "obs_fps": 125.0, + "registration_imports": [ + "latency_bench.envs.gymnasium_walker2d" + ], + "state_space": { + "labels": [ + "torso_height", + "torso_angle", + "right_thigh_angle", + "right_leg_angle", + "right_foot_angle", + "left_thigh_angle", + "left_leg_angle", + "left_foot_angle", + "torso_x_velocity", + "torso_z_velocity", + "torso_angular_velocity", + "right_thigh_angular_velocity", + "right_leg_angular_velocity", + "right_foot_angular_velocity", + "left_thigh_angular_velocity", + "left_leg_angular_velocity", + "left_foot_angular_velocity" + ] + }, + "task_name": "walker2d_rgb_state" +} diff --git a/zero-latency/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_l0_h1_5fps_g128_20260921/README.md b/zero-latency/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_l0_h1_5fps_g128_20260921/README.md new file mode 100644 index 0000000000000000000000000000000000000000..650e07073ae4c72cf95efbf7df0c5f55dec19d68 --- /dev/null +++ b/zero-latency/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_l0_h1_5fps_g128_20260921/README.md @@ -0,0 +1,31 @@ +# walker2d / qwenpi_v3 + +Training condition: `zero-latency`. Run: `walker2d_pi05_l0_h1_5fps_g128_20260921`. + +Copied from [Standard-Pipeline](https://huggingface.co/latency-sensitive-bench/Standard-Pipeline/tree/2208875f92b21e3f8ec2363a30a6705dfe56cc04/walker2d/zero_latency/Pi05) at `2208875f92b21e3f8ec2363a30a6705dfe56cc04`. + +- Checkpoint: `checkpoints/model.pt` +- Selection: only published VLA checkpoint +- Checkpoint SHA256: `3293bd997b44a633e894b98c17dde79fb0da70eb0e5ef55bd2d3b9175db3c36c` +- Configuration, FPS, actions, normalization and latency semantics retain their source values. +- This import does not run evaluations or change training/evaluation/acceptance status. + +Action horizon is 1. Use `config.yaml` and `dataset_statistics.json` with the matching +prepared backbone. `task_contract.json` is extracted from the original training configuration. +Original machine paths in historical configurations are provenance: bind checkpoint, +backbone, manifest and profile paths to downloaded local files when running inference. + +See [the original model notes](source/README.md) for experiment results and limitations. + +[Original provenance](source/provenance.json) retains dataset references and recorded experiment evidence. + +## Download + +```python +from huggingface_hub import snapshot_download +prefix = "zero-latency/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_l0_h1_5fps_g128_20260921" +snapshot_download("latency-sensitive-bench/benchmark-models", revision="", + allow_patterns=[prefix + "/**"]) +``` + +Use the verified import commit listed in the repository's import report. diff --git a/zero-latency/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_l0_h1_5fps_g128_20260921/checkpoints/model.pt b/zero-latency/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_l0_h1_5fps_g128_20260921/checkpoints/model.pt new file mode 100644 index 0000000000000000000000000000000000000000..86d1a121995eec4ab95346c49cb0bbde11d3eac9 --- /dev/null +++ b/zero-latency/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_l0_h1_5fps_g128_20260921/checkpoints/model.pt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3293bd997b44a633e894b98c17dde79fb0da70eb0e5ef55bd2d3b9175db3c36c +size 10922653853 diff --git a/zero-latency/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_l0_h1_5fps_g128_20260921/config.full.yaml b/zero-latency/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_l0_h1_5fps_g128_20260921/config.full.yaml new file mode 100644 index 0000000000000000000000000000000000000000..f9ce68994d520db9a4131f65facceea4a8e602b9 --- /dev/null +++ b/zero-latency/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_l0_h1_5fps_g128_20260921/config.full.yaml @@ -0,0 +1,265 @@ +framework: + name: QwenPI_v3 + qwenvl: + base_vlm: ${oc.env:PI05_BACKBONE_DIR} + attn_implementation: flash_attention_2 + vl_hidden_dim: 2560 + num_vl_layers: 36 + enable_gradient_checkpointing: false + action_model: + action_model_type: