diff --git a/latency-aware/hopper/small-policy/sample-factory-qwengr00t-rtx3090-v1/README.md b/latency-aware/hopper/small-policy/sample-factory-qwengr00t-rtx3090-v1/README.md new file mode 100644 index 0000000000000000000000000000000000000000..9e32436dc77e2a2f6e97af7ba75b39c9762a417d --- /dev/null +++ b/latency-aware/hopper/small-policy/sample-factory-qwengr00t-rtx3090-v1/README.md @@ -0,0 +1,26 @@ +# hopper / sample-factory-appo + +Training condition: `latency-aware`. Run: `hopper_profile_appo_h1_5fps_20260917`. + +Copied from [Standard-Pipeline](https://huggingface.co/latency-sensitive-bench/Standard-Pipeline/tree/2208875f92b21e3f8ec2363a30a6705dfe56cc04/hopper/profile_latency/small_model/GR00T) at `2208875f92b21e3f8ec2363a30a6705dfe56cc04`. + +- Checkpoint: `checkpoint.pth` +- Selection: training_best +- Checkpoint SHA256: `3dc620932950a33e730f0ba615f2a85dded0c11dd24d0d27ff08329090bbc9cc` +- 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/hopper/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/hopper/small-policy/sample-factory-qwengr00t-rtx3090-v1/checkpoint.pth b/latency-aware/hopper/small-policy/sample-factory-qwengr00t-rtx3090-v1/checkpoint.pth new file mode 100644 index 0000000000000000000000000000000000000000..e2f4517dfceced636362a591b158291d4eaf863f --- /dev/null +++ b/latency-aware/hopper/small-policy/sample-factory-qwengr00t-rtx3090-v1/checkpoint.pth @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3dc620932950a33e730f0ba615f2a85dded0c11dd24d0d27ff08329090bbc9cc +size 27445 diff --git a/latency-aware/hopper/small-policy/sample-factory-qwengr00t-rtx3090-v1/config.full.yaml b/latency-aware/hopper/small-policy/sample-factory-qwengr00t-rtx3090-v1/config.full.yaml new file mode 100644 index 0000000000000000000000000000000000000000..c9088443dac4871b3d8f790c6ef5a0fdba53bab5 --- /dev/null +++ b/latency-aware/hopper/small-policy/sample-factory-qwengr00t-rtx3090-v1/config.full.yaml @@ -0,0 +1,151 @@ +env: + state_space: + labels: + - torso_height + - torso_angle + - thigh_angle + - leg_angle + - foot_angle + - torso_x_velocity + - torso_z_velocity + - torso_angular_velocity + - thigh_angular_velocity + - leg_angular_velocity + - foot_angular_velocity + name: gymnasium + task_name: hopper + env_id: LatencyBench/Hopper-v0 + registration_imports: + - latency_bench.envs.gymnasium_hopper + make_kwargs: + base_env_id: Hopper-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: + - thigh_torque + - leg_torque + - foot_torque + low: + - -1.0 + - -1.0 + - -1.0 + high: + - 1.0 + - 1.0 + - 1.0 + dtype: float32 + noop_action: + - 0.0 + - 0.0 + - 0.0 + base_prompt: Move the Hopper robot forward while keeping its torso upright. Predict + three continuous torques in [-1, 1] ordered as thigh, leg, and foot. +experiment: + name: hopper_profile_appo_h1_5fps_20260917 + seed: 3333 +backend: + type: sample_factory + algo: APPO + device: cuda + train_dir: /mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/hopper/profile_latency/teacher/checkpoints + restart_behavior: overwrite + run_mode: train + extra_args: + - --wandb_user + - dongqianyu99-zhejiang-university +executor: + mode: simulated +latency: + method: iid + profile_path: /mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/hopper/profile_latency/profile_3090_20260917T141828786022Z/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 + num_batches_to_accumulate: 2 + policy_workers_per_policy: 1 + max_policy_lag: 10000 + learning_rate: 0.00295 + lr_schedule: linear_decay + lr_schedule_kl_threshold: 0.008 + gamma: 0.99 + gae_lambda: 0.95 + ppo_clip_ratio: 0.2 + ppo_clip_value: 1.0 + value_loss_coeff: 1.3 + max_grad_norm: 3.5 + exploration_loss: entropy + exploration_loss_coeff: 0.0 + kl_loss_coeff: 0.1 + reward_scale: 1.0 + reward_clip: 1000.0 + async_rl: false + serial_mode: false + batched_sampling: false + with_vtrace: false + use_rnn: false + env_framestack: 4 + encoder_mlp_layers: + - 64 + - 64 + nonlinearity: tanh + adaptive_stddev: false + policy_initialization: torch_default + initial_stddev: 1.0 + actor_critic_share_weights: true + shuffle_minibatches: false + value_bootstrap: false + normalize_input: true + normalize_returns: true + decorrelate_experience_max_seconds: 10 + decorrelate_envs_on_one_worker: true + set_workers_cpu_affinity: true + force_envs_single_thread: true + save_every_sec: 600 + keep_checkpoints: 3 + save_best_every_sec: 60 + save_best_after: 100000 +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/hopper/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/hopper/small-policy/sample-factory-qwengr00t-rtx3090-v1/config.json b/latency-aware/hopper/small-policy/sample-factory-qwengr00t-rtx3090-v1/config.json new file mode 100644 index 0000000000000000000000000000000000000000..df32746990463b046961a9796f104f50eff9479e --- /dev/null +++ b/latency-aware/hopper/small-policy/sample-factory-qwengr00t-rtx3090-v1/config.json @@ -0,0 +1,271 @@ +{ + "help": false, + "algo": "APPO", + "env": "latency_gymnasium", + "experiment": "hopper_profile_appo_h1_5fps_20260917", + "train_dir": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/hopper/profile_latency/teacher/checkpoints", + "restart_behavior": "overwrite", + "device": "gpu", + "seed": 3333, + "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": 10000, + "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": false, + "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": 1.0, + "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": 10, + "decorrelate_envs_on_one_worker": true, + "actor_worker_gpus": [], + "set_workers_cpu_affinity": true, + "force_envs_single_thread": true, + "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": 3, + "load_checkpoint_kind": "latest", + "save_milestones_sec": -1, + "save_best_every_sec": 60, + "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": 4, + "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": "hopper", + "gym_env_id": "LatencyBench/Hopper-v0", + "gym_make_kwargs_json": "{\"base_env_id\": \"Hopper-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_hopper\"]", + "gym_action_space_json": "{\"type\": \"box\", \"labels\": [\"thigh_torque\", \"leg_torque\", \"foot_torque\"], \"low\": [-1.0, -1.0, -1.0], \"high\": [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]", + "gym_base_prompt": "Move the Hopper robot forward while keeping its torso upright. Predict three continuous torques in [-1, 1] ordered as thigh, leg, and foot.", + "gym_state_labels_json": "[\"torso_height\", \"torso_angle\", \"thigh_angle\", \"leg_angle\", \"foot_angle\", \"torso_x_velocity\", \"torso_z_velocity\", \"torso_angular_velocity\", \"thigh_angular_velocity\", \"leg_angular_velocity\", \"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/hopper/profile_latency/profile_3090_20260917T141828786022Z/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/hopper/profile_latency/teacher/checkpoints/hopper_profile_appo_h1_5fps_20260917/episode_metrics.jsonl", + "ppo": null, + "command_line": "--mode train --algo APPO --env latency_gymnasium --experiment hopper_profile_appo_h1_5fps_20260917 --train_dir /mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/hopper/profile_latency/teacher/checkpoints --restart_behavior overwrite --device gpu --seed 3333 --episode_metrics_path /mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/hopper/profile_latency/teacher/checkpoints/hopper_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 10000 --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 --initial_stddev 1.0 --env_framestack 4 --gamma 0.99 --gae_lambda 0.95 --ppo_clip_ratio 0.2 --ppo_clip_value 1.0 --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 --decorrelate_experience_max_seconds 10 --save_every_sec 600 --keep_checkpoints 3 --save_best_every_sec 60 --save_best_after 100000 --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 False --with_vtrace False --decorrelate_envs_on_one_worker True --set_workers_cpu_affinity True --force_envs_single_thread True --actor_critic_share_weights 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/hopper/profile_latency/profile_3090_20260917T141828786022Z/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 hopper --gym-env-id LatencyBench/Hopper-v0 --gym-make-kwargs-json {\"base_env_id\": \"Hopper-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_hopper\"] --gym-action-space-json {\"type\": \"box\", \"labels\": [\"thigh_torque\", \"leg_torque\", \"foot_torque\"], \"low\": [-1.0, -1.0, -1.0], \"high\": [1.0, 1.0, 1.0], \"dtype\": \"float32\"} --gym-noop-action-json [0.0, 0.0, 0.0] --gym-base-prompt Move the Hopper robot forward while keeping its torso upright. Predict three continuous torques in [-1, 1] ordered as thigh, leg, and foot. --gym-state-labels-json [\"torso_height\", \"torso_angle\", \"thigh_angle\", \"leg_angle\", \"foot_angle\", \"torso_x_velocity\", \"torso_z_velocity\", \"torso_angular_velocity\", \"thigh_angular_velocity\", \"leg_angular_velocity\", \"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": "hopper_profile_appo_h1_5fps_20260917", + "train_dir": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/hopper/profile_latency/teacher/checkpoints", + "restart_behavior": "overwrite", + "device": "gpu", + "seed": 3333, + "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": 10000, + "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": false, + "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": 1.0, + "with_vtrace": false, + "max_grad_norm": 3.5, + "learning_rate": 0.00295, + "lr_schedule_kl_threshold": 0.008, + "normalize_input": true, + "decorrelate_experience_max_seconds": 10, + "decorrelate_envs_on_one_worker": true, + "set_workers_cpu_affinity": true, + "force_envs_single_thread": true, + "train_for_env_steps": 10000000, + "save_every_sec": 600, + "keep_checkpoints": 3, + "save_best_every_sec": 60, + "save_best_after": 100000, + "encoder_mlp_layers": [ + 64, + 64 + ], + "use_rnn": false, + "nonlinearity": "tanh", + "policy_initialization": "torch_default", + "actor_critic_share_weights": true, + "adaptive_stddev": false, + "initial_stddev": 1.0, + "use_env_info_cache": false, + "env_framestack": 4, + "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": "hopper", + "gym_env_id": "LatencyBench/Hopper-v0", + "gym_make_kwargs_json": "{\"base_env_id\": \"Hopper-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_hopper\"]", + "gym_action_space_json": "{\"type\": \"box\", \"labels\": [\"thigh_torque\", \"leg_torque\", \"foot_torque\"], \"low\": [-1.0, -1.0, -1.0], \"high\": [1.0, 1.0, 1.0], \"dtype\": \"float32\"}", + "gym_noop_action_json": "[0.0, 0.0, 0.0]", + "gym_base_prompt": "Move the Hopper robot forward while keeping its torso upright. Predict three continuous torques in [-1, 1] ordered as thigh, leg, and foot.", + "gym_state_labels_json": "[\"torso_height\", \"torso_angle\", \"thigh_angle\", \"leg_angle\", \"foot_angle\", \"torso_x_velocity\", \"torso_z_velocity\", \"torso_angular_velocity\", \"thigh_angular_velocity\", \"leg_angular_velocity\", \"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/hopper/profile_latency/profile_3090_20260917T141828786022Z/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/hopper/profile_latency/teacher/checkpoints/hopper_profile_appo_h1_5fps_20260917/episode_metrics.jsonl", + "output_dir": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/hopper/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/hopper/profile_latency/teacher/training", + "wandb_unique_id": "hopper_profile_appo_h1_5fps_20260917" +} \ No newline at end of file diff --git a/latency-aware/hopper/small-policy/sample-factory-qwengr00t-rtx3090-v1/profile/DONE b/latency-aware/hopper/small-policy/sample-factory-qwengr00t-rtx3090-v1/profile/DONE new file mode 100644 index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391 diff --git a/latency-aware/hopper/small-policy/sample-factory-qwengr00t-rtx3090-v1/profile/hardware.json b/latency-aware/hopper/small-policy/sample-factory-qwengr00t-rtx3090-v1/profile/hardware.json new file mode 100644 index 0000000000000000000000000000000000000000..e8204e81807044975c0e0a0953d16d6835192f31 --- /dev/null +++ b/latency-aware/hopper/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/hopper/small-policy/sample-factory-qwengr00t-rtx3090-v1/profile/latency_burst_model.json b/latency-aware/hopper/small-policy/sample-factory-qwengr00t-rtx3090-v1/profile/latency_burst_model.json new file mode 100644 index 0000000000000000000000000000000000000000..091d90d99241bd6c8b6c483c0ce3c57eca06e760 --- /dev/null +++ b/latency-aware/hopper/small-policy/sample-factory-qwengr00t-rtx3090-v1/profile/latency_burst_model.json @@ -0,0 +1,1315 @@ +{ + "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, + 1.6000000000000005, + 2.200000000000001, + 2.8000000000000016, + 3.3999999999999986, + 3.999999999999999, + 4.6, + 5.2, + 5.800000000000001, + 6.400000000000001, + 6.999999999999998, + 7.6, + 8.2, + 8.8, + 9.4, + 10.000000000000002, + 10.599999999999998, + 11.2, + 11.799999999999999, + 12.4, + 13.0, + 13.600000000000001, + 14.2, + 14.799999999999999, + 15.399999999999999, + 16.0, + 16.6, + 17.200000000000003, + 17.8, + 18.400000000000002, + 19.000000000000004, + 19.6, + 20.199999999999996, + 20.799999999999997, + 21.4, + 22.0, + 22.599999999999998, + 23.2, + 23.8, + 24.400000000000002, + 25.0, + 25.6, + 26.200000000000003, + 26.800000000000004, + 27.4, + 27.999999999999996, + 28.599999999999998, + 29.2, + 29.799999999999997, + 30.4, + 31.0, + 31.0, + 31.0, + 31.0, + 31.0, + 31.0, + 31.0, + 31.0, + 31.0, + 31.0, + 31.0, + 31.0, + 31.0, + 31.0, + 31.0, + 31.0, + 31.0, + 31.0, + 31.0, + 31.0, + 31.0, + 31.0, + 31.0, + 31.0, + 31.0, + 31.0, + 31.0, + 31.0, + 31.0, + 31.0, + 31.0, + 31.0, + 31.0, + 31.0, + 31.0, + 31.0, + 31.0 + ], + "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, + 31 + ], + "burst_merge_gap_records": 30, + "burst_rank_processes": [ + { + "draw_count": 32, + "dwell_length_spearman_rho": -1.0, + "level_ms": { + "distribution_type": "inverse_cdf", + "latency_ms": [ + 133.18363499641418, + 133.18363499641418, + 133.18363499641418, + 133.18363499641418, + 133.18363499641418, + 133.18363499641418, + 133.18363499641418, + 133.18363499641418, + 133.18363499641418, + 133.18363499641418, + 133.18363499641418, + 133.18363499641418, + 133.18363499641418, + 133.18363499641418, + 133.18363499641418, + 133.18363499641418, + 133.18363499641418, + 133.18363499641418, + 133.18363499641418, + 133.18363499641418, + 133.18363499641418, + 133.18363499641418, + 133.18363499641418, + 133.18363499641418, + 133.18363499641418, + 133.18363499641418, + 133.18363499641418, + 133.18363499641418, + 133.18363499641418, + 133.18363499641418, + 133.18363499641418, + 133.18363499641418, + 133.18363499641418, + 133.18363499641418, + 133.18363499641418, + 133.18363499641418, + 133.18363499641418, + 133.2983943605423, + 133.41315372467042, + 133.52791308879853, + 133.64267245292663, + 133.75743181705474, + 133.87219118118287, + 133.98695054531098, + 134.10170990943908, + 134.2164692735672, + 134.33122863769532, + 134.44598800182342, + 134.56074736595153, + 134.67550673007966, + 134.79026609420777, + 134.90502545833587, + 135.01978482246398, + 135.1345441865921, + 135.2493035507202, + 135.36406291484832, + 135.47882227897645, + 135.59358164310456, + 135.70834100723266, + 135.82310037136077, + 135.9378597354889, + 136.052619099617, + 136.1673784637451, + 136.28213782787324, + 136.39689719200135, + 136.51165655612945, + 136.62641592025756, + 136.7411752843857, + 136.8559346485138, + 136.9706940126419, + 137.08545337677003, + 137.20021274089814, + 137.31497210502624, + 137.42973146915435, + 137.54449083328248, + 137.65925019741059, + 137.7740095615387, + 137.8887689256668, + 138.00352828979493, + 138.11828765392303, + 138.23304701805114, + 138.34780638217927, + 138.46256574630738, + 138.57732511043548, + 138.6920844745636, + 138.80684383869172, + 138.92160320281982, + 138.92160320281982, + 138.92160320281982, + 138.92160320281982, + 138.92160320281982, + 138.92160320281982, + 138.92160320281982, + 138.92160320281982, + 138.92160320281982, + 138.92160320281982, + 138.92160320281982, + 138.92160320281982, + 138.92160320281982, + 138.92160320281982, + 138.92160320281982, + 138.92160320281982, + 138.92160320281982, + 138.92160320281982, + 138.92160320281982, + 138.92160320281982, + 138.92160320281982, + 138.92160320281982, + 138.92160320281982, + 138.92160320281982, + 138.92160320281982, + 138.92160320281982, + 138.92160320281982, + 138.92160320281982, + 138.92160320281982, + 138.92160320281982, + 138.92160320281982, + 138.92160320281982, + 138.92160320281982, + 138.92160320281982, + 138.92160320281982, + 138.92160320281982, + 138.92160320281982 + ], + "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": [ + -1.5463809967041016, + -1.5463809967041016, + -1.5463809967041016, + -1.5463809967041016, + -1.5463809967041016, + -1.5463809967041016, + -1.5463809967041016, + -1.5463809967041016, + -1.5463809967041016, + -1.5463809967041016, + -1.5463809967041016, + -1.5463809967041016, + -1.5463809967041016, + -1.5463809967041016, + -1.5463809967041016, + -1.5463809967041016, + -1.5463809967041016, + -1.5463809967041016, + -1.5463809967041016, + -1.5463809967041016, + -1.5463809967041016, + -1.5463809967041016, + -1.5463809967041016, + -1.5463809967041016, + -1.5463809967041016, + -1.5463809967041016, + -1.5463809967041016, + -1.5463809967041016, + -1.5309171867370606, + -1.4845257568359373, + -1.4381343269348144, + -1.3917428970336914, + -1.3453514671325684, + -1.2989600372314452, + -1.2525686073303222, + -1.2061771774291992, + -1.1597857475280762, + -1.1133943176269532, + -1.06700288772583, + -1.0206114578247067, + -0.9742200279235842, + -0.9278285980224611, + -0.881437168121338, + -0.8350457382202149, + -0.7886543083190918, + -0.7422628784179687, + -0.695871448516846, + -0.6494800186157226, + -0.6030885887145998, + -0.5566971588134764, + -0.5103057289123536, + -0.46391429901123016, + -0.4175228691101074, + -0.3711314392089844, + -0.3247400093078612, + -0.2783485794067382, + -0.2319571495056152, + -0.18556571960449197, + -0.1391742897033692, + -0.09278285980224621, + -0.046391429901122994, + 0.0, + 0.04639142990112309, + 0.09278285980224618, + 0.13917428970336926, + 0.18556571960449236, + 0.23195714950561544, + 0.2783485794067385, + 0.3247400093078613, + 0.371131439208984, + 0.41752286911010744, + 0.4639142990112302, + 0.5103057289123536, + 0.5566971588134764, + 0.6030885887145998, + 0.6494800186157226, + 0.695871448516846, + 0.7422628784179687, + 0.7886543083190921, + 0.8350457382202149, + 0.8814371681213377, + 0.9278285980224604, + 0.9742200279235839, + 1.0206114578247072, + 1.06700288772583, + 1.1133943176269527, + 1.1597857475280762, + 1.2061771774291996, + 1.2525686073303224, + 1.2989600372314452, + 1.3453514671325686, + 1.391742897033692, + 1.4381343269348148, + 1.4845257568359373, + 1.5309171867370601, + 1.5463809967041016, + 1.5463809967041016, + 1.5463809967041016, + 1.5463809967041016, + 1.5463809967041016, + 1.5463809967041016, + 1.5463809967041016, + 1.5463809967041016, + 1.5463809967041016, + 1.5463809967041016, + 1.5463809967041016, + 1.5463809967041016, + 1.5463809967041016, + 1.5463809967041016, + 1.5463809967041016, + 1.5463809967041016, + 1.5463809967041016, + 1.5463809967041016, + 1.5463809967041016, + 1.5463809967041016, + 1.5463809967041016, + 1.5463809967041016, + 1.5463809967041016, + 1.5463809967041016, + 1.5463809967041016, + 1.5463809967041016, + 1.5463809967041016, + 1.5463809967041016 + ], + "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": [ + 0.06451612903225806, + 0.06451612903225806, + 0.06451612903225806, + 0.06451612903225806, + 0.06451612903225806, + 0.06451612903225806, + 0.06451612903225806, + 0.06451612903225806, + 0.06451612903225806, + 0.06451612903225806, + 0.06451612903225806, + 0.06451612903225806, + 0.06451612903225806, + 0.06451612903225806, + 0.06451612903225806, + 0.06451612903225806, + 0.06451612903225806, + 0.06451612903225806, + 0.06451612903225806, + 0.06451612903225806, + 0.06451612903225806, + 0.06451612903225806, + 0.06451612903225806, + 0.06451612903225806, + 0.06451612903225806, + 0.06451612903225806, + 0.06451612903225806, + 0.06451612903225806, + 0.06451612903225806, + 0.06451612903225806, + 0.06451612903225806, + 0.06451612903225806, + 0.06451612903225806, + 0.06451612903225806, + 0.06451612903225806, + 0.06451612903225806, + 0.06451612903225806, + 0.08322580645161293, + 0.10193548387096778, + 0.12064516129032263, + 0.13935483870967738, + 0.15806451612903225, + 0.17677419354838708, + 0.19548387096774195, + 0.2141935483870968, + 0.23290322580645167, + 0.25161290322580643, + 0.27032258064516124, + 0.28903225806451616, + 0.30774193548387097, + 0.32645161290322583, + 0.3451612903225807, + 0.36387096774193545, + 0.3825806451612903, + 0.4012903225806452, + 0.42000000000000004, + 0.4387096774193549, + 0.4574193548387097, + 0.47612903225806447, + 0.49483870967741933, + 0.5135483870967742, + 0.532258064516129, + 0.5509677419354839, + 0.5696774193548388, + 0.5883870967741935, + 0.6070967741935485, + 0.6258064516129034, + 0.6445161290322583, + 0.6632258064516128, + 0.6819354838709677, + 0.7006451612903226, + 0.7193548387096773, + 0.7380645161290322, + 0.7567741935483872, + 0.7754838709677421, + 0.7941935483870968, + 0.8129032258064517, + 0.8316129032258066, + 0.8503225806451613, + 0.8690322580645162, + 0.8877419354838709, + 0.9064516129032258, + 0.9251612903225805, + 0.9438709677419355, + 0.9625806451612904, + 0.9812903225806453, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 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": 31, + "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": [ + 0.3191208839416504, + 0.3191208839416504, + 0.3191208839416504, + 0.3191208839416504, + 0.32492733001708984, + 0.33145958185195923, + 0.33652119636535643, + 0.3408474922180176, + 0.3436393737792969, + 0.3445132374763489, + 0.3459360122680664, + 0.34927997589111326, + 0.35262393951416016, + 0.363059401512146, + 0.3796498775482178, + 0.400412917137146, + 0.4131340980529785, + 0.5402920246124268, + 0.6453938484191895, + 0.6560739278793335, + 0.6599628925323486, + 0.6653448343276978, + 0.6689369678497314, + 0.6718119382858276, + 0.6738979816436768, + 0.6783915758132935, + 0.6841988563537598, + 0.6977783441543579, + 0.7126328945159912, + 0.8430026769638062, + 1.1771290302276611, + 1.182563066482544, + 1.19075608253479, + 1.1983259916305542, + 1.2016658782958984, + 1.2063746452331543, + 1.213541030883789, + 1.21583092212677, + 1.2193219661712646, + 1.221943974494934, + 1.2253408432006836, + 1.2371093034744263, + 1.2435901165008545, + 1.2467049360275269, + 1.2553789615631104, + 1.2627575397491455, + 1.2704319953918457, + 1.2774990797042847, + 1.28053879737854, + 1.2851709127426147, + 1.2871739864349365, + 1.289041519165039, + 1.2896828651428223, + 1.2917089462280273, + 1.2937769889831543, + 1.297810435295105, + 1.2987720966339111, + 1.3005419969558716, + 1.3019602298736572, + 1.3037190437316895, + 1.304184913635254, + 1.3053499460220337, + 1.3061630725860596, + 1.306815505027771, + 1.307631254196167, + 1.308297872543335, + 1.30967116355896, + 1.3108749389648438, + 1.312272071838379, + 1.3129324913024902, + 1.3139278888702393, + 1.3150349855422974, + 1.3155899047851562, + 1.317028522491455, + 1.3179378509521484, + 1.3186566829681396, + 1.3202378749847412, + 1.320967435836792, + 1.3217980861663818, + 1.3228120803833008, + 1.3244881629943848, + 1.3250499963760376, + 1.3255338668823242, + 1.3264724016189575, + 1.3279390335083008, + 1.3283764123916626, + 1.3290090560913086, + 1.3304675817489624, + 1.3320629596710205, + 1.3335751295089722, + 1.3356120586395264, + 1.3372207880020142, + 1.3389902114868164, + 1.3421059846878052, + 1.3440868854522705, + 1.346256971359253, + 1.3509929180145264, + 1.3518489599227905, + 1.3549509048461914, + 1.3592159748077393, + 1.3640799522399902, + 1.3707060813903809, + 1.3779151439666748, + 1.3878369331359863, + 1.39597487449646, + 1.4097869396209717, + 1.4323368072509766, + 1.4465961456298828, + 1.458968162536621, + 1.4640833139419556, + 1.490962028503418, + 1.5037924289703366, + 1.5166228294372552, + 1.5212754726409914, + 1.5235916137695313, + 1.5253555178642273, + 1.526677632331848, + 1.529145872592926, + 1.533906364440918, + 1.5381379127502441, + 1.5381379127502441, + 1.5381379127502441, + 1.5381379127502441 + ], + "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": 32, + "calm": 418 + }, + "regime_transition_counts": { + "burst": { + "burst": 30, + "calm": 2 + }, + "calm": { + "burst": 2, + "calm": 411 + } + }, + "reset_scope": "session", + "schema_version": 12, + "spike_median_multiplier": 1.25, + "spike_threshold_ms_by_worker_slot": { + "0": 129.78225141763687 + }, + "worker_count": 1 +} diff --git a/latency-aware/hopper/small-policy/sample-factory-qwengr00t-rtx3090-v1/profile/latency_distribution.json b/latency-aware/hopper/small-policy/sample-factory-qwengr00t-rtx3090-v1/profile/latency_distribution.json new file mode 100644 index 0000000000000000000000000000000000000000..e4f000fa446c37d93d3ff3300eb5f1749ff030d5 --- /dev/null +++ b/latency-aware/hopper/small-policy/sample-factory-qwengr00t-rtx3090-v1/profile/latency_distribution.json @@ -0,0 +1,1027 @@ +{ + "schema_version": 3, + "worker_slots": { + "0": { + "all": { + "count": 450, + "observation_to_action_latency_ms": { + "distribution_type": "inverse_cdf", + "latency_ms": [ + 100.25659894943237, + 100.25659894943237, + 100.25659894943237, + 100.25659894943237, + 100.42937173843384, + 100.62374112606048, + 100.69315276145934, + 100.70008552074432, + 100.74209804534912, + 100.8279602766037, + 100.90682971477509, + 100.9612243771553, + 101.01561903953552, + 101.23397159576416, + 101.69519209861755, + 101.78508400917053, + 101.98833203315735, + 102.13919639587402, + 102.20834183692932, + 102.35052299499512, + 102.51532506942749, + 103.0292284488678, + 103.20998811721802, + 103.40239250659943, + 104.22255086898804, + 104.35863304138184, + 104.58285403251648, + 104.8021981716156, + 105.0253598690033, + 105.11554050445557, + 105.21885323524475, + 105.49418652057648, + 105.55903506278992, + 105.62058639526367, + 105.69216680526733, + 105.97676956653595, + 106.14571690559387, + 106.40422940254211, + 106.54527497291565, + 106.83702743053436, + 107.07311296463013, + 107.39133262634277, + 109.45088505744934, + 111.85058748722076, + 114.33664631843567, + 115.26488494873047, + 115.66307210922241, + 115.97080898284912, + 116.19734692573547, + 116.43717241287231, + 116.82313919067383, + 117.02803337574005, + 117.19235110282898, + 117.28512907028198, + 117.40299415588379, + 117.48011445999146, + 117.64198112487793, + 117.74017536640167, + 117.82470107078552, + 117.8770592212677, + 117.98120903968811, + 118.08124113082886, + 118.18748998641968, + 118.2358021736145, + 118.37188076972961, + 118.39072239398956, + 118.46283721923828, + 118.5429459810257, + 118.57718801498413, + 118.63366198539734, + 118.70069098472595, + 118.75012600421906, + 118.82725787162781, + 118.85935461521149, + 118.89937210083008, + 118.95050764083862, + 118.98004794120789, + 119.02459692955017, + 119.133309841156, + 119.21458446979523, + 119.34236073493958, + 119.49238348007202, + 119.57314395904541, + 119.65250658988953, + 119.71456408500671, + 119.75180888175964, + 119.82160687446594, + 119.88000154495239, + 119.91864490509033, + 119.98319184780121, + 120.12973380088806, + 120.17876136302948, + 120.24415397644043, + 120.39544105529785, + 120.47922396659851, + 120.54887330532074, + 120.57981014251709, + 120.72355282306671, + 120.79654693603516, + 120.89436054229736, + 121.02687931060791, + 121.11260545253754, + 121.47096109390259, + 121.66697132587433, + 121.82130098342896, + 122.10503101348877, + 122.61133480072021, + 122.7961984872818, + 123.41785788536072, + 123.9133630990982, + 128.59473490715027, + 129.29327311515806, + 129.99181132316585, + 130.90294326543813, + 131.87481627464297, + 133.20605784654617, + 134.82479426860806, + 136.45590093135826, + 138.1117480754853, + 139.58361220359802, + 139.58361220359802, + 139.58361220359802, + 139.58361220359802 + ], + "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": 0.9992948113324016, + "worker_service_time_ms": { + "distribution_type": "inverse_cdf", + "latency_ms": [ + 99.58777499198914, + 99.58777499198914, + 99.58777499198914, + 99.58777499198914, + 99.7652349948883, + 99.96487749814987, + 100.03305070400238, + 100.03548926115036, + 100.07217817306518, + 100.15168002843856, + 100.22633543014527, + 100.28402824401856, + 100.34172105789185, + 100.52799248695374, + 101.02643918991089, + 101.29250264167786, + 101.42182898521423, + 101.531378865242, + 101.63294625282288, + 101.80644500255585, + 102.04786491394043, + 102.30684840679169, + 102.53749585151672, + 102.9086594581604, + 103.08049392700195, + 103.21372640132904, + 103.59838581085205, + 103.7831654548645, + 103.91463613510132, + 104.13139498233795, + 104.21336793899536, + 104.28853642940521, + 104.3345079421997, + 104.41410851478577, + 104.63475108146667, + 104.77252340316772, + 104.9023118019104, + 105.1790360212326, + 105.24173402786255, + 105.71528899669647, + 105.91539907455444, + 106.26289248466492, + 107.9127471446991, + 111.30565190315247, + 112.93897104263306, + 113.9259090423584, + 114.27574896812439, + 114.56841802597046, + 114.76836800575256, + 115.12542915344238, + 115.49577593803406, + 115.78303098678589, + 115.84675121307373, + 115.93612504005432, + 116.10223412513733, + 116.17524790763855, + 116.27443599700928, + 116.41248953342438, + 116.51627397537231, + 116.56277596950531, + 116.65111708641052, + 116.74887442588806, + 116.82825899124146, + 116.90236043930054, + 117.0376329421997, + 117.09950959682465, + 117.17189383506775, + 117.23926544189453, + 117.26328182220459, + 117.34407150745392, + 117.39512014389038, + 117.44214844703674, + 117.52338409423828, + 117.56467795372009, + 117.62358117103577, + 117.66721796989441, + 117.70292377471924, + 117.74845850467682, + 117.81770896911621, + 117.91695594787598, + 118.09892177581787, + 118.18935239315033, + 118.32236886024475, + 118.3527250289917, + 118.40837502479553, + 118.43709409236908, + 118.50040292739868, + 118.56967854499817, + 118.6290500164032, + 118.67505252361298, + 118.80858206748962, + 118.87486100196838, + 118.9444169998169, + 119.06965291500092, + 119.17551016807556, + 119.263507604599, + 119.32206797599792, + 119.40327894687653, + 119.46758484840393, + 119.56757807731628, + 119.70799684524536, + 119.83685052394867, + 120.14590215682983, + 120.3845944404602, + 120.49759197235107, + 120.85645306110382, + 121.28379487991333, + 121.49702835083008, + 122.1178719997406, + 122.6059000492096, + 127.31702494621277, + 128.2782228708267, + 129.23942079544062, + 130.21750211715704, + 131.20040726661685, + 132.41044449806213, + 133.80218739509579, + 135.35875407457343, + 137.2449683189393, + 138.92160320281982, + 138.92160320281982, + 138.92160320281982, + 138.92160320281982 + ], + "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": 418, + "observation_to_action_latency_ms": { + "distribution_type": "inverse_cdf", + "latency_ms": [ + 100.25659894943237, + 100.25659894943237, + 100.25659894943237, + 100.25659894943237, + 100.40172809219361, + 100.58227565670013, + 100.69118077659607, + 100.69762053966522, + 100.70546349334717, + 100.78521996593476, + 100.86497643852233, + 100.92641179323196, + 100.97693839073182, + 101.48090188503265, + 101.76543150901794, + 101.85746025085449, + 102.02494635581971, + 102.18352294921876, + 102.22856862068176, + 102.41133119106293, + 102.68664865493774, + 103.04449622631073, + 103.25476621627807, + 103.52145528793335, + 104.25664330482483, + 104.41195118427277, + 104.62633328437805, + 104.85828572750091, + 105.03275696754456, + 105.11564922332764, + 105.21573969841003, + 105.54319264888764, + 105.56258661270141, + 105.6345480632782, + 105.74041326522827, + 105.9784102678299, + 106.14571690559387, + 106.36495617866517, + 106.54436493873597, + 106.74350165367126, + 107.00653597831726, + 107.19780504703522, + 107.88429025650026, + 110.45301551818847, + 112.888581533432, + 114.8453804397583, + 115.2884108543396, + 115.85090047359466, + 116.15784915924073, + 116.41947458267212, + 116.82206436157226, + 117.01121060848236, + 117.1767498588562, + 117.25987484931946, + 117.37580160140992, + 117.4549824142456, + 117.54047169685364, + 117.69674455165863, + 117.77106837272645, + 117.83091792106629, + 117.9397297668457, + 118.03194200992584, + 118.11757529258728, + 118.19736300468445, + 118.2709280014038, + 118.37591128349304, + 118.40800461769103, + 118.50036959171295, + 118.56493553161621, + 118.63128381729126, + 118.70066559791564, + 118.74803605079651, + 118.83201357841492, + 118.86596692085266, + 118.92433373451233, + 118.9539349937439, + 118.99869818687439, + 119.06108468532562, + 119.17068294525147, + 119.30099086761474, + 119.46926037788391, + 119.57424767017365, + 119.65214509010315, + 119.70749708652497, + 119.74519910812378, + 119.81235805511474, + 119.87276411056519, + 119.90728704452515, + 119.97691723823547, + 120.03986695289612, + 120.17478412628174, + 120.24029574394226, + 120.29351853370667, + 120.42183836936951, + 120.52673645973205, + 120.56319220066071, + 120.67996163368225, + 120.73559628009797, + 120.80071730613709, + 120.90748084545136, + 121.04518411636353, + 121.11846363544464, + 121.44409932136536, + 121.65417563915253, + 121.79435591697693, + 121.98679324150085, + 122.51374492645263, + 122.71021111011505, + 123.00279578208924, + 123.46635902881623, + 124.38282137870787, + 124.87906891489028, + 125.39102749538422, + 125.92937257337572, + 126.46771765136721, + 127.3458736467362, + 128.23066009902948, + 128.9766024608612, + 129.625466840744, + 130.14704203605652, + 130.14704203605652, + 130.14704203605652, + 130.14704203605652 + ], + "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": 0.999272450580491, + "worker_service_time_ms": { + "distribution_type": "inverse_cdf", + "latency_ms": [ + 99.58777499198914, + 99.58777499198914, + 99.58777499198914, + 99.58777499198914, + 99.73684139442445, + 99.92228709745407, + 100.03235706996918, + 100.03462221860886, + 100.03825738143921, + 100.1121057715416, + 100.18595416164398, + 100.24710484313965, + 100.30069505691529, + 100.82506213665009, + 101.1387021446228, + 101.33871601581573, + 101.44854555130004, + 101.54180234909057, + 101.64674322128296, + 101.82540905475616, + 102.0620273399353, + 102.33945229053498, + 102.5968849658966, + 102.97159505844117, + 103.09481094360352, + 103.23617889404296, + 103.65310387611389, + 103.80324698925018, + 103.91589998245239, + 104.18816221237182, + 104.2256536769867, + 104.31202034950256, + 104.36575504302978, + 104.46129910945892, + 104.64864039421082, + 104.78796194076538, + 104.9023118019104, + 105.17897140026092, + 105.22289293289185, + 105.64354291915893, + 105.86738454818726, + 106.08066976070404, + 107.26280077934265, + 109.6323651075363, + 111.88637012481689, + 113.52936239242554, + 113.9621356010437, + 114.51033910274505, + 114.74266494750977, + 115.06721992015838, + 115.47237975120545, + 115.72886090278625, + 115.84143285751342, + 115.89359952926635, + 116.01603100776673, + 116.14374716758728, + 116.20214700698853, + 116.3483362865448, + 116.45014364242553, + 116.52146469593048, + 116.60253409385682, + 116.67599904537201, + 116.7781213092804, + 116.87534006595611, + 116.92153003692627, + 117.05918686389923, + 117.13113899230957, + 117.17549333572387, + 117.25790552139283, + 117.31309947013855, + 117.39715318679809, + 117.43717951774597, + 117.52794399261475, + 117.56680624961854, + 117.62714520454406, + 117.66932479381562, + 117.7031861782074, + 117.76872053146363, + 117.85444743156432, + 118.01769463062287, + 118.16539481163025, + 118.3232766866684, + 118.35119206428527, + 118.39980568885804, + 118.42484113693237, + 118.47398302555084, + 118.56194305419922, + 118.5902167224884, + 118.65561540603638, + 118.72641350269318, + 118.86620130538941, + 118.93647229671478, + 119.00474523544312, + 119.11499736785889, + 119.25242570877076, + 119.2844930934906, + 119.3654022693634, + 119.4202631855011, + 119.48397501945496, + 119.57837578773498, + 119.73762118339539, + 119.83856031894683, + 120.12005149841309, + 120.35054317474365, + 120.49533098220826, + 120.73445631980896, + 121.24480094909667, + 121.37678823471069, + 121.68138793945312, + 122.21880384922028, + 123.27626915931701, + 124.04111554765701, + 124.62335801601411, + 124.8989195418358, + 125.17448106765748, + 126.05603566169744, + 126.94941451072687, + 127.84247981166835, + 128.73532588386533, + 129.45302033424377, + 129.45302033424377, + 129.45302033424377, + 129.45302033424377 + ], + "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/hopper/small-policy/sample-factory-qwengr00t-rtx3090-v1/profile/latency_profile.png b/latency-aware/hopper/small-policy/sample-factory-qwengr00t-rtx3090-v1/profile/latency_profile.png new file mode 100644 index 0000000000000000000000000000000000000000..cd72217ae74c3000128b81f3e549d43b5819e869 Binary files /dev/null and b/latency-aware/hopper/small-policy/sample-factory-qwengr00t-rtx3090-v1/profile/latency_profile.png differ diff --git a/latency-aware/hopper/small-policy/sample-factory-qwengr00t-rtx3090-v1/profile/profile.json b/latency-aware/hopper/small-policy/sample-factory-qwengr00t-rtx3090-v1/profile/profile.json new file mode 100644 index 0000000000000000000000000000000000000000..cf5d7c83c160fea8ef7bc83706a877e0e32a9884 --- /dev/null +++ b/latency-aware/hopper/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": 450, + "n_capacity_drops": 0, + "n_observation_attempts": 450, + "per_slot_summary": { + "0": { + "admitted_count": 450, + "mean_observation_to_action_latency_ms": 114.61718838108911, + "mean_worker_service_time_ms": 113.43859605047437, + "p95_observation_to_action_latency_ms": 122.60570173263551, + "p95_worker_service_time_ms": 121.27957479953767, + "p99_worker_service_time_ms": 126.26976265907285 + } + }, + "provenance": { + "base_config": "/workspace/lzj/latency-sensitive-bench/runs/hopper/gr00t_h1_5fps_20260917/P.yaml", + "checkpoint_kind": "latest", + "model_artifact": { + "action_horizon": 1, + "checkpoint_sha256": "5ec4a5fbdfb7fac819a51b523f3735c6702b90a2440a2957c5cc699941ffbf1e", + "clock_fps": 5, + "hf_prefix": "hopper/zero_latency/GR00T", + "hf_repo": "latency-sensitive-bench/Standard-Pipeline", + "hf_revision": "50e51e28c1fc7f55159eba4c611857b2df56539b", + "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": "20260917T141828786022Z", + "summary": { + "frame_ms": 200.0, + "max_ms": 139.58361220359802, + "mean_effective_frames": 0.5730859419054455, + "mean_ms": 114.61718838108911, + "min_ms": 100.25659894943237, + "n_samples": 450, + "p50_frames": 0.5904062056541443, + "p50_ms": 118.08124113082886, + "p90_frames": 0.6055044454336166, + "p90_ms": 121.10088908672333, + "p95_frames": 0.6130285086631775, + "p95_ms": 122.6057017326355, + "p99_frames": 0.6377877252340316, + "p99_ms": 127.55754504680631, + "prob_latency_gt_1_frame": 0.0, + "prob_latency_gt_2_frames": 0.0, + "prob_latency_gt_3_frames": 0.0, + "std_ms": 7.419009630839383 + }, + "visualization_path": "latency_profile.png", + "workload_id": "hopper" +} diff --git a/latency-aware/hopper/small-policy/sample-factory-qwengr00t-rtx3090-v1/provenance.json b/latency-aware/hopper/small-policy/sample-factory-qwengr00t-rtx3090-v1/provenance.json new file mode 100644 index 0000000000000000000000000000000000000000..469b81bf2155a6b55274493a95c8fdfb53ce6519 --- /dev/null +++ b/latency-aware/hopper/small-policy/sample-factory-qwengr00t-rtx3090-v1/provenance.json @@ -0,0 +1,44 @@ +{ + "task": "hopper", + "model": "sample-factory-appo", + "training_condition": "latency-aware", + "training_run_id": "hopper_profile_appo_h1_5fps_20260917", + "source": { + "repo_id": "latency-sensitive-bench/Standard-Pipeline", + "revision": "2208875f92b21e3f8ec2363a30a6705dfe56cc04", + "path": "hopper/profile_latency/small_model/GR00T", + "url": "https://huggingface.co/latency-sensitive-bench/Standard-Pipeline/tree/2208875f92b21e3f8ec2363a30a6705dfe56cc04/hopper/profile_latency/small_model/GR00T" + }, + "hf_repo_id": "latency-sensitive-bench/benchmark-models", + "hf_path": "latency-aware/hopper/small-policy/sample-factory-qwengr00t-rtx3090-v1", + "checkpoint": { + "source_file": "hopper/profile_latency/small_model/GR00T/checkpoints/selected.pth", + "source_sha256": "f8b6089e67f129a7c17504bc8d6146003177b260e927cc1281a178d26f404b74", + "source_bytes": 76933, + "file": "checkpoint.pth", + "selection_rule": "training_best", + "method": "inference_export", + "sha256": "3dc620932950a33e730f0ba615f2a85dded0c11dd24d0d27ff08329090bbc9cc", + "bytes": 27445, + "train_step": 19512, + "env_steps": 9990144, + "tensor_count": 15, + "tensor_equality_verified": true, + "removed_fields": [ + "best_performance", + "curr_lr", + "optimizer" + ] + }, + "config_source": "hopper/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/hopper/small-policy/sample-factory-qwengr00t-rtx3090-v1/selection.json b/latency-aware/hopper/small-policy/sample-factory-qwengr00t-rtx3090-v1/selection.json new file mode 100644 index 0000000000000000000000000000000000000000..17ed7b6de5c48d8931c88dd722f11e37a63da8f4 --- /dev/null +++ b/latency-aware/hopper/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/hopper/profile_latency/teacher/checkpoints/hopper_profile_appo_h1_5fps_20260917/checkpoint_p0/checkpoint_000019536_10002432.pth", + "sha256": "d7bccf0797e5622069c1af0aa67796863bd68c40707cd96c7e7399fa27ca49e0", + "episodes": 20, + "returns": [ + 3813.902369620451, + 3814.717929846383, + 3815.678995110681, + 3813.5158104903485, + 3812.1721732868205, + 3822.869610539937, + 3816.213698287465, + 3819.3530723476006, + 3809.7239266677057, + 3819.2709564714014, + 3804.324024566971, + 3818.1522276093633, + 3829.262769910777, + 3815.7688288666577, + 3814.218115130112, + 3802.188637269843, + 3814.133230818948, + 3811.2177504405886, + 3803.348027121752, + 3819.7451215416377 + ], + "mean": 3814.488863797272, + "std": 6.329159120614397, + "strict_gt3000": 20 + }, + "training_best": { + "checkpoint": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/hopper/profile_latency/teacher/checkpoints/hopper_profile_appo_h1_5fps_20260917/checkpoint_p0/best_000019512_9990144_reward_3220.937.pth", + "sha256": "f8b6089e67f129a7c17504bc8d6146003177b260e927cc1281a178d26f404b74", + "episodes": 20, + "returns": [ + 3818.3299991080344, + 3801.949979062483, + 3799.39904827517, + 3806.0984441131777, + 3821.3468930346507, + 3818.6013992377857, + 3825.5948641715036, + 3817.364507562392, + 3814.7261006971084, + 3825.008281517956, + 3816.453969095774, + 3815.17650963725, + 3816.6675065195645, + 3810.7786398491785, + 3813.106914295106, + 3803.3492543466577, + 3815.0139371581436, + 3819.671617704402, + 3825.458467648868, + 3825.03671743075 + ], + "mean": 3815.4566525232976, + "std": 7.632802007802108, + "strict_gt3000": 20 + } + }, + "completed_at": "2026-09-17T14:44:34.549937+00:00" +} diff --git a/latency-aware/hopper/small-policy/sample-factory-qwengr00t-rtx3090-v1/teacher-preparation.json b/latency-aware/hopper/small-policy/sample-factory-qwengr00t-rtx3090-v1/teacher-preparation.json new file mode 100644 index 0000000000000000000000000000000000000000..412384090c4eb555a2563e65185e8e2e2c2ff630 --- /dev/null +++ b/latency-aware/hopper/small-policy/sample-factory-qwengr00t-rtx3090-v1/teacher-preparation.json @@ -0,0 +1,15 @@ +{ + "state": "PROFILE_TEACHER_CONFIG_READY", + "profile": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/hopper/profile_latency/profile_3090_20260917T141828786022Z/profile.json", + "profile_sha256": "9d2e175b36914cc07fe93711db34ea89593c6c71e2b2a3d7fbe9176173ab7184", + "source_recipe": "/home/ubuntu/lzj/latency-sensitive-bench/configs/examples/gymnasium/hopper/small_model_train.yaml", + "source_recipe_sha256": "2e22333ef7e8a77e175654f9952e5c22f03af26590fb6d5bc543c2d399b63350", + "config": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/hopper/profile_latency/teacher.yaml", + "algo": "APPO", + "budget_env_steps": 10000000, + "seed": 3333, + "fps": 5, + "latency": "iid", + "initialization": "fresh", + "next": "Run only after immutable new P is verified; Final/best20 tieFinal, E10, bounded12probe, strictreturn>3000." +} diff --git a/latency-aware/hopper/small-policy/sample-factory-qwengr00t-rtx3090-v1/validation/E10-summary.json b/latency-aware/hopper/small-policy/sample-factory-qwengr00t-rtx3090-v1/validation/E10-summary.json new file mode 100644 index 0000000000000000000000000000000000000000..194de73e020723518fa3877cf025c018dea79c1d --- /dev/null +++ b/latency-aware/hopper/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/hopper/profile_latency/teacher/checkpoints/hopper_profile_appo_h1_5fps_20260917/checkpoint_p0/best_000019512_9990144_reward_3220.937.pth", + "sha256": "f8b6089e67f129a7c17504bc8d6146003177b260e927cc1281a178d26f404b74", + "episodes": 10, + "returns": [ + 3818.3299991080344, + 3801.949979062483, + 3799.39904827517, + 3806.0984441131777, + 3821.3468930346507, + 3818.6013992377857, + 3825.5948641715036, + 3817.364507562392, + 3814.7261006971084, + 3825.008281517956 + ], + "mean": 3814.8419516780264, + "std": 8.80400721259514, + "strict_gt3000": 10 +} diff --git a/latency-aware/hopper/small-policy/sample-factory-qwengr00t-rtx3090-v1/validation/E10/episode_metrics.jsonl b/latency-aware/hopper/small-policy/sample-factory-qwengr00t-rtx3090-v1/validation/E10/episode_metrics.