LayerwiseFM + action_dim: 6 + state_dim: 17 + state_encoding: continuous_projector + task_objective: null + action_horizon: 1 + repeated_diffusion_steps: 2 + num_inference_timesteps: 4 + add_pos_embed: true + max_seq_len: 1024 + num_target_vision_tokens: 32 + noise_beta_alpha: 1.5 + noise_beta_beta: 1.0 + noise_s: 0.999 + num_timestep_buckets: 1000 + diffusion_model_cfg: + action_dit_hidden_dim: 1024 + dropout: 0.2 + final_dropout: true + interleave_self_attention: true + norm_type: ada_norm + positional_embeddings: null + attention_head_dim: 64 + num_layers: 36 + input_embedding_dim: 1024 + cross_attention_dim: 1024 + output_dim: 1024 + num_attention_heads: 16 + action_env_dim: 6 + future_action_window_size: 0 + past_action_window_size: 0 +datasets: + vla_data: + dataset_py: lerobot_datasets + include_state: true + data_root_dir: / + data_mix: train + eval_data_mix: validation + custom_mixtures_path: ${PI05_RUN_DIR}/zero_latency/vla/mixture.json + action_type: continuous + sequential_step_sampling: true + eval_sequential_step_sampling: null + num_workers: 4 + eval_num_workers: 0 + prefetch_factor: 2 + persistent_workers: true + pin_memory: true + shuffle: true + action_balance: + enabled: false + strategy: balanced_epoch + action_key: action_id + target_flap_fraction: 0.3 + noop_id: 0 + flap_id: 1 + latency_curriculum: + enabled: false + strategy: exclusive + latencies: null + phase_steps: null + phase_distributions: null + new_latency_passes: 1.0 + replay_passes: 0.25 + target_total_passes: 2.0 + final_equalization: true + step_budget_mode: auto + eval_at_phase_end: false + save_at_phase_end: false + computed_plan: null + per_device_batch_size: 64 + load_all_data_for_training: true + num_obs_frames: 1 + image_mode: single + prompt_mode: raw + stitch_grid: + - 2 + - 2 + obs_image_size: + - 224 + - 224 + video_backend: pyav + lerobot_version: v2.0 + gymnasium_task_contract: + action_space: + dtype: float32 + high: + - 1.0 + - 1.0 + - 1.0 + - 1.0 + - 1.0 + - 1.0 + labels: + - right_thigh_torque + - right_leg_torque + - right_foot_torque + - left_thigh_torque + - left_leg_torque + - left_foot_torque + low: + - -1.0 + - -1.0 + - -1.0 + - -1.0 + - -1.0 + - -1.0 + type: box + base_prompt: Move the Walker2d robot forward while keeping its torso upright. + Predict six continuous torques in [-1, 1] ordered as right thigh, right leg, + right foot, left thigh, left leg, and left foot. + env_fps: 5 + env_id: LatencyBench/Walker2dContinuous-v0 + frame_stack: 1 + make_kwargs: + base_env_id: Walker2d-v4 + base_make_kwargs: + ctrl_cost_weight: 0.001 + exclude_current_positions_from_observation: true + forward_reward_weight: 1.0 + healthy_reward: 1.0 + reset_noise_scale: 0.005 + terminate_when_unhealthy: true + render_mode: rgb_array + noop_action: + - 0.0 + - 0.0 + - 0.0 + - 0.0 + - 0.0 + - 0.0 + obs_fps: 5 + registration_imports: + - latency_bench.envs.gymnasium_walker2d + state_space: + labels: + - torso_height + - torso_angle + - right_thigh_angle + - right_leg_angle + - right_foot_angle + - left_thigh_angle + - left_leg_angle + - left_foot_angle + - torso_x_velocity + - torso_z_velocity + - torso_angular_velocity + - right_thigh_angular_velocity + - right_leg_angular_velocity + - right_foot_angular_velocity + - left_thigh_angular_velocity + - left_leg_angular_velocity + - left_foot_angular_velocity + task_name: walker2d_rgb_state + active_action_dim: 6 +dataset: + source_hf: '' + config_name: null + source_subdir: null + converted_name: flappy_train + single_source_hf: '' + mixed_source_hf: '' + single_converted_name: flappy_train + mixed_converted_name: flappy_mixed_latency_train + single_latency_filter: null + mixed_latency_filter: null + force_download: false + setup_force: false + skip_verification: false + target_latency_unit: observation_steps + verify_rows: 200 + max_episodes: null + episodes_per_latency: null + latency_filter: null + debug_subset: + enabled: false + max_episodes: 5 + suffix: debug +base_model: + repo_id: Qwen/Qwen3-VL-4B-Instruct +initialization: + checkpoint_local_dir: null + checkpoint_hf_repo_id: null + checkpoint_filename: null +trainer: + max_train_steps: 5000 + num_warmup_steps: 100 + save_interval: 500 + eval_interval: 500 + eval_num_batches: 200 + per_latency_eval_num_batches: null + eval_action_classification: false + eval_action_classification_interval: null + cc_f1_tolerance: 1 + learning_rate: + base: 2.0e-05 + qwen_vl_interface: 1.0e-05 + action_model: 0.0001 + lr_scheduler_type: cosine_with_min_lr + scheduler_specific_kwargs: + min_lr: 1.0e-06 + freeze_modules: '' + freeze_vit: false + freeze_tied_embedding: false + freeze_llm_layers: [] + loss_scale: + vla: 1.0 + vlm: 0.1 + max_grad_norm: 1.0 + weight_decay: 0.0 + logging_frequency: 1 + profile_timing: + enabled: false + log_interval: 10 + gradient_clipping: 1.0 + gradient_accumulation_steps: 1 + distributed_backend: deepspeed + is_resume: false + pretrained_checkpoint: null + resume_step: 0 + reload_modules: null + optimizer: + name: AdamW + betas: + - 0.9 + - 0.95 + eps: 1.0e-08 + weight_decay: 1.0e-08 + fused: true + save_format: pt +run_id: walker2d_pi05_l0_h1_5fps_g128_20260921 +run_root_dir: ${PI05_RUN_DIR}/zero_latency/vla/training +seed: 42 +is_debug: false +version_id: '0.21' +wandb_project: latency-sensitive-bench +wandb_entity: dongqianyu99-zhejiang-university +wandb_name: walker2d_pi05_l0_h1_5fps_g128_20260921 +wandb_group: pi05-seven-env +wandb_tags: +- walker2d +- zero_latency +- Pi05 +- h1 +training_latency_condition: zero_latency +backbone: + repo_id: Qwen/Qwen3-VL-4B-Instruct + revision: ebb281ec70b05090aa6165b016eac8ec08e71b17 +checkpoint: + save_best_model: false + save_final_model: true + save_pt_file: true + save_training_state: true + local: + keep_last_n: 1 +config_yaml: ${PI05_RUN_DIR}/zero_latency/train.yaml +output_dir: ${PI05_RUN_DIR}/zero_latency/vla/training/walker2d_pi05_l0_h1_5fps_g128_20260921 diff --git a/zero-latency/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_l0_h1_5fps_g128_20260921/config.yaml b/zero-latency/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_l0_h1_5fps_g128_20260921/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..11ceafdaf97f1d1c31dffed71c94a21d0a842cb8 --- /dev/null +++ b/zero-latency/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_l0_h1_5fps_g128_20260921/config.yaml @@ -0,0 +1,142 @@ +framework: + name: QwenPI_v3 + qwenvl: + base_vlm: ${oc.env:PI05_BACKBONE_DIR} + attn_implementation: flash_attention_2 + vl_hidden_dim: 2560 + num_vl_layers: 36 + enable_gradient_checkpointing: false + action_model: + action_model_type: LayerwiseFM + action_dim: 6 + state_dim: 17 + state_encoding: continuous_projector + task_objective: null + action_horizon: 1 + repeated_diffusion_steps: 2 + num_inference_timesteps: 4 + add_pos_embed: true + max_seq_len: 1024 + num_target_vision_tokens: 32 + noise_beta_alpha: 1.5 + noise_beta_beta: 1.0 + noise_s: 0.999 + num_timestep_buckets: 1000 + diffusion_model_cfg: + action_dit_hidden_dim: 1024 + dropout: 0.2 + final_dropout: true + interleave_self_attention: true + norm_type: ada_norm + positional_embeddings: null + attention_head_dim: 64 + num_layers: 36 + input_embedding_dim: 1024 + cross_attention_dim: 1024 + output_dim: 1024 + num_attention_heads: 16 + action_env_dim: 6 + future_action_window_size: 0 + past_action_window_size: 0 +version_id: '0.21' +datasets: + vla_data: + include_state: true + obs_image_size: + - 224 + - 224 + image_mode: single + stitch_grid: + - 2 + - 2 + num_obs_frames: 1 + gymnasium_task_contract: &id001 + action_space: + dtype: float32 + high: + - 1.0 + - 1.0 + - 1.0 + - 1.0 + - 1.0 + - 1.0 + labels: + - right_thigh_torque + - right_leg_torque + - right_foot_torque + - left_thigh_torque + - left_leg_torque + - left_foot_torque + low: + - -1.0 + - -1.0 + - -1.0 + - -1.0 + - -1.0 + - -1.0 + type: box + base_prompt: Move the Walker2d robot forward while keeping its torso upright. + Predict six continuous torques in [-1, 1] ordered as right thigh, right leg, + right foot, left thigh, left leg, and left foot. + env_fps: 5 + env_id: LatencyBench/Walker2dContinuous-v0 + frame_stack: 1 + make_kwargs: + base_env_id: Walker2d-v4 + base_make_kwargs: + ctrl_cost_weight: 0.001 + exclude_current_positions_from_observation: true + forward_reward_weight: 1.0 + healthy_reward: 1.0 + reset_noise_scale: 0.005 + terminate_when_unhealthy: true + render_mode: rgb_array + noop_action: + - 0.0 + - 0.0 + - 0.0 + - 0.0 + - 0.0 + - 0.0 + obs_fps: 5 + registration_imports: + - latency_bench.envs.gymnasium_walker2d + state_space: + labels: + - torso_height + - torso_angle + - right_thigh_angle + - right_leg_angle + - right_foot_angle + - left_thigh_angle + - left_leg_angle + - left_foot_angle + - torso_x_velocity + - torso_z_velocity + - torso_angular_velocity + - right_thigh_angular_velocity + - right_leg_angular_velocity + - right_foot_angular_velocity + - left_thigh_angular_velocity + - left_leg_angular_velocity + - left_foot_angular_velocity + task_name: walker2d_rgb_state +trainer: + pretrained_checkpoint: null + profile_timing: + enabled: false + log_interval: 10 +backbone: + repo_id: Qwen/Qwen3-VL-4B-Instruct + revision: ebb281ec70b05090aa6165b016eac8ec08e71b17 +rl_games: + model_alias: pi05 + task: gymnasium + action_carrier: native + gymnasium: + task_contract: *id001 + env_eval: + image_size: 224 + frameskip: 1 + image_transform: raw_rgb + prompt_mode: raw diff --git a/zero-latency/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_l0_h1_5fps_g128_20260921/dataset_statistics.json b/zero-latency/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_l0_h1_5fps_g128_20260921/dataset_statistics.json new file mode 100644 index 0000000000000000000000000000000000000000..bfb9e72f511b15f1aba80c6eb1faa007fa88cee2 --- /dev/null +++ b/zero-latency/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_l0_h1_5fps_g128_20260921/dataset_statistics.json @@ -0,0 +1,180 @@ +{ + "new_embodiment": { + "action": { + "mean": [ + 0.8840610980987549, + 0.34449484944343567, + 0.11654752492904663, + 0.8480036854743958, + 0.32758569717407227, + 0.9875396490097046 + ], + "std": [ + 0.32312333583831787, + 0.7096099853515625, + 0.9048961400985718, + 0.343753844499588, + 0.7393367886543274, + 0.11781969666481018 + ], + "max": [ + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0 + ], + "min": [ + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0 + ], + "q01": [ + -0.5243898153305053, + -1.0, + -1.0, + -0.5588397586345673, + -1.0, + 0.5681399703025818 + ], + "q99": [ + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0 + ], + "mask": [ + true, + true, + true, + true, + true, + true + ] + }, + "state": { + "mean": [ + 0.38908126950263977, + -0.06854692101478577, + 0.5603042840957642, + 0.295283704996109, + 0.20391421020030975, + 0.7933022975921631, + 0.32835325598716736, + 0.7283809781074524, + -0.11274679005146027, + 0.36438047885894775, + -0.041566070169210434, + -0.006785172037780285, + -0.21467362344264984, + -0.2302139699459076, + -0.008182121440768242, + -0.19842907786369324, + -0.0016649218741804361 + ], + "std": [ + 0.15210415422916412, + 0.28636232018470764, + 0.08957907557487488, + 0.40512198209762573, + 0.5324621796607971, + 0.09768694639205933, + 0.4175085127353668, + 0.07296281307935677, + 0.3539130389690399, + 0.21250014007091522, + 0.5710971355438232, + 0.19414709508419037, + 0.797458827495575, + 0.7378174662590027, + 0.32647520303726196, + 0.7747778296470642, + 0.10512100160121918 + ], + "max": [ + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0 + ], + "min": [ + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0 + ], + "q01": [ + -0.01641636550426483, + -0.5725254046916962, + 0.1497414314746857, + -0.5082594507932663, + -0.704446622133255, + 0.4100250256061554, + -0.7682463544607162, + 0.7055095672607422, + -0.8300059294700622, + -0.09350170254707335, + -1.0, + -0.7184711527824402, + -1.0, + -1.0, + -0.9244865983724594, + -1.0, + -0.37168214678764344 + ], + "q99": [ + 0.6826163899898535, + 0.5332732021808625, + 0.7265087735652924, + 0.9429451179504394, + 0.963300715684891, + 0.9189580070972443, + 