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..4dee0880e1001073d516a9a6a51e4759533defd8 --- /dev/null +++ b/latency-aware/hopper/small-policy/sample-factory-qwengr00t-rtx3090-v1/validation/E10/episode_metrics.jsonl @@ -0,0 +1,10 @@ +{"episode_id": 0, "episode_return": 3818.3299991080344, "episode_return_env": 3818.3299991080344, "game_score": null, "mean_latency_ms": 115.18995074102081, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/hopper/profile_latency/teacher/checkpoints/hopper_profile_appo_h1_5fps_20260917/checkpoint_p0/best_000019512_9990144_reward_3220.937.pth", "config_name": "hopper_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Hopper-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/hopper/profile_latency/E10", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "hopper_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/hopper/profile_latency/profile_3090_20260917T141828786022Z/profile.json", "source_run_id": "20260917T141828786022Z", "submitted_observation_frames": 1000, "workload_id": "hopper"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 121.69597162426244, "p99_latency_ms": 129.89549037880013, "return_raw": null, "survival_steps": 1000} +{"episode_id": 1, "episode_return": 3801.949979062483, "episode_return_env": 3801.949979062483, "game_score": null, "mean_latency_ms": 114.56449403686167, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/hopper/profile_latency/teacher/checkpoints/hopper_profile_appo_h1_5fps_20260917/checkpoint_p0/best_000019512_9990144_reward_3220.937.pth", "config_name": "hopper_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Hopper-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/hopper/profile_latency/E10", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "hopper_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/hopper/profile_latency/profile_3090_20260917T141828786022Z/profile.json", "source_run_id": "20260917T141828786022Z", "submitted_observation_frames": 1000, "workload_id": "hopper"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 121.50491640436555, "p99_latency_ms": 125.97119040661914, "return_raw": null, "survival_steps": 1000} +{"episode_id": 2, "episode_return": 3799.39904827517, "episode_return_env": 3799.39904827517, "game_score": null, "mean_latency_ms": 114.43209582448789, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/hopper/profile_latency/teacher/checkpoints/hopper_profile_appo_h1_5fps_20260917/checkpoint_p0/best_000019512_9990144_reward_3220.937.pth", "config_name": "hopper_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Hopper-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/hopper/profile_latency/E10", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "hopper_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/hopper/profile_latency/profile_3090_20260917T141828786022Z/profile.json", "source_run_id": "20260917T141828786022Z", "submitted_observation_frames": 1000, "workload_id": "hopper"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 121.02486631581769, "p99_latency_ms": 127.06192303846402, "return_raw": null, "survival_steps": 1000} +{"episode_id": 3, "episode_return": 3806.0984441131777, "episode_return_env": 3806.0984441131777, "game_score": null, "mean_latency_ms": 114.84414803452853, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/hopper/profile_latency/teacher/checkpoints/hopper_profile_appo_h1_5fps_20260917/checkpoint_p0/best_000019512_9990144_reward_3220.937.pth", "config_name": "hopper_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Hopper-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/hopper/profile_latency/E10", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "hopper_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/hopper/profile_latency/profile_3090_20260917T141828786022Z/profile.json", "source_run_id": "20260917T141828786022Z", "submitted_observation_frames": 1000, "workload_id": "hopper"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 121.02132326348054, "p99_latency_ms": 127.9190625902109, "return_raw": null, "survival_steps": 1000} +{"episode_id": 4, "episode_return": 3821.3468930346507, "episode_return_env": 3821.3468930346507, "game_score": null, "mean_latency_ms": 114.94717422490875, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/hopper/profile_latency/teacher/checkpoints/hopper_profile_appo_h1_5fps_20260917/checkpoint_p0/best_000019512_9990144_reward_3220.937.pth", "config_name": "hopper_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Hopper-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/hopper/profile_latency/E10", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "hopper_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/hopper/profile_latency/profile_3090_20260917T141828786022Z/profile.json", "source_run_id": "20260917T141828786022Z", "submitted_observation_frames": 1000, "workload_id": "hopper"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 121.00997960349433, "p99_latency_ms": 128.21732582655437, "return_raw": null, "survival_steps": 1000} +{"episode_id": 5, "episode_return": 3818.6013992377857, "episode_return_env": 3818.6013992377857, "game_score": null, "mean_latency_ms": 114.60245796451443, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/hopper/profile_latency/teacher/checkpoints/hopper_profile_appo_h1_5fps_20260917/checkpoint_p0/best_000019512_9990144_reward_3220.937.pth", "config_name": "hopper_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Hopper-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/hopper/profile_latency/E10", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "hopper_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/hopper/profile_latency/profile_3090_20260917T141828786022Z/profile.json", "source_run_id": "20260917T141828786022Z", "submitted_observation_frames": 1000, "workload_id": "hopper"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 120.87727201786771, "p99_latency_ms": 127.45375690975183, "return_raw": null, "survival_steps": 1000} +{"episode_id": 6, "episode_return": 3825.5948641715036, "episode_return_env": 3825.5948641715036, "game_score": null, "mean_latency_ms": 114.71892566828849, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/hopper/profile_latency/teacher/checkpoints/hopper_profile_appo_h1_5fps_20260917/checkpoint_p0/best_000019512_9990144_reward_3220.937.pth", "config_name": "hopper_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Hopper-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/hopper/profile_latency/E10", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "hopper_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/hopper/profile_latency/profile_3090_20260917T141828786022Z/profile.json", "source_run_id": "20260917T141828786022Z", "submitted_observation_frames": 1000, "workload_id": "hopper"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 121.09108318462013, "p99_latency_ms": 127.16757159428731, "return_raw": null, "survival_steps": 1000} +{"episode_id": 7, "episode_return": 3817.364507562392, "episode_return_env": 3817.364507562392, "game_score": null, "mean_latency_ms": 114.78291074661247, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/hopper/profile_latency/teacher/checkpoints/hopper_profile_appo_h1_5fps_20260917/checkpoint_p0/best_000019512_9990144_reward_3220.937.pth", "config_name": "hopper_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Hopper-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/hopper/profile_latency/E10", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "hopper_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/hopper/profile_latency/profile_3090_20260917T141828786022Z/profile.json", "source_run_id": "20260917T141828786022Z", "submitted_observation_frames": 1000, "workload_id": "hopper"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 121.54565631351461, "p99_latency_ms": 132.3236564844519, "return_raw": null, "survival_steps": 1000} +{"episode_id": 8, "episode_return": 3814.7261006971084, "episode_return_env": 3814.7261006971084, "game_score": null, "mean_latency_ms": 114.76045460737626, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/hopper/profile_latency/teacher/checkpoints/hopper_profile_appo_h1_5fps_20260917/checkpoint_p0/best_000019512_9990144_reward_3220.937.pth", "config_name": "hopper_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Hopper-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/hopper/profile_latency/E10", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "hopper_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/hopper/profile_latency/profile_3090_20260917T141828786022Z/profile.json", "source_run_id": "20260917T141828786022Z", "submitted_observation_frames": 1000, "workload_id": "hopper"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 121.3125299720794, "p99_latency_ms": 130.23741427327107, "return_raw": null, "survival_steps": 1000} +{"episode_id": 9, "episode_return": 3825.008281517956, "episode_return_env": 3825.008281517956, "game_score": null, "mean_latency_ms": 114.36901179788543, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/hopper/profile_latency/teacher/checkpoints/hopper_profile_appo_h1_5fps_20260917/checkpoint_p0/best_000019512_9990144_reward_3220.937.pth", "config_name": "hopper_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Hopper-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/hopper/profile_latency/E10", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "hopper_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/hopper/profile_latency/profile_3090_20260917T141828786022Z/profile.json", "source_run_id": "20260917T141828786022Z", "submitted_observation_frames": 1000, "workload_id": "hopper"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 121.07475619002756, "p99_latency_ms": 126.98905492821542, "return_raw": null, "survival_steps": 1000} diff --git a/latency-aware/hopper/small-policy/sample-factory-qwengr00t-rtx3090-v1/validation/probe_state.json b/latency-aware/hopper/small-policy/sample-factory-qwengr00t-rtx3090-v1/validation/probe_state.json new file mode 100644 index 0000000000000000000000000000000000000000..4377939712c3d509f76e6622beb17766f1ac195d --- /dev/null +++ b/latency-aware/hopper/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": 3814.5019599199295, + "seed": 3333, + "split": "train" + }, + { + "attempt_idx": 1, + "episode_idx": 1, + "episode_length": 1000, + "episode_raw_frames": 1000, + "episode_raw_return": 3808.423024535179, + "seed": 3334, + "split": "train" + }, + { + "attempt_idx": 2, + "episode_idx": 2, + "episode_length": 1000, + "episode_raw_frames": 1000, + "episode_raw_return": 3820.5545291900635, + "seed": 3335, + "split": "train" + }, + { + "attempt_idx": 3, + "episode_idx": 3, + "episode_length": 1000, + "episode_raw_frames": 1000, + "episode_raw_return": 3816.96435379982, + "seed": 3336, + "split": "train" + }, + { + "attempt_idx": 4, + "episode_idx": 4, + "episode_length": 1000, + "episode_raw_frames": 1000, + "episode_raw_return": 3823.0939115285873, + "seed": 3337, + "split": "train" + }, + { + "attempt_idx": 5, + "episode_idx": 5, + "episode_length": 1000, + "episode_raw_frames": 1000, + "episode_raw_return": 3824.567959845066, + "seed": 3338, + "split": "train" + }, + { + "attempt_idx": 6, + "episode_idx": 6, + "episode_length": 1000, + "episode_raw_frames": 1000, + "episode_raw_return": 3819.535971403122, + "seed": 3339, + "split": "train" + }, + { + "attempt_idx": 7, + "episode_idx": 7, + "episode_length": 1000, + "episode_raw_frames": 1000, + "episode_raw_return": 3822.3822045326233, + "seed": 3340, + "split": "train" + }, + { + "attempt_idx": 8, + "episode_idx": 8, + "episode_length": 1000, + "episode_raw_frames": 1000, + "episode_raw_return": 3794.1470665335655, + "seed": 3341, + "split": "train" + }, + { + "attempt_idx": 9, + "episode_idx": 9, + "episode_length": 1000, + "episode_raw_frames": 1000, + "episode_raw_return": 3814.092240333557, + "seed": 3342, + "split": "train" + }, + { + "attempt_idx": 10, + "episode_idx": 10, + "episode_length": 1000, + "episode_raw_frames": 1000, + "episode_raw_return": 3816.7141597270966, + "seed": 3343, + "split": "val" + }, + { + "attempt_idx": 11, + "episode_idx": 11, + "episode_length": 1000, + "episode_raw_frames": 1000, + "episode_raw_return": 3816.535435438156, + "seed": 3344, + "split": "train" + } + ], + "attempted_episodes": 12, + "next_attempt_idx": 12, + "rejected_episodes": 0, + "runtime_metadata": { + "checkpoint_train_step": 19512, + "env_fps": 5.0, + "frame_stack": 1, + "obs_fps": 5.0 + } +} diff --git a/latency-aware/hopper/small-policy/sample-factory-qwengr00t-rtx3090-v1/validation/selection_final/episode_metrics.jsonl b/latency-aware/hopper/small-policy/sample-factory-qwengr00t-rtx3090-v1/validation/selection_final/episode_metrics.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..363c9f1416faa4641f6290a73d5299f485d794e2 --- /dev/null +++ b/latency-aware/hopper/small-policy/sample-factory-qwengr00t-rtx3090-v1/validation/selection_final/episode_metrics.jsonl @@ -0,0 +1,20 @@ +{"episode_id": 0, "episode_return": 3813.902369620451, "episode_return_env": 3813.902369620451, "game_score": null, "mean_latency_ms": 115.18995074102081, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/hopper/profile_latency/teacher/checkpoints/hopper_profile_appo_h1_5fps_20260917/checkpoint_p0/checkpoint_000019536_10002432.pth", "config_name": "hopper_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Hopper-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/hopper/profile_latency/selection_final", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "hopper_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/hopper/profile_latency/profile_3090_20260917T141828786022Z/profile.json", "source_run_id": "20260917T141828786022Z", "submitted_observation_frames": 1000, "workload_id": "hopper"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 121.69597162426244, "p99_latency_ms": 129.89549037880013, "return_raw": null, "survival_steps": 1000} +{"episode_id": 1, "episode_return": 3814.717929846383, "episode_return_env": 3814.717929846383, "game_score": null, "mean_latency_ms": 114.56449403686167, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/hopper/profile_latency/teacher/checkpoints/hopper_profile_appo_h1_5fps_20260917/checkpoint_p0/checkpoint_000019536_10002432.pth", "config_name": "hopper_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Hopper-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/hopper/profile_latency/selection_final", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "hopper_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/hopper/profile_latency/profile_3090_20260917T141828786022Z/profile.json", "source_run_id": "20260917T141828786022Z", "submitted_observation_frames": 1000, "workload_id": "hopper"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 121.50491640436555, "p99_latency_ms": 125.97119040661914, "return_raw": null, "survival_steps": 1000} +{"episode_id": 2, "episode_return": 3815.678995110681, "episode_return_env": 3815.678995110681, "game_score": null, "mean_latency_ms": 114.43209582448789, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/hopper/profile_latency/teacher/checkpoints/hopper_profile_appo_h1_5fps_20260917/checkpoint_p0/checkpoint_000019536_10002432.pth", "config_name": "hopper_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Hopper-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/hopper/profile_latency/selection_final", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "hopper_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/hopper/profile_latency/profile_3090_20260917T141828786022Z/profile.json", "source_run_id": "20260917T141828786022Z", "submitted_observation_frames": 1000, "workload_id": "hopper"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 121.02486631581769, "p99_latency_ms": 127.06192303846402, "return_raw": null, "survival_steps": 1000} +{"episode_id": 3, "episode_return": 3813.5158104903485, "episode_return_env": 3813.5158104903485, "game_score": null, "mean_latency_ms": 114.84414803452853, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/hopper/profile_latency/teacher/checkpoints/hopper_profile_appo_h1_5fps_20260917/checkpoint_p0/checkpoint_000019536_10002432.pth", "config_name": "hopper_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Hopper-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/hopper/profile_latency/selection_final", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "hopper_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/hopper/profile_latency/profile_3090_20260917T141828786022Z/profile.json", "source_run_id": "20260917T141828786022Z", "submitted_observation_frames": 1000, "workload_id": "hopper"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 121.02132326348054, "p99_latency_ms": 127.9190625902109, "return_raw": null, "survival_steps": 1000} +{"episode_id": 4, "episode_return": 3812.1721732868205, "episode_return_env": 3812.1721732868205, "game_score": null, "mean_latency_ms": 114.94717422490875, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/hopper/profile_latency/teacher/checkpoints/hopper_profile_appo_h1_5fps_20260917/checkpoint_p0/checkpoint_000019536_10002432.pth", "config_name": "hopper_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Hopper-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/hopper/profile_latency/selection_final", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "hopper_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/hopper/profile_latency/profile_3090_20260917T141828786022Z/profile.json", "source_run_id": "20260917T141828786022Z", "submitted_observation_frames": 1000, "workload_id": "hopper"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 121.00997960349433, "p99_latency_ms": 128.21732582655437, "return_raw": null, "survival_steps": 1000} +{"episode_id": 5, "episode_return": 3822.869610539937, "episode_return_env": 3822.869610539937, "game_score": null, "mean_latency_ms": 114.60245796451443, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/hopper/profile_latency/teacher/checkpoints/hopper_profile_appo_h1_5fps_20260917/checkpoint_p0/checkpoint_000019536_10002432.pth", "config_name": "hopper_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Hopper-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/hopper/profile_latency/selection_final", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "hopper_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/hopper/profile_latency/profile_3090_20260917T141828786022Z/profile.json", "source_run_id": "20260917T141828786022Z", "submitted_observation_frames": 1000, "workload_id": "hopper"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 120.87727201786771, "p99_latency_ms": 127.45375690975183, "return_raw": null, "survival_steps": 1000} +{"episode_id": 6, "episode_return": 3816.213698287465, "episode_return_env": 3816.213698287465, "game_score": null, "mean_latency_ms": 114.71892566828849, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/hopper/profile_latency/teacher/checkpoints/hopper_profile_appo_h1_5fps_20260917/checkpoint_p0/checkpoint_000019536_10002432.pth", "config_name": "hopper_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Hopper-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/hopper/profile_latency/selection_final", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "hopper_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/hopper/profile_latency/profile_3090_20260917T141828786022Z/profile.json", "source_run_id": "20260917T141828786022Z", "submitted_observation_frames": 1000, "workload_id": "hopper"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 121.09108318462013, "p99_latency_ms": 127.16757159428731, "return_raw": null, "survival_steps": 1000} +{"episode_id": 7, "episode_return": 3819.3530723476006, "episode_return_env": 3819.3530723476006, "game_score": null, "mean_latency_ms": 114.78291074661247, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/hopper/profile_latency/teacher/checkpoints/hopper_profile_appo_h1_5fps_20260917/checkpoint_p0/checkpoint_000019536_10002432.pth", "config_name": "hopper_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Hopper-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/hopper/profile_latency/selection_final", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "hopper_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/hopper/profile_latency/profile_3090_20260917T141828786022Z/profile.json", "source_run_id": "20260917T141828786022Z", "submitted_observation_frames": 1000, "workload_id": "hopper"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 121.54565631351461, "p99_latency_ms": 132.3236564844519, "return_raw": null, "survival_steps": 1000} +{"episode_id": 8, "episode_return": 3809.7239266677057, "episode_return_env": 3809.7239266677057, "game_score": null, "mean_latency_ms": 114.76045460737626, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/hopper/profile_latency/teacher/checkpoints/hopper_profile_appo_h1_5fps_20260917/checkpoint_p0/checkpoint_000019536_10002432.pth", "config_name": "hopper_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Hopper-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/hopper/profile_latency/selection_final", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "hopper_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/hopper/profile_latency/profile_3090_20260917T141828786022Z/profile.json", "source_run_id": "20260917T141828786022Z", "submitted_observation_frames": 1000, "workload_id": "hopper"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 121.3125299720794, "p99_latency_ms": 130.23741427327107, "return_raw": null, "survival_steps": 1000} +{"episode_id": 9, "episode_return": 3819.2709564714014, "episode_return_env": 3819.2709564714014, "game_score": null, "mean_latency_ms": 114.36901179788543, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/hopper/profile_latency/teacher/checkpoints/hopper_profile_appo_h1_5fps_20260917/checkpoint_p0/checkpoint_000019536_10002432.pth", "config_name": "hopper_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Hopper-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/hopper/profile_latency/selection_final", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "hopper_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/hopper/profile_latency/profile_3090_20260917T141828786022Z/profile.json", "source_run_id": "20260917T141828786022Z", "submitted_observation_frames": 1000, "workload_id": "hopper"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 121.07475619002756, "p99_latency_ms": 126.98905492821542, "return_raw": null, "survival_steps": 1000} +{"episode_id": 10, "episode_return": 3804.324024566971, "episode_return_env": 3804.324024566971, "game_score": null, "mean_latency_ms": 114.78648178340455, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/hopper/profile_latency/teacher/checkpoints/hopper_profile_appo_h1_5fps_20260917/checkpoint_p0/checkpoint_000019536_10002432.pth", "config_name": "hopper_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Hopper-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/hopper/profile_latency/selection_final", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "hopper_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/hopper/profile_latency/profile_3090_20260917T141828786022Z/profile.json", "source_run_id": "20260917T141828786022Z", "submitted_observation_frames": 1000, "workload_id": "hopper"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 121.04216611012974, "p99_latency_ms": 128.9612933580124, "return_raw": null, "survival_steps": 1000} +{"episode_id": 11, "episode_return": 3818.1522276093633, "episode_return_env": 3818.1522276093633, "game_score": null, "mean_latency_ms": 114.8693603353378, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/hopper/profile_latency/teacher/checkpoints/hopper_profile_appo_h1_5fps_20260917/checkpoint_p0/checkpoint_000019536_10002432.pth", "config_name": "hopper_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Hopper-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/hopper/profile_latency/selection_final", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "hopper_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/hopper/profile_latency/profile_3090_20260917T141828786022Z/profile.json", "source_run_id": "20260917T141828786022Z", "submitted_observation_frames": 1000, "workload_id": "hopper"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 121.15181791640279, "p99_latency_ms": 126.56114195853236, "return_raw": null, "survival_steps": 1000} +{"episode_id": 12, "episode_return": 3829.262769910777, "episode_return_env": 3829.262769910777, "game_score": null, "mean_latency_ms": 114.56549116106042, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/hopper/profile_latency/teacher/checkpoints/hopper_profile_appo_h1_5fps_20260917/checkpoint_p0/checkpoint_000019536_10002432.pth", "config_name": "hopper_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Hopper-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/hopper/profile_latency/selection_final", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "hopper_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/hopper/profile_latency/profile_3090_20260917T141828786022Z/profile.json", "source_run_id": "20260917T141828786022Z", "submitted_observation_frames": 1000, "workload_id": "hopper"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 120.99124429827873, "p99_latency_ms": 129.0509902417293, "return_raw": null, "survival_steps": 1000} +{"episode_id": 13, "episode_return": 3815.7688288666577, "episode_return_env": 3815.7688288666577, "game_score": null, "mean_latency_ms": 114.7911034734444, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/hopper/profile_latency/teacher/checkpoints/hopper_profile_appo_h1_5fps_20260917/checkpoint_p0/checkpoint_000019536_10002432.pth", "config_name": "hopper_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Hopper-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/hopper/profile_latency/selection_final", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "hopper_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/hopper/profile_latency/profile_3090_20260917T141828786022Z/profile.json", "source_run_id": "20260917T141828786022Z", "submitted_observation_frames": 1000, "workload_id": "hopper"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 121.18068730074646, "p99_latency_ms": 125.48262594291727, "return_raw": null, "survival_steps": 1000} +{"episode_id": 14, "episode_return": 3814.218115130112, "episode_return_env": 3814.218115130112, "game_score": null, "mean_latency_ms": 114.65458686417112, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/hopper/profile_latency/teacher/checkpoints/hopper_profile_appo_h1_5fps_20260917/checkpoint_p0/checkpoint_000019536_10002432.pth", "config_name": "hopper_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Hopper-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/hopper/profile_latency/selection_final", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "hopper_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/hopper/profile_latency/profile_3090_20260917T141828786022Z/profile.json", "source_run_id": "20260917T141828786022Z", "submitted_observation_frames": 1000, "workload_id": "hopper"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 121.22762011132701, "p99_latency_ms": 128.9128408954642, "return_raw": null, "survival_steps": 1000} +{"episode_id": 15, "episode_return": 3802.188637269843, "episode_return_env": 3802.188637269843, "game_score": null, "mean_latency_ms": 114.66081796963667, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/hopper/profile_latency/teacher/checkpoints/hopper_profile_appo_h1_5fps_20260917/checkpoint_p0/checkpoint_000019536_10002432.pth", "config_name": "hopper_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Hopper-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/hopper/profile_latency/selection_final", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "hopper_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/hopper/profile_latency/profile_3090_20260917T141828786022Z/profile.json", "source_run_id": "20260917T141828786022Z", "submitted_observation_frames": 1000, "workload_id": "hopper"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 121.70391511101934, "p99_latency_ms": 130.96357131735851, "return_raw": null, "survival_steps": 1000} +{"episode_id": 16, "episode_return": 3814.133230818948, "episode_return_env": 3814.133230818948, "game_score": null, "mean_latency_ms": 114.5993409244354, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/hopper/profile_latency/teacher/checkpoints/hopper_profile_appo_h1_5fps_20260917/checkpoint_p0/checkpoint_000019536_10002432.pth", "config_name": "hopper_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Hopper-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/hopper/profile_latency/selection_final", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "hopper_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/hopper/profile_latency/profile_3090_20260917T141828786022Z/profile.json", "source_run_id": "20260917T141828786022Z", "submitted_observation_frames": 1000, "workload_id": "hopper"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 121.08062203544773, "p99_latency_ms": 133.82208613353552, "return_raw": null, "survival_steps": 1000} +{"episode_id": 17, "episode_return": 3811.2177504405886, "episode_return_env": 3811.2177504405886, "game_score": null, "mean_latency_ms": 114.40987987770984, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/hopper/profile_latency/teacher/checkpoints/hopper_profile_appo_h1_5fps_20260917/checkpoint_p0/checkpoint_000019536_10002432.pth", "config_name": "hopper_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Hopper-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/hopper/profile_latency/selection_final", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "hopper_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/hopper/profile_latency/profile_3090_20260917T141828786022Z/profile.json", "source_run_id": "20260917T141828786022Z", "submitted_observation_frames": 1000, "workload_id": "hopper"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 120.9849709868762, "p99_latency_ms": 127.41288039707122, "return_raw": null, "survival_steps": 1000} +{"episode_id": 18, "episode_return": 3803.348027121752, "episode_return_env": 3803.348027121752, "game_score": null, "mean_latency_ms": 114.52542775804694, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/hopper/profile_latency/teacher/checkpoints/hopper_profile_appo_h1_5fps_20260917/checkpoint_p0/checkpoint_000019536_10002432.pth", "config_name": "hopper_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Hopper-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/hopper/profile_latency/selection_final", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "hopper_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/hopper/profile_latency/profile_3090_20260917T141828786022Z/profile.json", "source_run_id": "20260917T141828786022Z", "submitted_observation_frames": 1000, "workload_id": "hopper"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 121.16078157488214, "p99_latency_ms": 127.99722266879454, "return_raw": null, "survival_steps": 1000} +{"episode_id": 19, "episode_return": 3819.7451215416377, "episode_return_env": 3819.7451215416377, "game_score": null, "mean_latency_ms": 114.89218387011628, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/hopper/profile_latency/teacher/checkpoints/hopper_profile_appo_h1_5fps_20260917/checkpoint_p0/checkpoint_000019536_10002432.pth", "config_name": "hopper_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Hopper-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/hopper/profile_latency/selection_final", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "hopper_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/hopper/profile_latency/profile_3090_20260917T141828786022Z/profile.json", "source_run_id": "20260917T141828786022Z", "submitted_observation_frames": 1000, "workload_id": "hopper"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 121.03912278649729, "p99_latency_ms": 129.46797282419809, "return_raw": null, "survival_steps": 1000} diff --git a/latency-aware/hopper/small-policy/sample-factory-qwengr00t-rtx3090-v1/validation/selection_training_best/episode_metrics.jsonl b/latency-aware/hopper/small-policy/sample-factory-qwengr00t-rtx3090-v1/validation/selection_training_best/episode_metrics.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..2a3ec7682eb9887e13b248cbbb3435a73ad91c60 --- /dev/null +++ b/latency-aware/hopper/small-policy/sample-factory-qwengr00t-rtx3090-v1/validation/selection_training_best/episode_metrics.jsonl @@ -0,0 +1,20 @@ +{"episode_id": 0, "episode_return": 3818.3299991080344, "episode_return_env": 3818.3299991080344, "game_score": null, "mean_latency_ms": 115.18995074102081, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/hopper/profile_latency/teacher/checkpoints/hopper_profile_appo_h1_5fps_20260917/checkpoint_p0/best_000019512_9990144_reward_3220.937.pth", "config_name": "hopper_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Hopper-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/hopper/profile_latency/selection_training_best", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "hopper_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/hopper/profile_latency/profile_3090_20260917T141828786022Z/profile.json", "source_run_id": "20260917T141828786022Z", "submitted_observation_frames": 1000, "workload_id": "hopper"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 121.69597162426244, "p99_latency_ms": 129.89549037880013, "return_raw": null, "survival_steps": 1000} +{"episode_id": 1, "episode_return": 3801.949979062483, "episode_return_env": 3801.949979062483, "game_score": null, "mean_latency_ms": 114.56449403686167, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/hopper/profile_latency/teacher/checkpoints/hopper_profile_appo_h1_5fps_20260917/checkpoint_p0/best_000019512_9990144_reward_3220.937.pth", "config_name": "hopper_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Hopper-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/hopper/profile_latency/selection_training_best", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "hopper_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/hopper/profile_latency/profile_3090_20260917T141828786022Z/profile.json", "source_run_id": "20260917T141828786022Z", "submitted_observation_frames": 1000, "workload_id": "hopper"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 121.50491640436555, "p99_latency_ms": 125.97119040661914, "return_raw": null, "survival_steps": 1000} +{"episode_id": 2, "episode_return": 3799.39904827517, "episode_return_env": 3799.39904827517, "game_score": null, "mean_latency_ms": 114.43209582448789, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/hopper/profile_latency/teacher/checkpoints/hopper_profile_appo_h1_5fps_20260917/checkpoint_p0/best_000019512_9990144_reward_3220.937.pth", "config_name": "hopper_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Hopper-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/hopper/profile_latency/selection_training_best", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "hopper_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/hopper/profile_latency/profile_3090_20260917T141828786022Z/profile.json", "source_run_id": "20260917T141828786022Z", "submitted_observation_frames": 1000, "workload_id": "hopper"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 121.02486631581769, "p99_latency_ms": 127.06192303846402, "return_raw": null, "survival_steps": 1000} +{"episode_id": 3, "episode_return": 3806.0984441131777, "episode_return_env": 3806.0984441131777, "game_score": null, "mean_latency_ms": 114.84414803452853, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/hopper/profile_latency/teacher/checkpoints/hopper_profile_appo_h1_5fps_20260917/checkpoint_p0/best_000019512_9990144_reward_3220.937.pth", "config_name": "hopper_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Hopper-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/hopper/profile_latency/selection_training_best", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "hopper_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/hopper/profile_latency/profile_3090_20260917T141828786022Z/profile.json", "source_run_id": "20260917T141828786022Z", "submitted_observation_frames": 1000, "workload_id": "hopper"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 121.02132326348054, "p99_latency_ms": 127.9190625902109, "return_raw": null, "survival_steps": 1000} +{"episode_id": 4, "episode_return": 3821.3468930346507, "episode_return_env": 3821.3468930346507, "game_score": null, "mean_latency_ms": 114.94717422490875, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/hopper/profile_latency/teacher/checkpoints/hopper_profile_appo_h1_5fps_20260917/checkpoint_p0/best_000019512_9990144_reward_3220.937.pth", "config_name": "hopper_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Hopper-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/hopper/profile_latency/selection_training_best", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "hopper_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/hopper/profile_latency/profile_3090_20260917T141828786022Z/profile.json", "source_run_id": "20260917T141828786022Z", "submitted_observation_frames": 1000, "workload_id": "hopper"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 121.00997960349433, "p99_latency_ms": 128.21732582655437, "return_raw": null, "survival_steps": 1000} +{"episode_id": 5, "episode_return": 3818.6013992377857, "episode_return_env": 3818.6013992377857, "game_score": null, "mean_latency_ms": 114.60245796451443, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/hopper/profile_latency/teacher/checkpoints/hopper_profile_appo_h1_5fps_20260917/checkpoint_p0/best_000019512_9990144_reward_3220.937.pth", "config_name": "hopper_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Hopper-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/hopper/profile_latency/selection_training_best", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "hopper_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/hopper/profile_latency/profile_3090_20260917T141828786022Z/profile.json", "source_run_id": "20260917T141828786022Z", "submitted_observation_frames": 1000, "workload_id": "hopper"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 120.87727201786771, "p99_latency_ms": 127.45375690975183, "return_raw": null, "survival_steps": 1000} +{"episode_id": 6, "episode_return": 3825.5948641715036, "episode_return_env": 3825.5948641715036, "game_score": null, "mean_latency_ms": 114.71892566828849, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/hopper/profile_latency/teacher/checkpoints/hopper_profile_appo_h1_5fps_20260917/checkpoint_p0/best_000019512_9990144_reward_3220.937.pth", "config_name": "hopper_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Hopper-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/hopper/profile_latency/selection_training_best", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "hopper_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/hopper/profile_latency/profile_3090_20260917T141828786022Z/profile.json", "source_run_id": "20260917T141828786022Z", "submitted_observation_frames": 1000, "workload_id": "hopper"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 121.09108318462013, "p99_latency_ms": 127.16757159428731, "return_raw": null, "survival_steps": 1000} +{"episode_id": 7, "episode_return": 3817.364507562392, "episode_return_env": 3817.364507562392, "game_score": null, "mean_latency_ms": 114.78291074661247, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/hopper/profile_latency/teacher/checkpoints/hopper_profile_appo_h1_5fps_20260917/checkpoint_p0/best_000019512_9990144_reward_3220.937.pth", "config_name": "hopper_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Hopper-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/hopper/profile_latency/selection_training_best", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "hopper_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/hopper/profile_latency/profile_3090_20260917T141828786022Z/profile.json", "source_run_id": "20260917T141828786022Z", "submitted_observation_frames": 1000, "workload_id": "hopper"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 121.54565631351461, "p99_latency_ms": 132.3236564844519, "return_raw": null, "survival_steps": 1000} +{"episode_id": 8, "episode_return": 3814.7261006971084, "episode_return_env": 3814.7261006971084, "game_score": null, "mean_latency_ms": 114.76045460737626, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/hopper/profile_latency/teacher/checkpoints/hopper_profile_appo_h1_5fps_20260917/checkpoint_p0/best_000019512_9990144_reward_3220.937.pth", "config_name": "hopper_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Hopper-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/hopper/profile_latency/selection_training_best", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "hopper_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/hopper/profile_latency/profile_3090_20260917T141828786022Z/profile.json", "source_run_id": "20260917T141828786022Z", "submitted_observation_frames": 1000, "workload_id": "hopper"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 121.3125299720794, "p99_latency_ms": 130.23741427327107, "return_raw": null, "survival_steps": 1000} +{"episode_id": 9, "episode_return": 3825.008281517956, "episode_return_env": 3825.008281517956, "game_score": null, "mean_latency_ms": 114.36901179788543, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/hopper/profile_latency/teacher/checkpoints/hopper_profile_appo_h1_5fps_20260917/checkpoint_p0/best_000019512_9990144_reward_3220.937.pth", "config_name": "hopper_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Hopper-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/hopper/profile_latency/selection_training_best", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "hopper_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/hopper/profile_latency/profile_3090_20260917T141828786022Z/profile.json", "source_run_id": "20260917T141828786022Z", "submitted_observation_frames": 1000, "workload_id": "hopper"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 121.07475619002756, "p99_latency_ms": 126.98905492821542, "return_raw": null, "survival_steps": 1000} +{"episode_id": 10, "episode_return": 3816.453969095774, "episode_return_env": 3816.453969095774, "game_score": null, "mean_latency_ms": 114.78648178340455, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/hopper/profile_latency/teacher/checkpoints/hopper_profile_appo_h1_5fps_20260917/checkpoint_p0/best_000019512_9990144_reward_3220.937.pth", "config_name": "hopper_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Hopper-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/hopper/profile_latency/selection_training_best", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "hopper_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/hopper/profile_latency/profile_3090_20260917T141828786022Z/profile.json", "source_run_id": "20260917T141828786022Z", "submitted_observation_frames": 1000, "workload_id": "hopper"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 121.04216611012974, "p99_latency_ms": 128.9612933580124, "return_raw": null, "survival_steps": 1000} +{"episode_id": 11, "episode_return": 3815.17650963725, "episode_return_env": 3815.17650963725, "game_score": null, "mean_latency_ms": 114.8693603353378, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/hopper/profile_latency/teacher/checkpoints/hopper_profile_appo_h1_5fps_20260917/checkpoint_p0/best_000019512_9990144_reward_3220.937.pth", "config_name": "hopper_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Hopper-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/hopper/profile_latency/selection_training_best", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "hopper_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/hopper/profile_latency/profile_3090_20260917T141828786022Z/profile.json", "source_run_id": "20260917T141828786022Z", "submitted_observation_frames": 1000, "workload_id": "hopper"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 121.15181791640279, "p99_latency_ms": 126.56114195853236, "return_raw": null, "survival_steps": 1000} +{"episode_id": 12, "episode_return": 3816.6675065195645, "episode_return_env": 3816.6675065195645, "game_score": null, "mean_latency_ms": 114.56549116106042, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/hopper/profile_latency/teacher/checkpoints/hopper_profile_appo_h1_5fps_20260917/checkpoint_p0/best_000019512_9990144_reward_3220.937.pth", "config_name": "hopper_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Hopper-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/hopper/profile_latency/selection_training_best", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "hopper_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/hopper/profile_latency/profile_3090_20260917T141828786022Z/profile.json", "source_run_id": "20260917T141828786022Z", "submitted_observation_frames": 1000, "workload_id": "hopper"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 120.99124429827873, "p99_latency_ms": 129.0509902417293, "return_raw": null, "survival_steps": 1000} +{"episode_id": 13, "episode_return": 3810.7786398491785, "episode_return_env": 3810.7786398491785, "game_score": null, "mean_latency_ms": 114.7911034734444, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/hopper/profile_latency/teacher/checkpoints/hopper_profile_appo_h1_5fps_20260917/checkpoint_p0/best_000019512_9990144_reward_3220.937.pth", "config_name": "hopper_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Hopper-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/hopper/profile_latency/selection_training_best", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "hopper_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/hopper/profile_latency/profile_3090_20260917T141828786022Z/profile.json", "source_run_id": "20260917T141828786022Z", "submitted_observation_frames": 1000, "workload_id": "hopper"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 121.18068730074646, "p99_latency_ms": 125.48262594291727, "return_raw": null, "survival_steps": 1000} +{"episode_id": 14, "episode_return": 3813.106914295106, "episode_return_env": 3813.106914295106, "game_score": null, "mean_latency_ms": 114.65458686417112, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/hopper/profile_latency/teacher/checkpoints/hopper_profile_appo_h1_5fps_20260917/checkpoint_p0/best_000019512_9990144_reward_3220.937.pth", "config_name": "hopper_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Hopper-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/hopper/profile_latency/selection_training_best", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "hopper_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/hopper/profile_latency/profile_3090_20260917T141828786022Z/profile.json", "source_run_id": "20260917T141828786022Z", "submitted_observation_frames": 1000, "workload_id": "hopper"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 121.22762011132701, "p99_latency_ms": 128.9128408954642, "return_raw": null, "survival_steps": 1000} +{"episode_id": 15, "episode_return": 3803.3492543466577, "episode_return_env": 3803.3492543466577, "game_score": null, "mean_latency_ms": 