0.8997541296482087, + 0.7992216396331789, + 0.7250751173496247, + 0.9080468678474427, + 1.0, + 0.8595866250991845, + 1.0, + 1.0, + 1.0, + 1.0, + 0.04421793460845973 + ] + }, + "num_transitions": 85160, + "num_trajectories": 90 + } +} \ No newline at end of file diff --git a/zero-latency/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_l0_h1_5fps_g128_20260921/latency_prompt_map.json b/zero-latency/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_l0_h1_5fps_g128_20260921/latency_prompt_map.json new file mode 100644 index 0000000000000000000000000000000000000000..727b2dd7642ac6926aa0451c1859cfa4ed29d397 --- /dev/null +++ b/zero-latency/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_l0_h1_5fps_g128_20260921/latency_prompt_map.json @@ -0,0 +1,7 @@ +{ + "0": { + "prompt": "Move the Walker2d robot forward while keeping its torso upright. Predict six continuous torques in [-1, 1] ordered as right thigh, right leg, right foot, left thigh, left leg, and left foot. Current action latency is 0 raw frames (0.00 ms). The environment runs at 5 FPS and observations are emitted at 5 FPS. Choose the best next action.", + "latency_raw_frames": 0, + "latency_ms": 0.0 + } +} \ No newline at end of file diff --git a/zero-latency/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_l0_h1_5fps_g128_20260921/manifest.json b/zero-latency/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_l0_h1_5fps_g128_20260921/manifest.json new file mode 100644 index 0000000000000000000000000000000000000000..619523ff88c52f8db705af4a0c065c5be884716b --- /dev/null +++ b/zero-latency/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_l0_h1_5fps_g128_20260921/manifest.json @@ -0,0 +1,184 @@ +{ + "dataset_name": "walker2d_h1_5fps_l0", + "env_name": "walker2d_rgb_state", + "episodes": 90, + "frames": 85160, + "task_prompts": [ + "Move the Walker2d robot forward while keeping its torso upright. Predict six continuous torques in [-1, 1] ordered as right thigh, right leg, right foot, left thigh, left leg, and left foot. Current action latency is 0 raw frames (0.00 ms). The environment runs at 5 FPS and observations are emitted at 5 FPS. Choose the best next action." + ], + "format": "lerobot_v2_image_parquet", + "source": "${PI05_RUN_DIR}/zero_latency/data/raw_5fps", + "integration_name": "gymnasium", + "task_name": "walker2d_rgb_state", + "action_layout": "gymnasium_continuous_v1", + "action_labels": [ + "right_thigh_torque", + "right_leg_torque", + "right_foot_torque", + "left_thigh_torque", + "left_leg_torque", + "left_foot_torque" + ], + "carrier_action_labels": [ + "right_thigh_torque", + "right_leg_torque", + "right_foot_torque", + "left_thigh_torque", + "left_leg_torque", + "left_foot_torque" + ], + "action_dim": 6, + "active_action_dim": 6, + "action_carrier": "native", + "rows_unit": "decision_step", + "obs_fps": 5, + "obs_stride_raw_frames": 1, + "uses_state": true, + "state_dim": 17, + "state_labels": [ + "torso_height", + "torso_angle", + "right_thigh_angle", + "right_leg_angle", + "right_foot_angle", + "left_thigh_angle", + "left_leg_angle", + "left_foot_angle", + "torso_x_velocity", + "torso_z_velocity", + "torso_angular_velocity", + "right_thigh_angular_velocity", + "right_leg_angular_velocity", + "right_foot_angular_velocity", + "left_thigh_angular_velocity", + "left_leg_angular_velocity", + "left_foot_angular_velocity" + ], + "state_normalization": { + "type": "min_max", + "min": [ + 0.8001501560211182, + -0.9963381290435791, + -1.2083065509796143, + -2.6031124591827393, + -1.4880801439285278, + -1.7395809888839722, + -2.8114991188049316, + -1.4173651933670044, + -1.9014521837234497, + -5.880380630493164, + -10.0, + -10.0, + -10.0, + -10.0, + -10.0, + -10.0, + -10.0 + ], + "max": [ + 1.4312387704849243, + 0.9942846298217773, + 0.35824716091156006, + 0.42332249879837036, + 1.4080666303634644, + 0.19403602182865143, + 0.508411705493927, + 1.1665540933609009, + 9.758368492126465, + 2.7117056846618652, + 10.0, + 10.0, + 10.0, + 10.0, + 10.0, + 10.0, + 10.0 + ] + }, + "robot_type": "rl_games_gymnasium", + "latency_prompt_map_path": "${PI05_RUN_DIR}/zero_latency/data/lerobot/walker2d_h1_5fps_l0/latency_prompt_map.json", + "custom_mixtures_path": "${PI05_RUN_DIR}/zero_latency/data/lerobot/_generated_mixtures/walker2d_h1_5fps_l0.json", + "gymnasium_task": { + "action_space": { + "dtype": "float32", + "high": [ + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0 + ], + "labels": [ + "right_thigh_torque", + "right_leg_torque", + "right_foot_torque", + "left_thigh_torque", + "left_leg_torque", + "left_foot_torque" + ], + "low": [ + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0 + ], + "type": "box" + }, + "base_prompt": "Move the Walker2d robot forward while keeping its torso upright. Predict six continuous torques in [-1, 1] ordered as right thigh, right leg, right foot, left thigh, left leg, and left foot.", + "env_fps": 5, + "env_id": "LatencyBench/Walker2dContinuous-v0", + "frame_stack": 1, + "make_kwargs": { + "base_env_id": "Walker2d-v4", + "base_make_kwargs": { + "ctrl_cost_weight": 0.001, + "exclude_current_positions_from_observation": true, + "forward_reward_weight": 1.0, + "healthy_reward": 1.0, + "reset_noise_scale": 0.005, + "terminate_when_unhealthy": true + }, + "render_mode": "rgb_array" + }, + "noop_action": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ], + "obs_fps": 5, + "registration_imports": [ + "latency_bench.envs.gymnasium_walker2d" + ], + "state_space": { + "labels": [ + "torso_height", + "torso_angle", + "right_thigh_angle", + "right_leg_angle", + "right_foot_angle", + "left_thigh_angle", + "left_leg_angle", + "left_foot_angle", + "torso_x_velocity", + "torso_z_velocity", + "torso_angular_velocity", + "right_thigh_angular_velocity", + "right_leg_angular_velocity", + "right_foot_angular_velocity", + "left_thigh_angular_velocity", + "left_leg_angular_velocity", + "left_foot_angular_velocity" + ] + }, + "task_name": "walker2d_rgb_state" + }, + "validation_dataset_name": "walker2d_h1_5fps_l0__val", + "validation_episodes": 10, + "validation_frames": 9664 +} \ No newline at end of file diff --git a/zero-latency/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_l0_h1_5fps_g128_20260921/provenance.json b/zero-latency/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_l0_h1_5fps_g128_20260921/provenance.json new file mode 100644 index 0000000000000000000000000000000000000000..139444e2d06e0e988a20e0d5be1914d4e64fc779 --- /dev/null +++ b/zero-latency/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_l0_h1_5fps_g128_20260921/provenance.json @@ -0,0 +1,34 @@ +{ + "task": "walker2d", + "model": "qwenpi_v3", + "training_condition": "zero-latency", + "training_run_id": "walker2d_pi05_l0_h1_5fps_g128_20260921", + "source": { + "repo_id": "latency-sensitive-bench/Standard-Pipeline", + "revision": "2208875f92b21e3f8ec2363a30a6705dfe56cc04", + "path": "walker2d/zero_latency/Pi05", + "url": "https://huggingface.co/latency-sensitive-bench/Standard-Pipeline/tree/2208875f92b21e3f8ec2363a30a6705dfe56cc04/walker2d/zero_latency/Pi05" + }, + "hf_repo_id": "latency-sensitive-bench/benchmark-models", + "hf_path": "zero-latency/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_l0_h1_5fps_g128_20260921", + "checkpoint": { + "source_file": "walker2d/zero_latency/Pi05/checkpoints/model.pt", + "source_sha256": "3293bd997b44a633e894b98c17dde79fb0da70eb0e5ef55bd2d3b9175db3c36c", + "source_bytes": 10922653853, + "file": "checkpoints/model.pt", + "selection_rule": "only published VLA checkpoint", + "method": "copy", + "sha256": "3293bd997b44a633e894b98c17dde79fb0da70eb0e5ef55bd2d3b9175db3c36c", + "bytes": 10922653853 + }, + "config_source": "walker2d/zero_latency/Pi05/config.yaml", + "task_contract_file": "task_contract.json", + "missing_source_files": [], + "migration": { + "operation": "copy_selected_inference_bundle", + "source_modified": false, + "evaluation_performed": false, + "experiment_status": "preserved_in_source_evidence" + }, + "action_horizon": 1 +} diff --git a/zero-latency/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_l0_h1_5fps_g128_20260921/source/README.md