114.66081796963667, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/hopper/profile_latency/teacher/checkpoints/hopper_profile_appo_h1_5fps_20260917/checkpoint_p0/best_000019512_9990144_reward_3220.937.pth", "config_name": "hopper_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Hopper-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/hopper/profile_latency/selection_training_best", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "hopper_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/hopper/profile_latency/profile_3090_20260917T141828786022Z/profile.json", "source_run_id": "20260917T141828786022Z", "submitted_observation_frames": 1000, "workload_id": "hopper"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 121.70391511101934, "p99_latency_ms": 130.96357131735851, "return_raw": null, "survival_steps": 1000} +{"episode_id": 16, "episode_return": 3815.0139371581436, "episode_return_env": 3815.0139371581436, "game_score": null, "mean_latency_ms": 114.5993409244354, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/hopper/profile_latency/teacher/checkpoints/hopper_profile_appo_h1_5fps_20260917/checkpoint_p0/best_000019512_9990144_reward_3220.937.pth", "config_name": "hopper_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Hopper-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/hopper/profile_latency/selection_training_best", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "hopper_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/hopper/profile_latency/profile_3090_20260917T141828786022Z/profile.json", "source_run_id": "20260917T141828786022Z", "submitted_observation_frames": 1000, "workload_id": "hopper"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 121.08062203544773, "p99_latency_ms": 133.82208613353552, "return_raw": null, "survival_steps": 1000} +{"episode_id": 17, "episode_return": 3819.671617704402, "episode_return_env": 3819.671617704402, "game_score": null, "mean_latency_ms": 114.40987987770984, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/hopper/profile_latency/teacher/checkpoints/hopper_profile_appo_h1_5fps_20260917/checkpoint_p0/best_000019512_9990144_reward_3220.937.pth", "config_name": "hopper_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Hopper-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/hopper/profile_latency/selection_training_best", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "hopper_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/hopper/profile_latency/profile_3090_20260917T141828786022Z/profile.json", "source_run_id": "20260917T141828786022Z", "submitted_observation_frames": 1000, "workload_id": "hopper"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 120.9849709868762, "p99_latency_ms": 127.41288039707122, "return_raw": null, "survival_steps": 1000} +{"episode_id": 18, "episode_return": 3825.458467648868, "episode_return_env": 3825.458467648868, "game_score": null, "mean_latency_ms": 114.52542775804694, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/hopper/profile_latency/teacher/checkpoints/hopper_profile_appo_h1_5fps_20260917/checkpoint_p0/best_000019512_9990144_reward_3220.937.pth", "config_name": "hopper_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Hopper-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/hopper/profile_latency/selection_training_best", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "hopper_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/hopper/profile_latency/profile_3090_20260917T141828786022Z/profile.json", "source_run_id": "20260917T141828786022Z", "submitted_observation_frames": 1000, "workload_id": "hopper"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 121.16078157488214, "p99_latency_ms": 127.99722266879454, "return_raw": null, "survival_steps": 1000} +{"episode_id": 19, "episode_return": 3825.03671743075, "episode_return_env": 3825.03671743075, "game_score": null, "mean_latency_ms": 114.89218387011628, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/hopper/profile_latency/teacher/checkpoints/hopper_profile_appo_h1_5fps_20260917/checkpoint_p0/best_000019512_9990144_reward_3220.937.pth", "config_name": "hopper_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Hopper-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/hopper/profile_latency/selection_training_best", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "hopper_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/hopper/profile_latency/profile_3090_20260917T141828786022Z/profile.json", "source_run_id": "20260917T141828786022Z", "submitted_observation_frames": 1000, "workload_id": "hopper"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 121.03912278649729, "p99_latency_ms": 129.46797282419809, "return_raw": null, "survival_steps": 1000} diff --git a/latency-aware/hopper/small-policy/sample-factory-qwengr00t-rtx3090-v1/validation/teacher_probe_audit.json b/latency-aware/hopper/small-policy/sample-factory-qwengr00t-rtx3090-v1/validation/teacher_probe_audit.json new file mode 100644 index 0000000000000000000000000000000000000000..cecb8c9713d61539c7b8ef92f17b29fddc643ae3 --- /dev/null +++ b/latency-aware/hopper/small-policy/sample-factory-qwengr00t-rtx3090-v1/validation/teacher_probe_audit.json @@ -0,0 +1,241 @@ +{ + "teacher_config": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/hopper/profile_latency/teacher.yaml", + "profile_sha256": "9d2e175b36914cc07fe93711db34ea89593c6c71e2b2a3d7fbe9176173ab7184", + "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 + ], + "returns": [ + 3813.902369620451, + 3814.717929846383, + 3815.678995110681, + 3813.5158104903485, + 3812.1721732868205, + 3822.869610539937, + 3816.213698287465, + 3819.3530723476006, + 3809.7239266677057, + 3819.2709564714014, + 3804.324024566971, + 3818.1522276093633, + 3829.262769910777, + 3815.7688288666577, + 3814.218115130112, + 3802.188637269843, + 3814.133230818948, + 3811.2177504405886, + 3803.348027121752, + 3819.7451215416377 + ], + "mean": 3814.488863797272, + "raw_steps": 20000, + "actions": 20000, + "invalid": 0, + "drops": 0, + "clock_ms": 200, + "latency_raw_frames": [ + 1 + ] + }, + "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 + ], + "returns": [ + 3818.3299991080344, + 3801.949979062483, + 3799.39904827517, + 3806.0984441131777, + 3821.3468930346507, + 3818.6013992377857, + 3825.5948641715036, + 3817.364507562392, + 3814.7261006971084, + 3825.008281517956, + 3816.453969095774, + 3815.17650963725, + 3816.6675065195645, + 3810.7786398491785, + 3813.106914295106, + 3803.3492543466577, + 3815.0139371581436, + 3819.671617704402, + 3825.458467648868, + 3825.03671743075 + ], + "mean": 3815.4566525232976, + "raw_steps": 20000, + "actions": 20000, + "invalid": 0, + "drops": 0, + "clock_ms": 200, + "latency_raw_frames": [ + 1 + ] + }, + "E10": { + "episodes": 10, + "seeds": [ + 104771, + 104772, + 104773, + 104774, + 104775, + 104776, + 104777, + 104778, + 104779, + 104780 + ], + "returns": [ + 3818.3299991080344, + 3801.949979062483, + 3799.39904827517, + 3806.0984441131777, + 3821.3468930346507, + 3818.6013992377857, + 3825.5948641715036, + 3817.364507562392, + 3814.7261006971084, + 3825.008281517956 + ], + "mean": 3814.8419516780264, + "raw_steps": 10000, + "actions": 10000, + "invalid": 0, + "drops": 0, + "clock_ms": 200, + "latency_raw_frames": [ + 1 + ] + } + }, + "probe": { + "attempts": 12, + "accepted": 12, + "rejected": 0, + "min_return": 3794.1470665335655, + "seeds": [ + 3333, + 3334, + 3335, + 3336, + 3337, + 3338, + 3339, + 3340, + 3341, + 3342, + 3343, + 3344 + ] + }, + "selection": { + "selected": "training_best", + "candidates": { + "final": { + "checkpoint": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/hopper/profile_latency/teacher/checkpoints/hopper_profile_appo_h1_5fps_20260917/checkpoint_p0/checkpoint_000019536_10002432.pth", + "sha256": "d7bccf0797e5622069c1af0aa67796863bd68c40707cd96c7e7399fa27ca49e0", + "episodes": 20, + "returns": [ + 3813.902369620451, + 3814.717929846383, + 3815.678995110681, + 3813.5158104903485, + 3812.1721732868205, + 3822.869610539937, + 3816.213698287465, + 3819.3530723476006, + 3809.7239266677057, + 3819.2709564714014, + 3804.324024566971, + 3818.1522276093633, + 3829.262769910777, + 3815.7688288666577, + 3814.218115130112, + 3802.188637269843, + 3814.133230818948, + 3811.2177504405886, + 3803.348027121752, + 3819.7451215416377 + ], + "mean": 3814.488863797272, + "std": 6.329159120614397, + "strict_gt3000": 20 + }, + "training_best": { + "checkpoint": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/hopper/profile_latency/teacher/checkpoints/hopper_profile_appo_h1_5fps_20260917/checkpoint_p0/best_000019512_9990144_reward_3220.937.pth", + "sha256": "f8b6089e67f129a7c17504bc8d6146003177b260e927cc1281a178d26f404b74", + "episodes": 20, + "returns": [ + 3818.3299991080344, + 3801.949979062483, + 3799.39904827517, + 3806.0984441131777, + 3821.3468930346507, + 3818.6013992377857, + 3825.5948641715036, + 3817.364507562392, + 3814.7261006971084, + 3825.008281517956, + 3816.453969095774, + 3815.17650963725, + 3816.6675065195645, + 3810.7786398491785, + 3813.106914295106, + 3803.3492543466577, + 3815.0139371581436, + 3819.671617704402, + 3825.458467648868, + 3825.03671743075 + ], + "mean": 3815.4566525232976, + "std": 7.632802007802108, + "strict_gt3000": 20 + } + }, + "completed_at": "2026-09-17T14:44:34.549937+00:00" + }, + "note": "E10 reuses first10selection seeds; not independent seeds" +} \ No newline at end of file diff --git a/latency-aware/hopper/small-policy/sample-factory-qwenoft-rtx3090-v1/README.md b/latency-aware/hopper/small-policy/sample-factory-qwenoft-rtx3090-v1/README.md new file mode 100644 index 0000000000000000000000000000000000000000..24ff506d7ca1bb642333ed103ff4472c289c98cd --- /dev/null +++ b/latency-aware/hopper/small-policy/sample-factory-qwenoft-rtx3090-v1/README.md @@ -0,0 +1,26 @@ +# hopper / sample-factory-appo + +Training condition: `latency-aware`. Run: `hopper_profile_20260911T180943Z`. + +Copied from [Standard-Pipeline](https://huggingface.co/latency-sensitive-bench/Standard-Pipeline/tree/2208875f92b21e3f8ec2363a30a6705dfe56cc04/hopper/profile_latency/small_model) at `2208875f92b21e3f8ec2363a30a6705dfe56cc04`. + +- Checkpoint: `checkpoint.pth` +- Selection: best_reward_after_full_training_budget +- Checkpoint SHA256: `257d8c101292bd9f49a26f3c79601c5461bb5320c8b99e35f198b503756d2251` +- 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/hopper/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/hopper/small-policy/sample-factory-qwenoft-rtx3090-v1/checkpoint.pth b/latency-aware/hopper/small-policy/sample-factory-qwenoft-rtx3090-v1/checkpoint.pth new file mode 100644 index 0000000000000000000000000000000000000000..34de3b9723ceae60da2f56db15f32bbe6ee3c56a --- /dev/null +++ b/latency-aware/hopper/small-policy/sample-factory-qwenoft-rtx3090-v1/checkpoint.pth @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:257d8c101292bd9f49a26f3c79601c5461bb5320c8b99e35f198b503756d2251 +size 27445 diff --git a/latency-aware/hopper/small-policy/sample-factory-qwenoft-rtx3090-v1/config.json b/latency-aware/hopper/small-policy/sample-factory-qwenoft-rtx3090-v1/config.json new file mode 100644 index 0000000000000000000000000000000000000000..88afea6145b08ccad1f681a1d1d5ea4d82bd9676 --- /dev/null +++ b/latency-aware/hopper/small-policy/sample-factory-qwenoft-rtx3090-v1/config.json @@ -0,0 +1,283 @@ +{ + "help": false, + "algo": "APPO", + "env": "latency_gymnasium", + "experiment": "hopper_profile_20260911T180943Z", + "train_dir": "/mnt/results/latency-sensitive-bench/profile-latency-new/hopper/20260911T180943Z-hopper/small_models", + "restart_behavior": "overwrite", + "device": "gpu", + "seed": 3333, + "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": 10000, + "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": false, + "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": 1.0, + "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": 10, + "decorrelate_envs_on_one_worker": true, + "actor_worker_gpus": [], + "set_workers_cpu_affinity": true, + "force_envs_single_thread": true, + "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": 3, + "load_checkpoint_kind": "latest", + "save_milestones_sec": -1, + "save_best_every_sec": 60, + "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": 4, + "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": "hopper_rgb_state", + "gym_env_id": "LatencyBench/Hopper-v0", + "gym_make_kwargs_json": "{\"base_env_id\": \"Hopper-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_hopper\"]", + "gym_action_space_json": "{\"dtype\": \"float32\", \"high\": [1.0, 1.0, 1.0], \"labels\": [\"thigh_torque\", \"leg_torque\", \"foot_torque\"], \"low\": [-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]", + "gym_base_prompt": "Move the Hopper robot forward while keeping its torso upright. Predict three continuous torques in [-1, 1] ordered as thigh, leg, and foot.", + "gym_state_labels_json": "[\"torso_height\", \"torso_angle\", \"thigh_angle\", \"leg_angle\", \"foot_angle\", \"torso_x_velocity\", \"torso_z_velocity\", \"torso_angular_velocity\", \"thigh_angular_velocity\", \"leg_angular_velocity\", \"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/hopper/20260911T180943Z-hopper/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": 10, + "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/hopper/20260911T180943Z-hopper/small_models/hopper_profile_20260911T180943Z/episode_metrics.jsonl", + "command_line": "--mode train --algo APPO --env latency_gymnasium --experiment hopper_profile_20260911T180943Z --train_dir /mnt/results/latency-sensitive-bench/profile-latency-new/hopper/20260911T180943Z-hopper/small_models --restart_behavior overwrite --device gpu --seed 3333 --episode_metrics_path /mnt/results/latency-sensitive-bench/profile-latency-new/hopper/20260911T180943Z-hopper/small_models/hopper_profile_20260911T180943Z/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 10000 --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 4 --gamma 0.99 --gae_lambda 0.95 --ppo_clip_ratio 0.2 --ppo_clip_value 1.0 --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 10 --default_niceness 0 --rnn_type gru --rnn_size 512 --save_every_sec 600 --keep_checkpoints 3 --save_milestones_sec -1 --save_best_every_sec 60 --save_best_after 100000 --stats_avg 100 --experiment_summaries_interval 10 --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 --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 False --with_vtrace False --decorrelate_envs_on_one_worker True --set_workers_cpu_affinity True --force_envs_single_thread True --actor_critic_share_weights True --with_wandb False --fasttd3_compile True --encoder_mlp_layers 64 64 --latency-type iid --latency-seed 0 --latency-profile-path /mnt/results/latency-sensitive-bench/profile-latency-new/hopper/20260911T180943Z-hopper/profile/profile.json --latency-profile-worker-slot 0 --add-latency-info False --eval-episodes 10 --eval-parallel-envs 1 --eval-max-steps 1000 --eval-deterministic True --gym-task-name hopper_rgb_state --gym-env-id LatencyBench/Hopper-v0 --gym-make-kwargs-json {\"base_env_id\": \"Hopper-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_hopper\"] --gym-action-space-json {\"dtype\": \"float32\", \"high\": [1.0, 1.0, 1.0], \"labels\": [\"thigh_torque\", \"leg_torque\", \"foot_torque\"], \"low\": [-1.0, -1.0, -1.0], \"type\": \"box\"} --gym-noop-action-json [0.0, 0.0, 0.0] --gym-base-prompt Move the Hopper robot forward while keeping its torso upright. Predict three continuous torques in [-1, 1] ordered as thigh, leg, and foot. --gym-state-labels-json [\"torso_height\", \"torso_angle\", \"thigh_angle\", \"leg_angle\", \"foot_angle\", \"torso_x_velocity\", \"torso_z_velocity\", \"torso_angular_velocity\", \"thigh_angular_velocity\", \"leg_angular_velocity\", \"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": "hopper_profile_20260911T180943Z", + "train_dir": "/mnt/results/latency-sensitive-bench/profile-latency-new/hopper/20260911T180943Z-hopper/small_models", + "restart_behavior": "overwrite", + "device": "gpu", + "seed": 3333, + "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": 10000, + "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, + "recurrence": 1, + "shuffle_minibatches": false, + "gamma": 0.99, + "reward_scale": 1.0, + "reward_clip": 1000.0, + "value_bootstrap": false, + "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": 1.0, + "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": 10, + "decorrelate_envs_on_one_worker": true, + "set_workers_cpu_affinity": true, + "force_envs_single_thread": true, + "default_niceness": 0, + "experiment_summaries_interval": 10, + "stats_avg": 100, + "train_for_env_steps": 10000000, + "save_every_sec": 600, + "keep_checkpoints": 3, + "save_milestones_sec": -1, + "save_best_every_sec": 60, + "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": 4, + "with_wandb": false, + "gym_task_name": "hopper_rgb_state", + "gym_env_id": "LatencyBench/Hopper-v0", + "gym_make_kwargs_json": "{\"base_env_id\": \"Hopper-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_hopper\"]", + "gym_action_space_json": "{\"dtype\": \"float32\", \"high\": [1.0, 1.0, 1.0], \"labels\": [\"thigh_torque\", \"leg_torque\", \"foot_torque\"], \"low\": [-1.0, -1.0, -1.0], \"type\": \"box\"}", + "gym_noop_action_json": "[0.0, 0.0, 0.0]", + "gym_base_prompt": "Move the Hopper robot forward while keeping its torso upright. Predict three continuous torques in [-1, 1] ordered as thigh, leg, and foot.", + "gym_state_labels_json": "[\"torso_height\", \"torso_angle\", \"thigh_angle\", \"leg_angle\", \"foot_angle\", \"torso_x_velocity\", \"torso_z_velocity\", \"torso_angular_velocity\", \"thigh_angular_velocity\", \"leg_angular_velocity\", \"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/hopper/20260911T180943Z-hopper/profile/profile.json", + "latency_profile_worker_slot": 0, + "add_latency_info": false, + "hold_policy": "hold", + "ordering_policy": "issue_order_fifo", + "eval_episodes": 10, + "eval_parallel_envs": 1, + "eval_max_steps": 1000, + "eval_deterministic": true, + "episode_metrics_path": "/mnt/results/latency-sensitive-bench/profile-latency-new/hopper/20260911T180943Z-hopper/small_models/hopper_profile_20260911T180943Z/episode_metrics.jsonl", + "output_dir": "/mnt/results/latency-sensitive-bench/profile-latency-new/hopper/20260911T180943Z-hopper/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/hopper/20260911T180943Z-hopper/small_train" +} \ No newline at end of file diff --git a/latency-aware/hopper/small-policy/sample-factory-qwenoft-rtx3090-v1/profile/DONE b/latency-aware/hopper/small-policy/sample-factory-qwenoft-rtx3090-v1/profile/DONE new file mode 100644 index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391 diff --git a/latency-aware/hopper/small-policy/sample-factory-qwenoft-rtx3090-v1/profile/hardware.json b/latency-aware/hopper/small-policy/sample-factory-qwenoft-rtx3090-v1/profile/hardware.json new file mode 100644 index 0000000000000000000000000000000000000000..e8204e81807044975c0e0a0953d16d6835192f31 --- /dev/null +++ b/latency-aware/hopper/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/hopper/small-policy/sample-factory-qwenoft-rtx3090-v1/profile/latency_burst_model.json b/latency-aware/hopper/small-policy/sample-factory-qwenoft-rtx3090-v1/profile/latency_burst_model.json new file mode 100644 index 0000000000000000000000000000000000000000..1d402e53f0f814c206fb40d3f5366493189ae014 --- /dev/null +++ b/latency-aware/hopper/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": [ + 92.92644906044006, + 92.92644906044006, + 92.92644906044006, + 92.92644906044006, + 92.92644906044006, + 92.92644906044006, + 92.92644906044006, + 92.92644906044006, + 92.92644906044006, + 92.92644906044006, + 92.92644906044006, + 92.92644906044006, + 92.92644906044006, + 92.92644906044006, + 92.92644906044006, + 92.92644906044006, + 92.92644906044006, + 92.92644906044006, + 92.92644906044006, + 92.92644906044006, + 92.92644906044006, + 92.92644906044006, + 92.92644906044006, + 92.92644906044006, + 92.92644906044006, + 92.92644906044006, + 92.92644906044006, + 92.92644906044006, + 92.92818662166596, + 92.93339930534363, + 92.9386119890213, + 92.94382467269898, + 92.94903735637665, + 92.95425004005432, + 92.95946272373199, + 92.96467540740967, + 92.96988809108734, + 92.97510077476501, + 92.9803134584427, + 92.98552614212036, + 92.99073882579803, + 92.9959515094757, + 93.00116419315339, + 93.00637687683106, + 93.01158956050872, + 93.01680224418641, + 93.02201492786408, + 93.02722761154175, + 93.03244029521942, + 93.0376529788971, + 93.04286566257477, + 93.04807834625244, + 93.05329102993011, + 93.05850371360779, + 93.06371639728546, + 93.06892908096313, + 93.07414176464081, + 93.07935444831848, + 93.08456713199615, + 93.08977981567382, + 93.0949924993515, + 93.10020518302917, + 93.65123818159104, + 94.20227118015289, + 94.75330417871476, + 95.30433717727661, + 95.85537017583847, + 96.40640317440034, + 96.95743617296219, + 97.50846917152404, + 98.0595021700859, + 98.61053516864776, + 99.16156816720962, + 99.71260116577149, + 100.26363416433334, + 100.8146671628952, + 101.36570016145707, + 101.91673316001892, + 102.46776615858079, + 103.01879915714264, + 103.56983215570449, + 104.12086515426635, + 104.67189815282822, + 105.22293115139007, + 105.77396414995194, + 106.32499714851379, + 106.87603014707565, + 107.42706314563752, + 107.97809614419937, + 108.52912914276124, + 109.08016214132309, + 109.63119513988495, + 110.18222813844682, + 110.73326113700867, + 111.28429413557052, + 111.46797180175781, + 111.46797180175781, + 111.46797180175781, + 111.46797180175781, + 111.46797180175781, + 111.46797180175781, + 111.46797180175781, + 111.46797180175781, + 111.46797180175781, + 111.46797180175781, + 111.46797180175781, + 111.46797180175781, + 111.46797180175781, + 111.46797180175781, + 111.46797180175781, + 111.46797180175781, + 111.46797180175781, + 111.46797180175781, + 111.46797180175781, + 111.46797180175781, + 111.46797180175781, + 111.46797180175781, + 111.46797180175781, + 111.46797180175781, + 111.46797180175781, + 111.46797180175781, + 111.46797180175781, + 111.46797180175781 + ], + "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": [ + 0.31311678886413574, + 0.31311678886413574, + 0.3165763258934021, + 0.32325554490089414, + 0.3290378570556641, + 0.34071471095085143, + 0.3509213447570801, + 0.35257425904273987, + 0.3555735468864441, + 0.3601503908634186, + 0.3644925594329834, + 0.3818242847919464, + 0.3849796652793884, + 0.43671417236328125, + 0.4749835133552551, + 0.4945000410079956, + 0.5159845948219299, + 0.5300548076629639, + 0.5394640564918518, + 0.547340989112854, + 0.5548242330551147, + 0.5603091716766357, + 0.5666623711585999, + 0.5706260204315186, + 0.5731815695762634, + 0.5773470401763916, + 0.5807499289512634, + 0.5849419832229614, + 0.5887086987495422, + 0.5918290615081787, + 0.5939978957176208, + 0.5969480276107788, + 0.5994340777397156, + 0.6025159358978271, + 0.6040467619895935, + 0.6064813137054443, + 0.6088453531265259, + 0.6113431453704834, + 0.6135084629058838, + 0.6157195568084717, + 0.6180734634399414, + 0.6201481819152832, + 0.6233775615692139, + 0.6259479522705078, + 0.6277621388435364, + 0.6300039291381836, + 0.6324813365936279, + 0.6342650651931763, + 0.6360381841659546, + 0.6377451419830322, + 0.639049768447876, + 0.640360951423645, + 0.6418464183807373, + 0.6428799629211426, + 0.6441222429275513, + 0.6458930969238281, + 0.6482654809951782, + 0.6496939659118652, + 0.6526684761047363, + 0.6541144847869873, + 0.6560317873954773, + 0.658473014831543, + 0.659576952457428, + 0.6612474918365479, + 0.6628846526145935, + 0.6641979217529297, + 0.6656185388565063, + 0.6666920185089111, + 0.6680689454078674, + 0.6693181991577148, + 0.6703876256942749, + 0.6712313890457153, + 0.6727586388587952, + 0.6742727756500244, + 0.6758086681365967, + 0.6774241924285889, + 0.6792400479316711, + 0.6818399429321289, + 0.683896005153656, + 0.6866179704666138, + 0.6883576512336731, + 0.6901149749755859, + 0.692481517791748, + 0.6947081089019775, + 0.697051465511322, + 0.698936939239502, + 0.7017368674278259, + 0.7050299644470215, + 0.70822674036026, + 0.7101070880889893, + 0.7132005095481873, + 0.7152360677719116, + 0.7168729901313782, + 0.7200279235839844, + 0.7235506176948547, + 0.7269525527954102, + 0.7312224507331848, + 0.7360110282897949, + 0.7416163682937622, + 0.7470424175262451, + 0.7538335919380188, + 0.7591180801391602, + 0.7654633522033691, + 0.7710039615631104, + 0.7774527072906494, + 0.7881929874420166, + 0.8031460642814636, + 0.8192280530929565, + 0.8465723991394043, + 0.8769099712371826, + 0.9416267275810242, + 0.9441980957984923, + 0.9461143732070924, + 0.9521929323673245, + 0.9566469669342051, + 0.9747528433799744, + 0.979430532455442, + 1.0052420735359195, + 1.0194490909576401, + 1.0595047175884362, + 1.2158348947763677, + 1.3550479412078857, + 1.3550479412078857 + ], + "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": 2072 + }, + "regime_transition_counts": { + "burst": { + "burst": 0, + "calm": 3 + }, + "calm": { + "burst": 3, + "calm": 2064 + } + }, + "reset_scope": "session", + "schema_version": 12, + "spike_median_multiplier": 1.25, + "spike_threshold_ms_by_worker_slot": { + "0": 87.82212123274803 + }, + "worker_count": 1 +} diff --git a/latency-aware/hopper/small-policy/sample-factory-qwenoft-rtx3090-v1/profile/latency_distribution.json b/latency-aware/hopper/small-policy/sample-factory-qwenoft-rtx3090-v1/profile/latency_distribution.json new file mode 100644 index 0000000000000000000000000000000000000000..59cde1722b922a5e61b79b6e5c535c5fe9b10c7c --- /dev/null +++ b/latency-aware/hopper/small-policy/sample-factory-qwenoft-rtx3090-v1/profile/latency_distribution.json @@ -0,0 +1,1027 @@ +{ + "schema_version": 3, + "worker_slots": { + "0": { + "all": { + "count": 2075, + "observation_to_action_latency_ms": { + "distribution_type": "inverse_cdf", + "latency_ms": [ + 65.19399881362915, + 65.19399881362915, + 65.21452818512917, + 65.37617138624191, + 65.54988111257553, + 65.60978311300278, + 65.85620288848877, + 66.11677795648575, + 66.1667419910431, + 66.25579528808593, + 66.3200498342514, + 66.37821425199509, + 66.44661968946457, + 67.40388679504395, + 67.9565160870552, + 68.30614387989044, + 68.86876726150513, + 69.23061203956604, + 69.40943151712418, + 69.708052277565, + 69.82890748977661, + 69.99191188812256, + 70.07996428012848, + 70.20106554031372, + 70.2888200879097, + 70.34249114990234, + 70.42751479148865, + 70.50662112236023, + 70.5581624507904, + 70.60912585258484, + 70.65479511022568, + 70.69437503814697, + 70.74371039867401, + 70.80917811393738, + 70.83664780855179, + 70.88321900367737, + 70.91019904613495, + 70.95217895507812, + 70.97798812389374, + 71.01112449169159, + 71.03684651851654, + 71.07607984542847, + 71.12114173173904, + 71.14498901367188, + 71.169668674469, + 71.19082713127136, + 71.21074295043945, + 71.24556457996368, + 71.27112966775894, + 71.28998637199402, + 71.31158643960953, + 71.34597957134247, + 71.36409616470337, + 71.39237093925476, + 71.41189247369766, + 71.44127345085144, + 71.47214812040329, + 71.49239921569824, + 71.51301550865173, + 71.53032338619232, + 71.5557844042778, + 71.5749728679657, + 71.59549480676651, + 71.6154580116272, + 71.63657933473587, + 71.6589138507843, + 71.68088454008102, + 71.70165157318115, + 71.73091351985931, + 71.75631403923035, + 71.77797484397888, + 71.82077741622925, + 71.85434252023697, + 71.88232922554016, + 71.9016506075859, + 71.91977643966675, + 71.95200443267822, + 71.98209404945374, + 72.0122202038765, + 72.03439712524414, + 72.06454348564148, + 72.09861707687378, + 72.12148833274841, + 72.1564769744873, + 72.1896619796753, + 72.22209119796753, + 72.27087277173996, + 72.31662225723267, + 72.35859662294388, + 72.4115800857544, + 72.43648117780685, + 72.47373795509338, + 72.49812859296799, + 72.53804278373718, + 72.58772403001785, + 72.64632666110992, + 72.70239788293839, + 72.79391694068909, + 72.83307409286499, + 72.89201402664185, + 72.95603406429291, + 73.05068588256836, + 73.21339404582977, + 73.34779798984528, + 73.54553669691086, + 73.78722310066223, + 74.11802303791046, + 74.58567237854004, + 75.28934186697006, + 76.44343304634094, + 77.00749260187149, + 77.320473575592, + 77.51085505485536, + 78.1938346505165, + 78.28879662752153, + 78.42176166176796, + 79.36353144645672, + 81.95102273225795, + 83.81126960515967, + 93.65813264846805, + 102.26191507876086, + 112.13530778884888, + 112.13530778884888 + ], + "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": 0.9980124420738008, + "worker_service_time_ms": { + "distribution_type": "inverse_cdf", + "latency_ms": [ + 64.81042885780334, + 64.81042885780334, + 64.8304044932127, + 64.86529658436775, + 65.09479154348374, + 65.18544340133667, + 65.45056700706482, + 65.71252375841141, + 65.7791820526123, + 65.83106318712234, + 65.96416923999786, + 65.99974170923232, + 66.05729752779007, + 66.85541200637817, + 67.4124995470047, + 67.70765841007233, + 68.38320344686508, + 68.66614484786987, + 68.86086118221283, + 69.11453056335449, + 69.2644213438034, + 69.38441181182861, + 69.49850225448608, + 69.59222090244293, + 69.66973400115967, + 69.74627423286438, + 69.80929481983185, + 69.88673007488251, + 69.92713761329651, + 69.97099304199219, + 70.01391452550888, + 70.07167446613312, + 70.1147900223732, + 70.16800880432129, + 70.20960211753845, + 70.24057853221893, + 70.27172857522964, + 70.30395483970642, + 70.33524876832962, + 70.36733794212341, + 70.41446840763092, + 70.43638014793396, + 70.47007262706757, + 70.49437403678894, + 70.52942305803299, + 70.54843187332153, + 70.56563359498978, + 70.58997797966003, + 70.61523360013962, + 70.63820695877075, + 70.65534472465515, + 70.68713617324829, + 70.72156208753586, + 70.74655199050903, + 70.77104866504669, + 70.794921875, + 70.81391477584839, + 70.83501505851746, + 70.85863208770752, + 70.87805342674255, + 70.89673417806625, + 70.91481804847717, + 70.93379074335098, + 70.95436489582062, + 70.9727611541748, + 70.99777722358704, + 71.01445686817169, + 71.04416513442993, + 71.07488167285919, + 71.09937310218811, + 71.1260045170784, + 71.14588451385498, + 71.17757761478424, + 71.20317506790161, + 71.22922629117966, + 71.26177847385406, + 71.28523004055023, + 71.31312823295593, + 71.33135783672333, + 71.35811042785645, + 71.38167172670364, + 71.40552186965942, + 71.44026362895966, + 71.4722410440445, + 71.50916689634323, + 71.55513215065002, + 71.59390467405319, + 71.63793969154358, + 71.6769991517067, + 71.70788717269897, + 71.74298250675201, + 71.7811050415039, + 71.81722241640091, + 71.84697794914246, + 71.89713567495346, + 71.96147060394287, + 72.00608330965042, + 72.07237100601196, + 72.13394355773926, + 72.19530701637268, + 72.25350522994995, + 72.34730696678162, + 72.50979733467102, + 72.63740396499634, + 72.84575146436691, + 73.07518124580383, + 73.36473077535629, + 73.83938801288605, + 74.32575643062592, + 75.57429695129395, + 76.11366033554077, + 76.45826106667515, + 76.83195266723635, + 77.16471412181853, + 77.66161592006684, + 77.73926329612732, + 78.54240913391091, + 81.13938184976583, + 83.1561977505683, + 93.00029541254047, + 101.59529724419284, + 111.46797180175781, + 111.46797180175781 + ], + "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": 2072, + "observation_to_action_latency_ms": { + "distribution_type": "inverse_cdf", + "latency_ms": [ + 65.19399881362915, + 65.19399881362915, + 65.21447089385987, + 65.37542019462586, + 65.54987103176117, + 65.60966022491455, + 65.85360446739197, + 66.113779296875, + 66.16652703666686, + 66.25572955894471, + 66.31964143466949, + 66.37805958938598, + 66.44557957172394, + 67.40347755432128, + 67.95384462833404, + 68.3040751028061, + 68.86173973083496, + 69.22473413944245, + 69.40377369403839, + 69.70454369068146, + 69.82808116436004, + 69.99165661334992, + 70.07831597328186, + 70.19712745666504, + 70.2876419210434, + 70.3415141248703, + 70.42468881607056, + 70.50353671073914, + 70.55655439853668, + 70.60617962837219, + 70.65110831737519, + 70.69143896102905, + 70.74109094619752, + 70.8089477443695, + 70.83344268321991, + 70.88257723331452, + 70.90947210788727, + 70.94953246593475, + 70.9757197189331, + 71.01045125961303, + 71.03570990562439, + 71.07419350147248, + 71.11981529712676, + 71.14466162681579, + 71.16853910923004, + 71.19018000125885, + 71.20882227420807, + 71.24291341304779, + 71.26965581893921, + 71.28910193443298, + 71.30903896331787, + 71.34541156291962, + 71.3635185432434, + 71.3916244506836, + 71.41094114780427, + 71.43986271858215, + 71.47048921585083, + 71.49185324668885, + 71.51238204956054, + 71.52781995773316, + 71.55354582309722, + 71.57390081882477, + 71.59174089431762, + 71.61493656635284, + 71.6342034482956, + 71.65483741283417, + 71.68058023452758, + 71.70002612113953, + 71.72443329334259, + 71.75478495121003, + 71.77438542366028, + 71.81701493263245, + 71.84875475883484, + 71.87945019245147, + 71.90045991420746, + 71.91874521255494, + 71.94669437408447, + 71.98143951892852, + 72.01119516849518, + 72.0328631067276, + 72.06257130146027, + 72.09504079818726, + 72.11566365242004, + 72.15310551643371, + 72.18850584506988, + 72.21814324855805, + 72.26514148712158, + 72.31500059127808, + 72.35579874992371, + 72.40864204883576, + 72.43190546512604, + 72.46693255901337, + 72.49548690319061, + 72.53481489181519, + 72.58008307933807, + 72.64314253807068, + 72.69793500900269, + 72.78377663612366, + 72.82674848079681, + 72.88684828281403, + 72.93725447654724, + 73.03033211231232, + 73.20079249858856, + 73.33228698730468, + 73.50521060466767, + 73.73466797351837, + 74.04130051136016, + 74.44832262992858, + 75.16131977558135, + 76.2820675754547, + 76.88091934204103, + 76.94890914344788, + 77.14981128501891, + 77.46755316638948, + 78.03063362503056, + 78.24113952636718, + 78.38724931049347, + 78.70385138702399, + 80.69423282146485, + 83.1145306940078, + 83.92277062034601, + 84.44702100753784, + 84.44702100753784 + ], + "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": 0.9980037963516655, + "worker_service_time_ms": { + "distribution_type": "inverse_cdf", + "latency_ms": [ + 64.81042885780334, + 64.81042885780334, + 64.83034874725342, + 64.8652042169571, + 65.09434249591827, + 65.18473916053772, + 65.44755630970002, + 65.70904842376709, + 65.7790016708374, + 65.82948631858825, + 65.96403185939789, + 65.99917395210267, + 66.05419579982758, + 66.85226134777069, + 67.41213180541992, + 67.70523978710175, + 68.37709033489227, + 68.6659171819687, + 68.85855054855347, + 69.11082427978516, + 69.26426580905914, + 69.38339400291443, + 69.49818974494934, + 69.59014869213104, + 69.66760065555573, + 69.74625440597534, + 69.80519919395446, + 69.88650359630584, + 69.92661521434783, + 69.96951825618744, + 70.01145081996918, + 70.07099788188934, + 70.11137570858001, + 70.16706781387329, + 70.20938957214355, + 70.24025132656098, + 70.26918590068817, + 70.30251583099366, + 70.33475484848023, + 70.36707436561585, + 70.4130216884613, + 70.43617479801178, + 70.4693127822876, + 70.49378530979156, + 70.52594029903412, + 70.54812473297119, + 70.5624710559845, + 70.58988870143891, + 70.61189743995666, + 70.63650825500488, + 70.65514147758483, + 70.68650047779083, + 70.72054006099701, + 70.74340092658997, + 70.77034149169921, + 70.79453890323639, + 70.81376779079437, + 70.83277872562408, + 70.85774645805358, + 70.87755789279937, + 70.8952799797058, + 70.91331553459167, + 70.93301747322083, + 70.95305408000947, + 70.97179078102111, + 70.99544564723969, + 71.01263489723206, + 71.04232339382172, + 71.07317966461181, + 71.09767675876617, + 71.12224251270294, + 71.14307947158814, + 71.1765646982193, + 71.20110731601716, + 71.22757455825806, + 71.26078465461731, + 71.28300664424896, + 71.31065947055816, + 71.32941711902619, + 71.35752510547638, + 71.3804721879959, + 71.40329875946045, + 71.4374256324768, + 71.47053267478942, + 71.50463972568512, + 71.55248319149017, + 71.58942151069641, + 71.63497022628785, + 71.67235461711884, + 71.70529507637023, + 71.73997480869293, + 71.77655177116394, + 71.81513391971588, + 71.84466456413269, + 71.89355994224549, + 71.95444693088531, + 72.0025963306427, + 72.05938720703125, + 72.1252225112915, + 72.19015914440155, + 72.24081826210022, + 72.335604429245, + 72.48756666660309, + 72.61886465549469, + 72.82262350082398, + 73.05337915420532, + 73.28593082427979, + 73.7462468290329, + 74.27689339637756, + 75.49606513023376, + 75.9689773273468, + 76.06074317169188, + 76.27554173469544, + 76.56049769020082, + 77.12394835662843, + 77.4304215621948, + 77.7176802444458, + 77.9243575134278, + 80.04920254230532, + 82.06012575626357, + 83.2668972358703, + 83.78737902641296, + 83.78737902641296 + ], + "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/hopper/small-policy/sample-factory-qwenoft-rtx3090-v1/profile/latency_profile.png b/latency-aware/hopper/small-policy/sample-factory-qwenoft-rtx3090-v1/profile/latency_profile.png new file mode 100644 index 0000000000000000000000000000000000000000..9e1413a55e254fa732a9544836153c7f3f798a8b Binary files /dev/null and b/latency-aware/hopper/small-policy/sample-factory-qwenoft-rtx3090-v1/profile/latency_profile.png differ diff --git a/latency-aware/hopper/small-policy/sample-factory-qwenoft-rtx3090-v1/profile/profile.json b/latency-aware/hopper/small-policy/sample-factory-qwenoft-rtx3090-v1/profile/profile.json new file mode 100644 index 0000000000000000000000000000000000000000..f326bbb79f0cad25de15e0e64806a9ac00a43afc --- /dev/null +++ b/latency-aware/hopper/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": 2075, + "n_capacity_drops": 1, + "n_observation_attempts": 2076, + "per_slot_summary": { + "0": { + "admitted_count": 2075, + "mean_observation_to_action_latency_ms": 71.62769928219807, + "mean_worker_service_time_ms": 70.97048402751784, + "p95_observation_to_action_latency_ms": 74.10973887443542, + "p95_worker_service_time_ms": 73.34952919483185, + "p99_worker_service_time_ms": 76.0833806705475 + } + }, + "provenance": { + "base_config": "/workspace/tasks/20260911T023128Z-p-only4/hopper/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/hopper_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": "20260911T034939618615Z", + "summary": { + "frame_ms": 100.0, + "max_ms": 112.13530778884888, + "mean_effective_frames": 0.7162769928219807, + "mean_ms": 71.62769928219807, + "min_ms": 65.19399881362915, + "n_samples": 2075, + "p50_frames": 0.715749728679657, + "p50_ms": 71.5749728679657, + "p90_frames": 0.7304598674774171, + "p90_ms": 73.04598674774171, + "p95_frames": 0.7410973887443543, + "p95_ms": 74.10973887443542, + "p99_frames": 0.7696984185695647, + "p99_ms": 76.96984185695646, + "prob_latency_gt_1_frame": 0.00048192771084337347, + "prob_latency_gt_2_frames": 0.0, + "prob_latency_gt_3_frames": 0.0, + "std_ms": 2.032231876669252 + }, + "visualization_path": "latency_profile.png", + "workload_id": "hopper" +} diff --git a/latency-aware/hopper/small-policy/sample-factory-qwenoft-rtx3090-v1/provenance.json b/latency-aware/hopper/small-policy/sample-factory-qwenoft-rtx3090-v1/provenance.json new file mode 100644 index 0000000000000000000000000000000000000000..f38fffd29ae49fec0c2f07a74b40b4bcbae81c92 --- /dev/null +++ b/latency-aware/hopper/small-policy/sample-factory-qwenoft-rtx3090-v1/provenance.json @@ -0,0 +1,44 @@ +{ + "task": "hopper", + "model": "sample-factory-appo", + "training_condition": "latency-aware", + "training_run_id": "hopper_profile_20260911T180943Z", + "source": { + "repo_id": "latency-sensitive-bench/Standard-Pipeline", + "revision": "2208875f92b21e3f8ec2363a30a6705dfe56cc04", + "path": "hopper/profile_latency/small_model", + "url": "https://huggingface.co/latency-sensitive-bench/Standard-Pipeline/tree/2208875f92b21e3f8ec2363a30a6705dfe56cc04/hopper/profile_latency/small_model" + }, + "hf_repo_id": "latency-sensitive-bench/benchmark-models", + "hf_path": "latency-aware/hopper/small-policy/sample-factory-qwenoft-rtx3090-v1", + "checkpoint": { + "source_file": "hopper/profile_latency/small_model/checkpoint_p0/best_000015008_7684096_reward_3018.662.pth", + "source_sha256": "0cb2eb721fe26b34c39132082d2b15e098a5daccc6d0ed560f21ee7cacf6cb33", + "source_bytes": 76933, + "file": "checkpoint.pth", + "selection_rule": "best_reward_after_full_training_budget", + "method": "inference_export", + "sha256": "257d8c101292bd9f49a26f3c79601c5461bb5320c8b99e35f198b503756d2251", + "bytes": 27445, + "train_step": 15008, + "env_steps": 7684096, + "tensor_count": 15, + "tensor_equality_verified": true, + "removed_fields": [ + "best_performance", + "curr_lr", + "optimizer" + ] + }, + "config_source": "hopper/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/hopper/small-policy/sample-factory-qwenoft-rtx3090-v1/rollout_descriptor.json b/latency-aware/hopper/small-policy/sample-factory-qwenoft-rtx3090-v1/rollout_descriptor.json new file mode 100644 index 0000000000000000000000000000000000000000..e64193aba869ce585734cfba9cad269b5725320b --- /dev/null +++ b/latency-aware/hopper/small-policy/sample-factory-qwenoft-rtx3090-v1/rollout_descriptor.json @@ -0,0 +1,314 @@ +{ + "env_name": "hopper_rgb_state", + "integration_name": "gymnasium", + "action_spec": { + "layout": "gymnasium_continuous_v1", + "labels": [ + "thigh_torque", + "leg_torque", + "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 Hopper robot forward while keeping its torso upright. Predict three continuous torques in [-1, 1] ordered as thigh, leg, and foot.", + "checkpoint_experiment_dir": "/mnt/results/latency-sensitive-bench/profile-latency-new/hopper/20260911T180943Z-hopper/small_models/hopper_profile_20260911T180943Z", + "checkpoint_path": "/mnt/results/latency-sensitive-bench/profile-latency-new/hopper/20260911T180943Z-hopper/small_models/hopper_profile_20260911T180943Z/checkpoint_p0/best_000015008_7684096_reward_3018.662.pth", + "flat_cfg": { + "help": false, + "algo": "APPO", + "env": "latency_gymnasium", + "experiment": "hopper_profile_20260911T180943Z", + "train_dir": "/mnt/results/latency-sensitive-bench/profile-latency-new/hopper/20260911T180943Z-hopper/small_models", + "restart_behavior": "overwrite", + "device": "gpu", + "seed": 3333, + "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": 10000, + "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": false, + "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": 1.0, + "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": 10, + "decorrelate_envs_on_one_worker": true, + "actor_worker_gpus": [], + "set_workers_cpu_affinity": true, + "force_envs_single_thread": true, + "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": 3, + "load_checkpoint_kind": "latest", + "save_milestones_sec": -1, + "save_best_every_sec": 60, + "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": 4, + "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": "hopper_rgb_state", + "gym_env_id": "LatencyBench/Hopper-v0", + "gym_make_kwargs_json": "{\"base_env_id\": \"Hopper-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_hopper\"]", + "gym_action_space_json": "{\"dtype\": \"float32\", \"high\": [1.0, 1.0, 1.0], \"labels\": [\"thigh_torque\", \"leg_torque\", \"foot_torque\"], \"low\": [-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]", + "gym_base_prompt": "Move the Hopper robot forward while keeping its torso upright. Predict three continuous torques in [-1, 1] ordered as thigh, leg, and foot.", + "gym_state_labels_json": "[\"torso_height\", \"torso_angle\", \"thigh_angle\", \"leg_angle\", \"foot_angle\", \"torso_x_velocity\", \"torso_z_velocity\", \"torso_angular_velocity\", \"thigh_angular_velocity\", \"leg_angular_velocity\", \"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/hopper/20260911T180943Z-hopper/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": 10, + "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/hopper/20260911T180943Z-hopper/small_models/hopper_profile_20260911T180943Z/episode_metrics.jsonl", + "command_line": "--mode train --algo APPO --env latency_gymnasium --experiment hopper_profile_20260911T180943Z --train_dir /mnt/results/latency-sensitive-bench/profile-latency-new/hopper/20260911T180943Z-hopper/small_models --restart_behavior overwrite --device gpu --seed 3333 --episode_metrics_path /mnt/results/latency-sensitive-bench/profile-latency-new/hopper/20260911T180943Z-hopper/small_models/hopper_profile_20260911T180943Z/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 10000 --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 4 --gamma 0.99 --gae_lambda 0.95 --ppo_clip_ratio 0.2 --ppo_clip_value 1.0 --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 10 --default_niceness 0 --rnn_type gru --rnn_size 512 --save_every_sec 600 --keep_checkpoints 3 --save_milestones_sec -1 --save_best_every_sec 60 --save_best_after 100000 --stats_avg 100 --experiment_summaries_interval 10 --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 --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 False --with_vtrace False --decorrelate_envs_on_one_worker True --set_workers_cpu_affinity True --force_envs_single_thread True --actor_critic_share_weights True --with_wandb False --fasttd3_compile True --encoder_mlp_layers 64 64 --latency-type iid --latency-seed 0 --latency-profile-path /mnt/results/latency-sensitive-bench/profile-latency-new/hopper/20260911T180943Z-hopper/profile/profile.json --latency-profile-worker-slot 0 --add-latency-info False --eval-episodes 10 --eval-parallel-envs 1 --eval-max-steps 1000 --eval-deterministic True --gym-task-name hopper_rgb_state --gym-env-id LatencyBench/Hopper-v0 --gym-make-kwargs-json {\"base_env_id\": \"Hopper-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_hopper\"] --gym-action-space-json {\"dtype\": \"float32\", \"high\": [1.0, 1.0, 1.0], \"labels\": [\"thigh_torque\", \"leg_torque\", \"foot_torque\"], \"low\": [-1.0, -1.0, -1.0], \"type\": \"box\"} --gym-noop-action-json [0.0, 0.0, 0.0] --gym-base-prompt Move the Hopper robot forward while keeping its torso upright. Predict three continuous torques in [-1, 1] ordered as thigh, leg, and foot. --gym-state-labels-json [\"torso_height\", \"torso_angle\", \"thigh_angle\", \"leg_angle\", \"foot_angle\", \"torso_x_velocity\", \"torso_z_velocity\", \"torso_angular_velocity\", \"thigh_angular_velocity\", \"leg_angular_velocity\", \"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": "hopper_profile_20260911T180943Z", + "train_dir": "/mnt/results/latency-sensitive-bench/profile-latency-new/hopper/20260911T180943Z-hopper/small_models", + "restart_behavior": "overwrite", + "device": "gpu", + "seed": 