b/zero-latency/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_l0_h1_5fps_g128_20260921/source/README.md new file mode 100644 index 0000000000000000000000000000000000000000..fb6580cf77eb8a2b5ba5311c25e438bb5846f8ba --- /dev/null +++ b/zero-latency/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_l0_h1_5fps_g128_20260921/source/README.md @@ -0,0 +1,3 @@ +# Walker2d Pi0.5 H1 — zero latency + +Fresh QwenPI_v3 action head, pinned Qwen3-VL-4B-Instruct, 5000 updates, seed42, global128=micro64 x accumulation1 x2H200, GCfalse, ZeRO2. RGB224/state17/native continuous action6/H1, 5/5FPS, train-only state minmax once. Training and saved real-state forward completed; L0sim20, RTX3090 P/B/F remain separate pending stages. Inference bundle excludes optimizer and full training state. diff --git a/zero-latency/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_l0_h1_5fps_g128_20260921/source/provenance.json b/zero-latency/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_l0_h1_5fps_g128_20260921/source/provenance.json new file mode 100644 index 0000000000000000000000000000000000000000..3ead5048adf46f9949e773f18f6a62cf62267db5 --- /dev/null +++ b/zero-latency/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_l0_h1_5fps_g128_20260921/source/provenance.json @@ -0,0 +1,115 @@ +{ + "task": "walker2d", + "backbone": { + "repo_id": "Qwen/Qwen3-VL-4B-Instruct", + "revision": "ebb281ec70b05090aa6165b016eac8ec08e71b17" + }, + "checkpoint": "steps_5000_pytorch_model.pt", + "training_steps": 5000, + "seed": 42, + "global_batch": 128, + "training_run_id": "walker2d_pi05_l0_h1_5fps_g128_20260921", + "condition": "zero_latency", + "source": { + "task": "walker2d", + "source_code": { + "archive_sha256": { + "lsb-pi05-h1-20260918.tar.gz": "08989cd9c3518cd79cb5ec065300755b8afe7555e1b6d65fed829549ef9127c4", + "pi05-training-package.tar.gz": "0e1c6f7cceb8979b525ceaa3c209f005049166b97eb97a1602e13ed8d7384b65" + }, + "original_code_root": "2de3816f07ecea4002964d046c4dfa581a658671", + "starvla": "3430c45edf6a08e4cfcaa2fce58bb4ea05995617", + "sample_factory": "4b11d8754ffc989cdd15a340cfc9be0277377d5b", + "task_overlay": "batch128 and AirRaid manifest clock; recorded separately" + }, + "code_overlay": { + "files": { + "scripts/gym_adapt/gr00t.py": "07fddb83f5a7f0766fcbf025c521703fe295ee3754017187c5ac4b66e1bcdce1", + "tests/data/test_starvla_tasks.py": "4fd24cea648baa1b3af95fddd3311610a80fd346cb633a83bb6f94ea6423c797" + }, + "tests": "16 relevant config/export tests passed,11 deselected", + "changes": [ + "global_batch parameter default64,newrun128", + "microbatch64 CLI support", + "AirRaid evaluation uses manifest FPS" + ], + "verified_at": "2026-09-20T07:12:04.593097+00:00" + }, + "dataset": { + "repo": "latency-sensitive-bench/Standard-Pipeline", + "revision": "b0cdd0505f6c6f2dadfca67638e4fec1b252756c", + "prefix": "walker2d/zero_latency/shared/source_100ep_v1/demonstrations/raw/", + "sha256": { + "metadata.json": "c8ea2669493beabd97d9e7ecd06cc443adcb1760d8e71dd03588ea4adea20ae9", + "train.parquet": "6a105846e7def09f10fd2a985413cc9a04b9433f57519d330ce33a1033ba7e65", + "val.parquet": "687c1a5d08b3641b5ae76d218dd6afe483f8873b057aa3ab7a81108a6772f424" + }, + "timing_derivation": { + "source_env_fps": 125, + "source_obs_fps": 125, + "target_env_fps": 5, + "target_obs_fps": 5, + "source_revision": "b0cdd0505f6c6f2dadfca67638e4fec1b252756c", + "source_prefix": "walker2d/zero_latency/shared/source_100ep_v1/demonstrations/raw/", + "changed": [ + "prompt", + "dataset FPS metadata" + ], + "unchanged": [ + "images", + "state", + "action", + "reward", + "episode split", + "zero latency", + "native physics" + ], + "source_checkpoint_config_is_historical": true + }, + "splits": { + "train": { + "rows": 85160, + "episodes": 90, + "return_min": 3019.899699331727, + "return_mean": 4039.562214606017, + "action_min": -1.0, + "action_max": 1.0 + }, + "val": { + "rows": 9664, + "episodes": 10, + "return_min": 3668.9183055646718, + "return_mean": 4200.107031565485, + "action_min": -1.0, + "action_max": 1.0 + } + } + }, + "backbone": { + "repo_id": "Qwen/Qwen3-VL-4B-Instruct", + "revision": "ebb281ec70b05090aa6165b016eac8ec08e71b17" + }, + "protocol": { + "env_fps": 5, + "obs_fps": 5, + "action_horizon": 1, + "state_dim": 17, + "action_dim": 6, + "action_units": "native continuous torques, environment clips to [-1,1]", + "native_physics_unchanged": true, + "updates": 5000, + "seed": 42, + "global_batch": 128, + "micro_batch": 64, + "accumulation": 1, + "gpus": 2, + "physical_gpu_indices": null, + "gradient_checkpointing": false, + "zero_stage": 2, + "allocator": "expandable_segments:True", + "state_encoding": "continuous_projector" + } + }, + "training_config_sha256": "1dafa02de089ac0979ab0c21172ffce03b7ad1edd96a4805232051520c191cce", + "dataset_manifest_sha256": "3a1ff8bca0476c8bf524e835fe4b8fafa06f2c78fc2dfb63586b82f5d121655c" +} diff --git a/zero-latency/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_l0_h1_5fps_g128_20260921/task_contract.json b/zero-latency/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_l0_h1_5fps_g128_20260921/task_contract.json new file mode 100644 index 0000000000000000000000000000000000000000..7567af8871901a6d8a45081d77c7b0ca38924e72 --- /dev/null +++ b/zero-latency/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_l0_h1_5fps_g128_20260921/task_contract.json @@ -0,0 +1,80 @@ +{ + "action_space": { + "dtype": "float32", + "high": [ + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0 + ], + "labels": [ + "right_thigh_torque", + "right_leg_torque", + "right_foot_torque", + "left_thigh_torque", + "left_leg_torque", + "left_foot_torque" + ], + "low": [ + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0 + ], + "type": "box" + }, + "base_prompt": "Move the Walker2d robot forward while keeping its torso upright. Predict six continuous torques in [-1, 1] ordered as right thigh, right leg, right foot, left thigh, left leg, and left foot.", + "env_fps": 5, + "env_id": "LatencyBench/Walker2dContinuous-v0", + "frame_stack": 1, + "make_kwargs": { + "base_env_id": "Walker2d-v4", + "base_make_kwargs": { + "ctrl_cost_weight": 0.001, + "exclude_current_positions_from_observation": true, + "forward_reward_weight": 1.0, + "healthy_reward": 1.0, + "reset_noise_scale": 0.005, + "terminate_when_unhealthy": true + }, + "render_mode": "rgb_array" + }, + "noop_action": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ], + "obs_fps": 5, + "registration_imports": [ + "latency_bench.envs.gymnasium_walker2d" + ], + "state_space": { + "labels": [ + "torso_height", + "torso_angle", + "right_thigh_angle", + "right_leg_angle", + "right_foot_angle", + "left_thigh_angle", + "left_leg_angle", + "left_foot_angle", + "torso_x_velocity", + "torso_z_velocity", + "torso_angular_velocity", + "right_thigh_angular_velocity", + "right_leg_angular_velocity", + "right_foot_angular_velocity", + "left_thigh_angular_velocity", + "left_leg_angular_velocity", + "left_foot_angular_velocity" + ] + }, + "task_name": "walker2d_rgb_state" +} diff --git a/zero-latency/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_l0_h1_5fps_g128_20260921/validation.json b/zero-latency/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_l0_h1_5fps_g128_20260921/validation.json new file mode 100644 index 0000000000000000000000000000000000000000..b9da968c602ae8573cf6235b59ee38be0d493908 --- /dev/null +++ b/zero-latency/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_l0_h1_5fps_g128_20260921/validation.json @@ -0,0 +1,18 @@ +{ + "state": "SAVED_BUNDLE_RELOAD_AND_REAL_SAMPLE_FORWARD_VERIFIED", + "verified_at": "2026-09-21T05:22:44.997608+00:00", + "forward_shape": [ + 1, + 1, + 6 + ], + "forward_finite": true, + "heldout_sample": 0, + "loader_rows": 9664, + "action_horizon": 1, + "state_dim": 17, + "action_dim": 6, + "state_encoding": "continuous_projector", + "checkpoint_sha256": "3293bd997b44a633e894b98c17dde79fb0da70eb0e5ef55bd2d3b9175db3c36c", + "reload": "Existing strict key check with documented tied Qwen lm_head equivalence only" +}