3333, + "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": 10000, + "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, + "recurrence": 1, + "shuffle_minibatches": false, + "gamma": 0.99, + "reward_scale": 1.0, + "reward_clip": 1000.0, + "value_bootstrap": false, + "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": 1.0, + "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": 10, + "decorrelate_envs_on_one_worker": true, + "set_workers_cpu_affinity": true, + "force_envs_single_thread": true, + "default_niceness": 0, + "experiment_summaries_interval": 10, + "stats_avg": 100, + "train_for_env_steps": 10000000, + "save_every_sec": 600, + "keep_checkpoints": 3, + "save_milestones_sec": -1, + "save_best_every_sec": 60, + "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": 4, + "with_wandb": false, + "gym_task_name": "hopper_rgb_state", + "gym_env_id": "LatencyBench/Hopper-v0", + "gym_make_kwargs_json": "{\"base_env_id\": \"Hopper-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_hopper\"]", + "gym_action_space_json": "{\"dtype\": \"float32\", \"high\": [1.0, 1.0, 1.0], \"labels\": [\"thigh_torque\", \"leg_torque\", \"foot_torque\"], \"low\": [-1.0, -1.0, -1.0], \"type\": \"box\"}", + "gym_noop_action_json": "[0.0, 0.0, 0.0]", + "gym_base_prompt": "Move the Hopper robot forward while keeping its torso upright. Predict three continuous torques in [-1, 1] ordered as thigh, leg, and foot.", + "gym_state_labels_json": "[\"torso_height\", \"torso_angle\", \"thigh_angle\", \"leg_angle\", \"foot_angle\", \"torso_x_velocity\", \"torso_z_velocity\", \"torso_angular_velocity\", \"thigh_angular_velocity\", \"leg_angular_velocity\", \"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/hopper/20260911T180943Z-hopper/profile/profile.json", + "latency_profile_worker_slot": 0, + "add_latency_info": false, + "hold_policy": "hold", + "ordering_policy": "issue_order_fifo", + "eval_episodes": 10, + "eval_parallel_envs": 1, + "eval_max_steps": 1000, + "eval_deterministic": true, + "episode_metrics_path": "/mnt/results/latency-sensitive-bench/profile-latency-new/hopper/20260911T180943Z-hopper/small_models/hopper_profile_20260911T180943Z/episode_metrics.jsonl", + "output_dir": "/mnt/results/latency-sensitive-bench/profile-latency-new/hopper/20260911T180943Z-hopper/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/hopper/20260911T180943Z-hopper/small_train" + }, + "device_override": "gpu", + "latency_override": { + "method": "iid", + "fixed_latency_ms": null, + "profile_path": "/mnt/results/latency-sensitive-bench/profile-latency-new/hopper/20260911T180943Z-hopper/profile/profile.json", + "profile_worker_slot": 0, + "seed": 0, + "add_latency_info": false + } +} diff --git a/latency-aware/hopper/small-policy/sample-factory-qwenoft-rtx3090-v1/rollout_metadata.json b/latency-aware/hopper/small-policy/sample-factory-qwenoft-rtx3090-v1/rollout_metadata.json new file mode 100644 index 0000000000000000000000000000000000000000..e9a0de4a68ba55c1d49ee71613d4aa61448cd884 --- /dev/null +++ b/latency-aware/hopper/small-policy/sample-factory-qwenoft-rtx3090-v1/rollout_metadata.json @@ -0,0 +1,359 @@ +{ + "checkpoint_experiment_dir": "/mnt/results/latency-sensitive-bench/profile-latency-new/hopper/20260911T180943Z-hopper/small_models/hopper_profile_20260911T180943Z", + "checkpoint_path": "/mnt/results/latency-sensitive-bench/profile-latency-new/hopper/20260911T180943Z-hopper/small_models/hopper_profile_20260911T180943Z/checkpoint_p0/best_000015008_7684096_reward_3018.662.pth", + "config_source": "/mnt/results/latency-sensitive-bench/profile-latency-new/hopper/20260911T180943Z-hopper/small_models/hopper_profile_20260911T180943Z/config.json", + "latency_source": "/mnt/results/latency-sensitive-bench/profile-latency-new/hopper/20260911T180943Z-hopper/h100/small_train.yaml", + "checkpoint_config": { + "help": false, + "algo": "APPO", + "env": "latency_gymnasium", + "experiment": "hopper_profile_20260911T180943Z", + "train_dir": "/mnt/results/latency-sensitive-bench/profile-latency-new/hopper/20260911T180943Z-hopper/small_models", + "restart_behavior": "overwrite", + "device": "gpu", + "seed": 3333, + "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": 10000, + "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": false, + "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": 1.0, + "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": 10, + "decorrelate_envs_on_one_worker": true, + "actor_worker_gpus": [], + "set_workers_cpu_affinity": true, + "force_envs_single_thread": true, + "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": 3, + "load_checkpoint_kind": "latest", + "save_milestones_sec": -1, + "save_best_every_sec": 60, + "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": 4, + "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": "hopper_rgb_state", + "gym_env_id": "LatencyBench/Hopper-v0", + "gym_make_kwargs_json": "{\"base_env_id\": \"Hopper-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_hopper\"]", + "gym_action_space_json": "{\"dtype\": \"float32\", \"high\": [1.0, 1.0, 1.0], \"labels\": [\"thigh_torque\", \"leg_torque\", \"foot_torque\"], \"low\": [-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]", + "gym_base_prompt": "Move the Hopper robot forward while keeping its torso upright. Predict three continuous torques in [-1, 1] ordered as thigh, leg, and foot.", + "gym_state_labels_json": "[\"torso_height\", \"torso_angle\", \"thigh_angle\", \"leg_angle\", \"foot_angle\", \"torso_x_velocity\", \"torso_z_velocity\", \"torso_angular_velocity\", \"thigh_angular_velocity\", \"leg_angular_velocity\", \"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/hopper/20260911T180943Z-hopper/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": 10, + "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/hopper/20260911T180943Z-hopper/small_models/hopper_profile_20260911T180943Z/episode_metrics.jsonl", + "command_line": "--mode train --algo APPO --env latency_gymnasium --experiment hopper_profile_20260911T180943Z --train_dir /mnt/results/latency-sensitive-bench/profile-latency-new/hopper/20260911T180943Z-hopper/small_models --restart_behavior overwrite --device gpu --seed 3333 --episode_metrics_path /mnt/results/latency-sensitive-bench/profile-latency-new/hopper/20260911T180943Z-hopper/small_models/hopper_profile_20260911T180943Z/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 10000 --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 4 --gamma 0.99 --gae_lambda 0.95 --ppo_clip_ratio 0.2 --ppo_clip_value 1.0 --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 10 --default_niceness 0 --rnn_type gru --rnn_size 512 --save_every_sec 600 --keep_checkpoints 3 --save_milestones_sec -1 --save_best_every_sec 60 --save_best_after 100000 --stats_avg 100 --experiment_summaries_interval 10 --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 --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 False --with_vtrace False --decorrelate_envs_on_one_worker True --set_workers_cpu_affinity True --force_envs_single_thread True --actor_critic_share_weights True --with_wandb False --fasttd3_compile True --encoder_mlp_layers 64 64 --latency-type iid --latency-seed 0 --latency-profile-path /mnt/results/latency-sensitive-bench/profile-latency-new/hopper/20260911T180943Z-hopper/profile/profile.json --latency-profile-worker-slot 0 --add-latency-info False --eval-episodes 10 --eval-parallel-envs 1 --eval-max-steps 1000 --eval-deterministic True --gym-task-name hopper_rgb_state --gym-env-id LatencyBench/Hopper-v0 --gym-make-kwargs-json {\"base_env_id\": \"Hopper-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_hopper\"] --gym-action-space-json {\"dtype\": \"float32\", \"high\": [1.0, 1.0, 1.0], \"labels\": [\"thigh_torque\", \"leg_torque\", \"foot_torque\"], \"low\": [-1.0, -1.0, -1.0], \"type\": \"box\"} --gym-noop-action-json [0.0, 0.0, 0.0] --gym-base-prompt Move the Hopper robot forward while keeping its torso upright. Predict three continuous torques in [-1, 1] ordered as thigh, leg, and foot. --gym-state-labels-json [\"torso_height\", \"torso_angle\", \"thigh_angle\", \"leg_angle\", \"foot_angle\", \"torso_x_velocity\", \"torso_z_velocity\", \"torso_angular_velocity\", \"thigh_angular_velocity\", \"leg_angular_velocity\", \"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": "hopper_profile_20260911T180943Z", + "train_dir": "/mnt/results/latency-sensitive-bench/profile-latency-new/hopper/20260911T180943Z-hopper/small_models", + "restart_behavior": "overwrite", + "device": "gpu", + "seed": 3333, + "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": 10000, + "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, + "recurrence": 1, + "shuffle_minibatches": false, + "gamma": 0.99, + "reward_scale": 1.0, + "reward_clip": 1000.0, + "value_bootstrap": false, + "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": 1.0, + "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": 10, + "decorrelate_envs_on_one_worker": true, + "set_workers_cpu_affinity": true, + "force_envs_single_thread": true, + "default_niceness": 0, + "experiment_summaries_interval": 10, + "stats_avg": 100, + "train_for_env_steps": 10000000, + "save_every_sec": 600, + "keep_checkpoints": 3, + "save_milestones_sec": -1, + "save_best_every_sec": 60, + "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": 4, + "with_wandb": false, + "gym_task_name": "hopper_rgb_state", + "gym_env_id": "LatencyBench/Hopper-v0", + "gym_make_kwargs_json": "{\"base_env_id\": \"Hopper-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_hopper\"]", + "gym_action_space_json": "{\"dtype\": \"float32\", \"high\": [1.0, 1.0, 1.0], \"labels\": [\"thigh_torque\", \"leg_torque\", \"foot_torque\"], \"low\": [-1.0, -1.0, -1.0], \"type\": \"box\"}", + "gym_noop_action_json": "[0.0, 0.0, 0.0]", + "gym_base_prompt": "Move the Hopper robot forward while keeping its torso upright. Predict three continuous torques in [-1, 1] ordered as thigh, leg, and foot.", + "gym_state_labels_json": "[\"torso_height\", \"torso_angle\", \"thigh_angle\", \"leg_angle\", \"foot_angle\", \"torso_x_velocity\", \"torso_z_velocity\", \"torso_angular_velocity\", \"thigh_angular_velocity\", \"leg_angular_velocity\", \"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/hopper/20260911T180943Z-hopper/profile/profile.json", + "latency_profile_worker_slot": 0, + "add_latency_info": false, + "hold_policy": "hold", + "ordering_policy": "issue_order_fifo", + "eval_episodes": 10, + "eval_parallel_envs": 1, + "eval_max_steps": 1000, + "eval_deterministic": true, + "episode_metrics_path": "/mnt/results/latency-sensitive-bench/profile-latency-new/hopper/20260911T180943Z-hopper/small_models/hopper_profile_20260911T180943Z/episode_metrics.jsonl", + "output_dir": "/mnt/results/latency-sensitive-bench/profile-latency-new/hopper/20260911T180943Z-hopper/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/hopper/20260911T180943Z-hopper/small_train" + }, + "latency_override": { + "method": "iid", + "fixed_latency_ms": null, + "profile_path": "/mnt/results/latency-sensitive-bench/profile-latency-new/hopper/20260911T180943Z-hopper/profile/profile.json", + "profile_worker_slot": 0, + "seed": 0, + "add_latency_info": false + }, + "gymnasium_task": { + "task_name": "hopper_rgb_state", + "env_id": "LatencyBench/Hopper-v0", + "make_kwargs": { + "base_env_id": "Hopper-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_hopper" + ], + "base_prompt": "Move the Hopper robot forward while keeping its torso upright. Predict three continuous torques in [-1, 1] ordered as thigh, leg, and foot.", + "env_fps": 10.0, + "obs_fps": 10.0, + "frame_stack": 1, + "state_space": { + "labels": [ + "torso_height", + "torso_angle", + "thigh_angle", + "leg_angle", + "foot_angle", + "torso_x_velocity", + "torso_z_velocity", + "torso_angular_velocity", + "thigh_angular_velocity", + "leg_angular_velocity", + "foot_angular_velocity" + ] + }, + "action_space": { + "dtype": "float32", + "high": [ + 1.0, + 1.0, + 1.0 + ], + "labels": [ + "thigh_torque", + "leg_torque", + "foot_torque" + ], + "low": [ + -1.0, + -1.0, + -1.0 + ], + "type": "box" + }, + "noop_action": [ + 0.0, + 0.0, + 0.0 + ] + } +} diff --git a/latency-aware/hopper/small-policy/sample-factory-qwenoft-rtx3090-v1/selection.json b/latency-aware/hopper/small-policy/sample-factory-qwenoft-rtx3090-v1/selection.json new file mode 100644 index 0000000000000000000000000000000000000000..61b22c9fb8f1f7a3346c05349e6f57a081340afb --- /dev/null +++ b/latency-aware/hopper/small-policy/sample-factory-qwenoft-rtx3090-v1/selection.json @@ -0,0 +1,14 @@ +{ + "run_id": "20260911T180943Z-hopper", + "source_profile_run_id": "20260911T034939618615Z", + "selection_rule": "best_reward_after_full_training_budget", + "best_checkpoint": "/mnt/results/latency-sensitive-bench/profile-latency-new/hopper/20260911T180943Z-hopper/small_models/hopper_profile_20260911T180943Z/checkpoint_p0/best_000015008_7684096_reward_3018.662.pth", + "best_bundle_path": "checkpoint_p0/best_000015008_7684096_reward_3018.662.pth", + "final_checkpoint": "/mnt/results/latency-sensitive-bench/profile-latency-new/hopper/20260911T180943Z-hopper/small_models/hopper_profile_20260911T180943Z/checkpoint_p0/checkpoint_000019544_10006528.pth", + "final_env_steps": 10006528, + "final_train_step": 19544, + "training_exit_code": 0, + "profile_sha256": "f8fe2d26f5fee49b2df550cbef421169b930888c02ea62624479d7998fc53f8f", + "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/hopper/small-policy/sample-factory-qwenoft-rtx3090-v1/small_train.yaml b/latency-aware/hopper/small-policy/sample-factory-qwenoft-rtx3090-v1/small_train.yaml new file mode 100644 index 0000000000000000000000000000000000000000..665843a081ff2122af2333b819af493dafb53d7a --- /dev/null +++ b/latency-aware/hopper/small-policy/sample-factory-qwenoft-rtx3090-v1/small_train.yaml @@ -0,0 +1,167 @@ +experiment: + name: hopper_profile_20260911T180943Z + seed: 3333 +backend: + type: sample_factory + algo: APPO + device: cuda + train_dir: /mnt/results/latency-sensitive-bench/profile-latency-new/hopper/20260911T180943Z-hopper/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 + labels: + - thigh_torque + - leg_torque + - foot_torque + low: + - -1.0 + - -1.0 + - -1.0 + type: box + base_prompt: Move the Hopper robot forward while keeping its torso upright. Predict + three continuous torques in [-1, 1] ordered as thigh, leg, and foot. + env_fps: 10.0 + env_id: LatencyBench/Hopper-v0 + frame_stack: 1 + make_kwargs: + base_env_id: Hopper-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 + obs_fps: 10.0 + registration_imports: + - latency_bench.envs.gymnasium_hopper + state_space: + labels: + - torso_height + - torso_angle + - thigh_angle + - leg_angle + - foot_angle + - torso_x_velocity + - torso_z_velocity + - torso_angular_velocity + - thigh_angular_velocity + - leg_angular_velocity + - foot_angular_velocity + task_name: hopper_rgb_state +latency: + method: iid + fixed_latency_ms: null + profile_path: /mnt/results/latency-sensitive-bench/profile-latency-new/hopper/20260911T180943Z-hopper/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 + num_batches_to_accumulate: 2 + policy_workers_per_policy: 1 + max_policy_lag: 10000 + learning_rate: 0.00295 + lr_schedule: linear_decay + lr_schedule_kl_threshold: 0.008 + gamma: 0.99 + gae_lambda: 0.95 + ppo_clip_ratio: 0.2 + ppo_clip_value: 1.0 + value_loss_coeff: 1.3 + max_grad_norm: 3.5 + exploration_loss: entropy + exploration_loss_coeff: 0.0 + kl_loss_coeff: 0.1 + reward_scale: 1.0 + reward_clip: 1000.0 + async_rl: false + serial_mode: false + batched_sampling: false + with_vtrace: false + use_rnn: false + env_framestack: 4 + encoder_mlp_layers: + - 64 + - 64 + nonlinearity: tanh + adaptive_stddev: false + policy_initialization: torch_default + initial_stddev: 1.0 + actor_critic_share_weights: true + shuffle_minibatches: false + value_bootstrap: false + normalize_input: true + normalize_returns: true + decorrelate_experience_max_seconds: 10 + decorrelate_envs_on_one_worker: true + set_workers_cpu_affinity: true + force_envs_single_thread: true + save_every_sec: 600 + keep_checkpoints: 3 + save_best_every_sec: 60 + save_best_after: 100000 + continuous_tanh_scale: 0.0 + optimizer: adam + adam_eps: 1.0e-06 + adam_beta1: 0.9 + adam_beta2: 0.999 + obs_subtract_mean: 0.0 + obs_scale: 1.0 + default_niceness: 0 + rnn_type: gru + rnn_size: 512 + save_milestones_sec: -1 + stats_avg: 100 + experiment_summaries_interval: 10 + 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_sonic_decoder_path: null + with_wandb: false + fasttd3_compile: true +evaluation: + eval_interval_steps: null + eval_episodes: 10 + eval_parallel_envs: 1 + eval_max_steps: 1000 + eval_deterministic: true +logging: + output_dir: /mnt/results/latency-sensitive-bench/profile-latency-new/hopper/20260911T180943Z-hopper/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/hopper/small-policy/sample-factory-qwenpi_v3-rtx3090-v1/README.md b/latency-aware/hopper/small-policy/sample-factory-qwenpi_v3-rtx3090-v1/README.md new file mode 100644 index 0000000000000000000000000000000000000000..644caf5aef4f53d6d19f9204c8653f5416c6c42b --- /dev/null +++ b/latency-aware/hopper/small-policy/sample-factory-qwenpi_v3-rtx3090-v1/README.md @@ -0,0 +1,30 @@ +# hopper / sample-factory-appo + +Training condition: `latency-aware`. Run: `pi05_hopper_profile_appo_h1_5fps_g128_20260921`. + +Copied from [Standard-Pipeline](https://huggingface.co/latency-sensitive-bench/Standard-Pipeline/tree/2208875f92b21e3f8ec2363a30a6705dfe56cc04/hopper/profile_latency/small_model/Pi05) at `2208875f92b21e3f8ec2363a30a6705dfe56cc04`. + +- Checkpoint: `checkpoint.pth` +- Selection: final +- Checkpoint SHA256: `fe04cfa766b0953fcb63331a7277f432582585edc30f03651f69095f105c552c` +- 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/hopper/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/hopper/small-policy/sample-factory-qwenpi_v3-rtx3090-v1/checkpoint.pth b/latency-aware/hopper/small-policy/sample-factory-qwenpi_v3-rtx3090-v1/checkpoint.pth new file mode 100644 index 0000000000000000000000000000000000000000..cb37f42716e07e8b7fb8cd8b4cbcac14f460105a --- /dev/null +++ b/latency-aware/hopper/small-policy/sample-factory-qwenpi_v3-rtx3090-v1/checkpoint.pth @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fe04cfa766b0953fcb63331a7277f432582585edc30f03651f69095f105c552c +size 27403 diff --git a/latency-aware/hopper/small-policy/sample-factory-qwenpi_v3-rtx3090-v1/config.json b/latency-aware/hopper/small-policy/sample-factory-qwenpi_v3-rtx3090-v1/config.json new file mode 100644 index 0000000000000000000000000000000000000000..8d99ca8ff1111442e533a0834e71ae205860b205 --- /dev/null +++ b/latency-aware/hopper/small-policy/sample-factory-qwenpi_v3-rtx3090-v1/config.json @@ -0,0 +1,287 @@ +{ + "help": false, + "algo": "APPO", + "env": "latency_gymnasium", + "experiment": "pi05_hopper_profile_appo_h1_5fps_g128_20260921", + "train_dir": "${PI05_RUN_DIR}/profile_latency/teacher/checkpoints", + "restart_behavior": "overwrite", + "device": "gpu", + "seed": 3333, + "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": 10000, + "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": false, + "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": 1.0, + "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": 10, + "decorrelate_envs_on_one_worker": true, + "actor_worker_gpus": [], + "set_workers_cpu_affinity": true, + "force_envs_single_thread": true, + "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": 3, + "load_checkpoint_kind": "latest", + "save_milestones_sec": -1, + "save_best_every_sec": 60, + "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": 4, + "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": [ + "hopper", + "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": "hopper", + "gym_env_id": "LatencyBench/Hopper-v0", + "gym_make_kwargs_json": "{\"base_env_id\": \"Hopper-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_hopper\"]", + "gym_action_space_json": "{\"type\": \"box\", \"labels\": [\"thigh_torque\", \"leg_torque\", \"foot_torque\"], \"low\": [-1.0, -1.0, -1.0], \"high\": [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]", + "gym_base_prompt": "Move the Hopper robot forward while keeping its torso upright. Predict three continuous torques in [-1, 1] ordered as thigh, leg, and foot.", + "gym_state_labels_json": "[\"torso_height\", \"torso_angle\", \"thigh_angle\", \"leg_angle\", \"foot_angle\", \"torso_x_velocity\", \"torso_z_velocity\", \"torso_angular_velocity\", \"thigh_angular_velocity\", \"leg_angular_velocity\", \"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": 10, + "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_hopper_profile_appo_h1_5fps_g128_20260921/episode_metrics.jsonl", + "ppo": null, + "command_line": "--mode train --algo APPO --env latency_gymnasium --experiment pi05_hopper_profile_appo_h1_5fps_g128_20260921 --train_dir ${PI05_RUN_DIR}/profile_latency/teacher/checkpoints --restart_behavior overwrite --device gpu --seed 3333 --episode_metrics_path ${PI05_RUN_DIR}/profile_latency/teacher/checkpoints/pi05_hopper_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 10000 --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 --initial_stddev 1.0 --env_framestack 4 --gamma 0.99 --gae_lambda 0.95 --ppo_clip_ratio 0.2 --ppo_clip_value 1.0 --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 --decorrelate_experience_max_seconds 10 --save_every_sec 600 --keep_checkpoints 3 --save_best_every_sec 60 --save_best_after 100000 --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 False --with_vtrace False --decorrelate_envs_on_one_worker True --set_workers_cpu_affinity True --force_envs_single_thread True --actor_critic_share_weights True --encoder_mlp_layers 64 64 --latency-type iid --latency-seed 0 --latency-profile-path ${PI05_RUN_DIR}/profile_latency/profile/profile.json --latency-profile-worker-slot 0 --add-latency-info False --eval-episodes 10 --eval-parallel-envs 1 --eval-max-steps 1000 --eval-deterministic True --with_wandb True --wandb_project latency-sensitive-bench --wandb_group pi05-profile-teachers-h1-5fps-20260921 --wandb_job_type profile_teacher --wandb_tags hopper APPO Pi05 measured-IID-profile H1 5FPS --wandb_user dongqianyu99-zhejiang-university --gym-task-name hopper --gym-env-id LatencyBench/Hopper-v0 --gym-make-kwargs-json {\"base_env_id\": \"Hopper-v4\", \"render_mode\": \"rgb_array\", \"base_make_kwargs\": {\"forward_reward_weight\": 1.0, \"ctrl_cost_weight\": 0.001, \"healthy_reward\": 1.0, \"terminate_when_unhealthy\": true, \"reset_noise_scale\": 0.005, \"exclude_current_positions_from_observation\": true}} --gym-registration-imports-json [\"latency_bench.envs.gymnasium_hopper\"] --gym-action-space-json {\"type\": \"box\", \"labels\": [\"thigh_torque\", \"leg_torque\", \"foot_torque\"], \"low\": [-1.0, -1.0, -1.0], \"high\": [1.0, 1.0, 1.0], \"dtype\": \"float32\"} --gym-noop-action-json [0.0, 0.0, 0.0] --gym-base-prompt Move the Hopper robot forward while keeping its torso upright. Predict three continuous torques in [-1, 1] ordered as thigh, leg, and foot. --gym-state-labels-json [\"torso_height\", \"torso_angle\", \"thigh_angle\", \"leg_angle\", \"foot_angle\", \"torso_x_velocity\", \"torso_z_velocity\", \"torso_angular_velocity\", \"thigh_angular_velocity\", \"leg_angular_velocity\", \"foot_angular_velocity\"] --env-fps 5 --obs-fps 5 --frame-stack 1 --hold-policy hold --ordering-policy issue_order_fifo --use_env_info_cache False", + "cli_args": { + "algo": "APPO", + "env": "latency_gymnasium", + "experiment": "pi05_hopper_profile_appo_h1_5fps_g128_20260921", + "train_dir": "${PI05_RUN_DIR}/profile_latency/teacher/checkpoints", + "restart_behavior": "overwrite", + "device": "gpu", + "seed": 3333, + "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": 10000, + "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": false, + "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": 1.0, + "with_vtrace": false, + "max_grad_norm": 3.5, + "learning_rate": 0.00295, + "lr_schedule_kl_threshold": 0.008, + "normalize_input": true, + "decorrelate_experience_max_seconds": 10, + "decorrelate_envs_on_one_worker": true, + "set_workers_cpu_affinity": true, + "force_envs_single_thread": true, + "train_for_env_steps": 10000000, + "save_every_sec": 600, + "keep_checkpoints": 3, + "save_best_every_sec": 60, + "save_best_after": 100000, + "encoder_mlp_layers": [ + 64, + 64 + ], + "use_rnn": false, + "nonlinearity": "tanh", + "policy_initialization": "torch_default", + "actor_critic_share_weights": true, + "adaptive_stddev": false, + "initial_stddev": 1.0, + "use_env_info_cache": false, + "env_framestack": 4, + "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": [ + "hopper", + "APPO", + "Pi05", + "measured-IID-profile", + "H1", + "5FPS" + ], + "gym_task_name": "hopper", + "gym_env_id": "LatencyBench/Hopper-v0", + "gym_make_kwargs_json": "{\"base_env_id\": \"Hopper-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_hopper\"]", + "gym_action_space_json": "{\"type\": \"box\", \"labels\": [\"thigh_torque\", \"leg_torque\", \"foot_torque\"], \"low\": [-1.0, -1.0, -1.0], \"high\": [1.0, 1.0, 1.0], \"dtype\": \"float32\"}", + "gym_noop_action_json": "[0.0, 0.0, 0.0]", + "gym_base_prompt": "Move the Hopper robot forward while keeping its torso upright. Predict three continuous torques in [-1, 1] ordered as thigh, leg, and foot.", + "gym_state_labels_json": "[\"torso_height\", \"torso_angle\", \"thigh_angle\", \"leg_angle\", \"foot_angle\", \"torso_x_velocity\", \"torso_z_velocity\", \"torso_angular_velocity\", \"thigh_angular_velocity\", \"leg_angular_velocity\", \"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": "${PI05_RUN_DIR}/profile_latency/profile/profile.json", + "latency_profile_worker_slot": 0, + "add_latency_info": false, + "hold_policy": "hold", + "ordering_policy": "issue_order_fifo", + "eval_episodes": 10, + "eval_parallel_envs": 1, + "eval_max_steps": 1000, + "eval_deterministic": true, + "episode_metrics_path": "${PI05_RUN_DIR}/profile_latency/teacher/checkpoints/pi05_hopper_profile_appo_h1_5fps_g128_20260921/episode_metrics.jsonl", + "output_dir": "${PI05_RUN_DIR}/profile_latency/teacher/training", + "wandb_unique_id": "pi05_hopper_profile_appo_h1_5fps_g128_20260921" + }, + "git_hash": "unknown", + "git_repo_name": "not a git repository", + "eval_env_frameskip": 1, + "output_dir": "${PI05_RUN_DIR}/profile_latency/teacher/training", + "wandb_unique_id": "pi05_hopper_profile_appo_h1_5fps_g128_20260921" +} \ No newline at end of file diff --git a/latency-aware/hopper/small-policy/sample-factory-qwenpi_v3-rtx3090-v1/provenance.json b/latency-aware/hopper/small-policy/sample-factory-qwenpi_v3-rtx3090-v1/provenance.json new file mode 100644 index 0000000000000000000000000000000000000000..de6ae0d04d0400d4a4a61c3aea4a2a853ac22dc6 --- /dev/null +++ b/latency-aware/hopper/small-policy/sample-factory-qwenpi_v3-rtx3090-v1/provenance.json @@ -0,0 +1,40 @@ +{ + "task": "hopper", + "model": "sample-factory-appo", + "training_condition": "latency-aware", + "training_run_id": "pi05_hopper_profile_appo_h1_5fps_g128_20260921", + "source": { + "repo_id": "latency-sensitive-bench/Standard-Pipeline", + "revision": "2208875f92b21e3f8ec2363a30a6705dfe56cc04", + "path": "hopper/profile_latency/small_model/Pi05", + "url": "https://huggingface.co/latency-sensitive-bench/Standard-Pipeline/tree/2208875f92b21e3f8ec2363a30a6705dfe56cc04/hopper/profile_latency/small_model/Pi05" + }, + "hf_repo_id": "latency-sensitive-bench/benchmark-models", + "hf_path": "latency-aware/hopper/small-policy/sample-factory-qwenpi_v3-rtx3090-v1", + "checkpoint": { + "source_file": "hopper/profile_latency/small_model/Pi05/checkpoint_p0/selected.pth", + "source_sha256": "fe04cfa766b0953fcb63331a7277f432582585edc30f03651f69095f105c552c", + "source_bytes": 27403, + "file": "checkpoint.pth", + "selection_rule": "final", + "method": "copy", + "sha256": "fe04cfa766b0953fcb63331a7277f432582585edc30f03651f69095f105c552c", + "bytes": 27403, + "train_step": 19536, + "env_steps": 10002432, + "tensor_count": 15, + "tensor_equality_verified": true, + "removed_fields": [] + }, + "config_source": "hopper/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/hopper/small-policy/sample-factory-qwenpi_v3-rtx3090-v1/selection.json b/latency-aware/hopper/small-policy/sample-factory-qwenpi_v3-rtx3090-v1/selection.json new file mode 100644 index 0000000000000000000000000000000000000000..cef839948188afdc7258b6febe4f67d42e833d42 --- /dev/null +++ b/latency-aware/hopper/small-policy/sample-factory-qwenpi_v3-rtx3090-v1/selection.json @@ -0,0 +1,112 @@ +{ + "selected": "final", + "candidates": { + "final": { + "checkpoint": "${PI05_RUN_DIR}/profile_latency/teacher/checkpoints/pi05_hopper_profile_appo_h1_5fps_g128_20260921/checkpoint_p0/checkpoint_000019536_10002432.pth", + "sha256": "9b78ae32c795cf0eecbd2b66e04f2775091a9756a40e5f5635791ac2b2ca4b86", + "episodes": 20, + "seeds": [ + 104771, + 104772, + 104773, + 104774, + 104775, + 104776, + 104777, + 104778, + 104779, + 104780, + 104781, + 104782, + 104783, + 104784, + 104785, + 104786, + 104787, + 104788, + 104789, + 104790 + ], + "returns": [ + 2224.1403163381674, + 2819.128114077857, + 3489.646074260514, + 2009.7554991069278, + 2113.0389978032463, + 2530.889359173486, + 3528.0257119878297, + 3480.216905777773, + 1947.2163728885328, + 2019.6803983866998, + 271.37905453360946, + 2083.796228363727, + 3447.6906733909796, + 2591.8151761056497, + 265.78937272271213, + 1892.4932081793627, + 1981.7056002712736, + 3621.6136520722443, + 3614.0425221471028, + 3542.033646827985 + ], + "mean": 2473.704844220784, + "population_std": 983.5990191484575, + "strict_gt3000": 7 + }, + "training_best": { + "checkpoint": "${PI05_RUN_DIR}/profile_latency/teacher/checkpoints/pi05_hopper_profile_appo_h1_5fps_g128_20260921/checkpoint_p0/best_000019536_10002432_reward_2535.456.pth", + "sha256": "f0d71c11d6f468510a6b1e98d528060bfe213e094ae81a13a7c9129af2f5b2ec", + "episodes": 20, + "seeds": [ + 104771, + 104772, + 104773, + 104774, + 104775, + 104776, + 104777, + 104778, + 104779, + 104780, + 104781, + 104782, + 104783, + 104784, + 104785, + 104786, + 104787, + 104788, + 104789, + 104790 + ], + "returns": [ + 2224.1403163381674, + 2819.128114077857, + 3489.646074260514, + 2009.7554991069278, + 2113.0389978032463, + 2530.889359173486, + 3528.0257119878297, + 3480.216905777773, + 1947.2163728885328, + 2019.6803983866998, + 271.37905453360946, + 2083.796228363727, + 3447.6906733909796, + 2591.8151761056497, + 265.78937272271213, + 1892.4932081793627, + 1981.7056002712736, + 3621.6136520722443, + 3614.0425221471028, + 3542.033646827985 + ], + "mean": 2473.704844220784, + "population_std": 983.5990191484575, + "strict_gt3000": 7 + } + }, + "completed_at": "2026-09-21T14:06:00.799410+00:00", + "final_checkpoint_env_steps": 10002432, + "final_checkpoint_train_step": 19536 +} diff --git a/latency-aware/hopper/small-policy/sample-factory-qwenpi_v3-rtx3090-v1/source/README.md b/latency-aware/hopper/small-policy/sample-factory-qwenpi_v3-rtx3090-v1/source/README.md new file mode 100644 index 0000000000000000000000000000000000000000..84887ce5b7ebf59d3ee18cc6fc6d77a5a636df43 --- /dev/null +++ b/latency-aware/hopper/small-policy/sample-factory-qwenpi_v3-rtx3090-v1/source/README.md @@ -0,0 +1,7 @@ +# Hopper Pi0.5 profile-latency APPO checkpoint + +Selected `final` at train step 19536 / 10002432 Sample Factory environment steps, seed 3333, native state/action 11/3, and 5/5 FPS. The selected paired evaluation mean was 2473.704844 (population SD 983.599019); 7 of 20 returns were strictly above 3000. P profile SHA256 `1289653bf6e8591cdd5cf805ad539ddd1cb041cdad08a9f2833597fe72271e1e` (publication revision `e1a9bdd566b96d5da4703f769d40eab656af491e`). + +Original probe accepted 2/12 under episode_raw_return >3000. A later, separately authorized dataset used >2500; that data gate does not change this teacher audit. P session-1 warmup-drift warning is retained in verification.json. + +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/hopper/small-policy/sample-factory-qwenpi_v3-rtx3090-v1/source/provenance.json b/latency-aware/hopper/small-policy/sample-factory-qwenpi_v3-rtx3090-v1/source/provenance.json new file mode 100644 index 0000000000000000000000000000000000000000..be3962fa2368ead70f1c5b436c197490db6f998a --- /dev/null +++ b/latency-aware/hopper/small-policy/sample-factory-qwenpi_v3-rtx3090-v1/source/provenance.json @@ -0,0 +1,51 @@ +{ + "state": "SELECTED_TEACHER_INFERENCE_EXPORT_CPU_RELOAD_VERIFIED", + "task": "hopper", + "selected": "final", + "source_checkpoint": "checkpoint_000019536_10002432.pth", + "source_checkpoint_sha256": "9b78ae32c795cf0eecbd2b66e04f2775091a9756a40e5f5635791ac2b2ca4b86", + "export_checkpoint": "checkpoint_p0/selected.pth", + "export_checkpoint_sha256": "fe04cfa766b0953fcb63331a7277f432582585edc30f03651f69095f105c552c", + "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": 19536, + "selected_env_steps": 10002432, + "actual_final_train_step": 19536, + "actual_final_env_steps": 10002432, + "budget_sample_factory_env_steps": 10000000, + "seed": 3333, + "native_state_dim": 11, + "native_action_dim": 3, + "profile_sha256": "1289653bf6e8591cdd5cf805ad539ddd1cb041cdad08a9f2833597fe72271e1e", + "profile_publication_revision": "e1a9bdd566b96d5da4703f769d40eab656af491e", + "audit_state": "TEACHER_FULL_TRACE_AUDIT_PASSED", + "audit_gate": "PASSED", + "selected_mean": 2473.704844220784, + "selected_population_std": 983.5990191484575, + "selected_strict_gt3000_episodes": 7, + "data": { + "state": "PUBLISHED_SEPARATELY", + "revision": "c974fe66922af6cadd425696b60133753e955566", + "accepted": 100, + "attempts": 218, + "rows": 97180, + "retained_plus_new": "53+47", + "gate": "episode_raw_return >2500" + }, + "source_sha256": { + "config.json": "b3c5bdc0727ff0358ed5f04c957687fd4ac2f838b74c4d1842d6cf5b8efb4f1b", + "teacher.yaml": "522195b46f4a951b43682c8c186cb1509d1ac8e179f62c7874e55a034f05aaa4", + "selection.json": "68eeacb6897edba524166194486359ca0278efeb6f27dba236ff3c4a7d0c4a8e", + "verification.json": "0c132bddf56fabe6dcf9d2cfddd37942e459e97b718b4816a6a9549ad50c801c" + } +} diff --git a/latency-aware/hopper/small-policy/sample-factory-qwenpi_v3-rtx3090-v1/teacher.yaml b/latency-aware/hopper/small-policy/sample-factory-qwenpi_v3-rtx3090-v1/teacher.yaml new file mode 100644 index 0000000000000000000000000000000000000000..de6cfc577074ddb09c0f414b604358964125f837 --- /dev/null +++ b/latency-aware/hopper/small-policy/sample-factory-qwenpi_v3-rtx3090-v1/teacher.yaml @@ -0,0 +1,158 @@ +env: + state_space: + labels: + - torso_height + - torso_angle + - thigh_angle + - leg_angle + - foot_angle + - torso_x_velocity + - torso_z_velocity + - torso_angular_velocity + - thigh_angular_velocity + - leg_angular_velocity + - foot_angular_velocity + name: gymnasium + task_name: hopper + env_id: LatencyBench/Hopper-v0 + registration_imports: + - latency_bench.envs.gymnasium_hopper + make_kwargs: + base_env_id: Hopper-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: + - thigh_torque + - leg_torque + - foot_torque + low: + - -1.0 + - -1.0 + - -1.0 + high: + - 1.0 + - 1.0 + - 1.0 + dtype: float32 + noop_action: + - 0.0 + - 0.0 + - 0.0 + base_prompt: Move the Hopper robot forward while keeping its torso upright. Predict + three continuous torques in [-1, 1] ordered as thigh, leg, and foot. +experiment: + name: pi05_hopper_profile_appo_h1_5fps_g128_20260921 + seed: 3333 +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 +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 + num_batches_to_accumulate: 2 + policy_workers_per_policy: 1 + max_policy_lag: 10000 + learning_rate: 0.00295 + lr_schedule: linear_decay + lr_schedule_kl_threshold: 0.008 + gamma: 0.99 + gae_lambda: 0.95 + ppo_clip_ratio: 0.2 + ppo_clip_value: 1.0 + value_loss_coeff: 1.3 + max_grad_norm: 3.5 + exploration_loss: entropy + exploration_loss_coeff: 0.0 + kl_loss_coeff: 0.1 + reward_scale: 1.0 + reward_clip: 1000.0 + async_rl: false + serial_mode: false + batched_sampling: false + with_vtrace: false + use_rnn: false + env_framestack: 4 + encoder_mlp_layers: + - 64 + - 64 + nonlinearity: tanh + adaptive_stddev: false + policy_initialization: torch_default + initial_stddev: 1.0 + actor_critic_share_weights: true + shuffle_minibatches: false + value_bootstrap: false + normalize_input: true + normalize_returns: true + decorrelate_experience_max_seconds: 10 + decorrelate_envs_on_one_worker: true + set_workers_cpu_affinity: true + force_envs_single_thread: true + save_every_sec: 600 + keep_checkpoints: 3 + save_best_every_sec: 60 + save_best_after: 100000 +evaluation: + eval_interval_steps: null + eval_episodes: 10 + eval_parallel_envs: 1 + eval_max_steps: 1000 + eval_deterministic: true +logging: + output_dir: ${PI05_RUN_DIR}/profile_latency/teacher/training + video: + enabled: false + num_bins: 1 + save_step_records: false + save_action_records: false + save_latency_records: false + wandb_project: latency-sensitive-bench + wandb_group: pi05-profile-teachers-h1-5fps-20260921 + wandb_name: pi05_hopper_profile_appo_h1_5fps_g128_20260921 + wandb_job_type: profile_teacher + wandb_tags: + - hopper + - APPO + - Pi05 + - measured-IID-profile + - H1 + - 5FPS diff --git a/latency-aware/hopper/small-policy/sample-factory-qwenpi_v3-rtx3090-v1/verification.json b/latency-aware/hopper/small-policy/sample-factory-qwenpi_v3-rtx3090-v1/verification.json new file mode 100644 index 0000000000000000000000000000000000000000..3dcee7f2f5ff412c7278488850c0dba92522e695 --- /dev/null +++ b/latency-aware/hopper/small-policy/sample-factory-qwenpi_v3-rtx3090-v1/verification.json @@ -0,0 +1,386 @@ +{ + "state": "TEACHER_FULL_TRACE_AUDIT_PASSED", + "verified_at": "2026-09-21T14:09:31.112834+00:00", + "task": "hopper", + "model_revision": "408930d8a85f0cb316dc3bdebfb8142526570355", + "P_profile_sha256": "1289653bf6e8591cdd5cf805ad539ddd1cb041cdad08a9f2833597fe72271e1e", + "P_warnings": [ + { + "session": 1, + "warmup_status": "unstable", + "warmup_stability_reason": "drifting", + "samples_excluded": 100 + } + ], + "actual_final_env_steps": 10002432, + "actual_final_train_step": 19536, + "selection": { + "selected": "final", + "candidates": { + "final": { + "checkpoint": "${PI05_RUN_DIR}/profile_latency/teacher/checkpoints/pi05_hopper_profile_appo_h1_5fps_g128_20260921/checkpoint_p0/checkpoint_000019536_10002432.pth", + "sha256": "9b78ae32c795cf0eecbd2b66e04f2775091a9756a40e5f5635791ac2b2ca4b86", + "episodes": 20, + "seeds": [ + 104771, + 104772, + 104773, + 104774, + 104775, + 104776, + 104777, + 104778, + 104779, + 104780, + 104781, + 104782, + 104783, + 104784, + 104785, + 104786, + 104787, + 104788, + 104789, + 104790 + ], + "returns": [ + 2224.1403163381674, + 2819.128114077857, + 3489.646074260514, + 2009.7554991069278, + 2113.0389978032463, + 2530.889359173486, + 3528.0257119878297, + 3480.216905777773, + 1947.2163728885328, + 2019.6803983866998, + 271.37905453360946, + 2083.796228363727, + 3447.6906733909796, + 2591.8151761056497, + 265.78937272271213, + 1892.4932081793627, + 1981.7056002712736, + 3621.6136520722443, + 3614.0425221471028, + 3542.033646827985 + ], + "mean": 2473.704844220784, + "population_std": 983.5990191484575, + "strict_gt3000": 7 + }, + "training_best": { + "checkpoint": "${PI05_RUN_DIR}/profile_latency/teacher/checkpoints/pi05_hopper_profile_appo_h1_5fps_g128_20260921/checkpoint_p0/best_000019536_10002432_reward_2535.456.pth", + "sha256": "f0d71c11d6f468510a6b1e98d528060bfe213e094ae81a13a7c9129af2f5b2ec", + "episodes": 20, + "seeds": [ + 104771, + 104772, + 104773, + 104774, + 104775, + 104776, + 104777, + 104778, + 104779, + 104780, + 104781, + 104782, + 104783, + 104784, + 104785, + 104786, + 104787, + 104788, + 104789, + 104790 + ], + "returns": [ + 2224.1403163381674, + 2819.128114077857, + 3489.646074260514, + 2009.7554991069278, + 2113.0389978032463, + 2530.889359173486, + 3528.0257119878297, + 3480.216905777773, + 1947.2163728885328, + 2019.6803983866998, + 271.37905453360946, + 2083.796228363727, + 3447.6906733909796, + 2591.8151761056497, + 265.78937272271213, + 1892.4932081793627, + 1981.7056002712736, + 3621.6136520722443, + 3614.0425221471028, + 3542.033646827985 + ], + "mean": 2473.704844220784, + "population_std": 983.5990191484575, + "strict_gt3000": 7 + } + }, + "completed_at": "2026-09-21T14:06:00.799410+00:00", + "final_checkpoint_env_steps": 10002432, + "final_checkpoint_train_step": 19536 + }, + "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": [ + 2224.1403163381674, + 2819.128114077857, + 3489.646074260514, + 2009.7554991069278, + 2113.0389978032463, + 2530.889359173486, + 3528.0257119878297, + 3480.216905777773, + 1947.2163728885328, + 2019.6803983866998, + 271.37905453360946, + 2083.796228363727, + 3447.6906733909796, + 2591.8151761056497, + 265.78937272271213, + 1892.4932081793627, + 1981.7056002712736, + 3621.6136520722443, + 3614.0425221471028, + 3542.033646827985 + ], + "lengths": [ + 594, + 757, + 1000, + 544, + 593, + 682, + 1000, + 1000, + 528, + 539, + 134, + 566, + 919, + 698, + 132, + 508, + 539, + 1000, + 983, + 1000 + ], + "mean_return": 2473.704844220784, + "population_std_return": 983.5990191484575, + "mean_length": 685.8, + "raw_steps": 13716, + "strict_gt3000": 7, + "actions": 10606, + "action_latency_mean_ms": 199.08640804375509, + "action_latency_p95_ms": 256.6691669408085, + "commands_requiring_native_clip": 0, + "invalid_actions": 0, + "dropped_actions": 0, + "submitted_observations": 10606, + "dropped_observations": 3110, + "observation_drop_fraction": 0.2267424905220181, + "config_sha256": "39166d114a7d4b46e64da14382bd644f4095e7eb36fc6eae52479c2545093f81", + "state_contract": "Native state11 is bound to manifest/loader QA/saved reload; 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": [ + 2224.1403163381674, + 2819.128114077857, + 3489.646074260514, + 2009.7554991069278, + 2113.0389978032463, + 2530.889359173486, + 3528.0257119878297, + 3480.216905777773, + 1947.2163728885328, + 2019.6803983866998, + 271.37905453360946, + 2083.796228363727, + 3447.6906733909796, + 2591.8151761056497, + 265.78937272271213, + 1892.4932081793627, + 1981.7056002712736, + 3621.6136520722443, + 3614.0425221471028, + 3542.033646827985 + ], + "lengths": [ + 594, + 757, + 1000, + 544, + 593, + 682, + 1000, + 1000, + 528, + 539, + 134, + 566, + 919, + 698, + 132, + 508, + 539, + 1000, + 983, + 1000 + ], + "mean_return": 2473.704844220784, + "population_std_return": 983.5990191484575, + "mean_length": 685.8, + "raw_steps": 13716, + "strict_gt3000": 7, + "actions": 10606, + "action_latency_mean_ms": 199.08640804375509, + "action_latency_p95_ms": 256.6691669408085, + "commands_requiring_native_clip": 0, + "invalid_actions": 0, + "dropped_actions": 0, + "submitted_observations": 10606, + "dropped_observations": 3110, + "observation_drop_fraction": 0.2267424905220181, + "config_sha256": "8447bbc671595a369933dff026480db4f94846a7970e4e8ac6731abdf21d0138", + "state_contract": "Native state11 is bound to manifest/loader QA/saved reload; no per-step state vector is recorded." + } + }, + "E10": { + "episodes": 10, + "seeds": [ + 104771, + 104772, + 104773, + 104774, + 104775, + 104776, + 104777, + 104778, + 104779, + 104780 + ], + "returns": [ + 2224.1403163381674, + 2819.128114077857, + 3489.646074260514, + 2009.7554991069278, + 2113.0389978032463, + 2530.889359173486, + 3528.0257119878297, + 3480.216905777773, + 1947.2163728885328, + 2019.6803983866998 + ], + "lengths": [ + 594, + 757, + 1000, + 544, + 593, + 682, + 1000, + 1000, + 528, + 539 + ], + "mean_return": 2616.173774980103, + "population_std_return": 629.868919379035, + "mean_length": 723.7, + "raw_steps": 7237, + "strict_gt3000": 3, + "actions": 5577, + "action_latency_mean_ms": 199.29742636488007, + "action_latency_p95_ms": 257.7097693485515, + "commands_requiring_native_clip": 0, + "invalid_actions": 0, + "dropped_actions": 0, + "submitted_observations": 5577, + "dropped_observations": 1660, + "observation_drop_fraction": 0.22937681359679427, + "config_sha256": "8a5246d083413a735ed88e2e3b9a5c4f7b49896abfd3153488d9fcd1edd62176", + "state_contract": "Native state11 is bound to manifest/loader QA/saved reload; no per-step state vector is recorded." + }, + "probe": { + "attempts": 12, + "accepted": 2, + "rejected": 10, + "source_state_sha256": "fad846336f923f7f0bf755c2d5cdbbcc2a8c4f955556c2299ef088cbd20d1f72", + "accepted_specs": [ + { + "attempt_idx": 4, + "episode_idx": 0, + "episode_length": 1000, + "episode_raw_frames": 1000, + "episode_raw_return": 3573.1784443855286, + "seed": 3337, + "split": "train" + }, + { + "attempt_idx": 9, + "episode_idx": 1, + "episode_length": 1000, + "episode_raw_frames": 1000, + "episode_raw_return": 3552.9572883844376, + "seed": 3342, + "split": "val" + } + ] + }, + "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/hopper/vla/starvla-qwengr00t-h1/hopper_profile_gr00t_h1_5fps_2h100_20260917/F-summary.json b/latency-aware/hopper/vla/starvla-qwengr00t-h1/hopper_profile_gr00t_h1_5fps_2h100_20260917/F-summary.json new file mode 100644 index 0000000000000000000000000000000000000000..e516fd2b08adec7150b3ce6df921ddf98a0cf676 --- /dev/null +++ b/latency-aware/hopper/vla/starvla-qwengr00t-h1/hopper_profile_gr00t_h1_5fps_2h100_20260917/F-summary.json @@ -0,0 +1,163 @@ +{ + "verified_at": "2026-09-17T21:34:33.830761+00:00", + "task": "hopper", + "B": { + "episodes": 20, + "mean_return": 8.553798918701421, + "population_std": 0.09164445349807723, + "mean_length": 9, + "mean_episode_latency_ms": 111.10969192814082, + "observation_drops": 0, + "submitted_observations": 180, + "action_drops": 0, + "invalid_actions": 0, + "metrics_sha256": "15e0fff2f282cce88a408993c1d1c1c608a8e534af55df802f5019924bb2e9b0" + }, + "F": { + "episodes": 20, + "mean_return": 242.55055665478992, + "population_std": 127.57369659782253, + "mean_length": 100.1, + "mean_episode_latency_ms": 110.7924141313037, + "observation_drops": 0, + "submitted_observations": 2002, + "action_drops": 0, + "invalid_actions": 0, + "metrics_sha256": "7e98ffea9db1d79cea521fcba68cf2a1ad76d818eae318a09aaa98e37a70aceb" + }, + "paired": [ + { + "seed": 42, + "B": 8.601128245693165, + "F": 343.6761117591365, + "delta": 335.0749835134433 + }, + { + "seed": 43, + "B": 8.698451451997723, + "F": 77.48860293690754, + "delta": 68.79015148490981 + }, + { + "seed": 44, + "B": 8.675803868427353, + "F": 112.99006876226332, + "delta": 104.31426489383597 + }, + { + "seed": 45, + "B": 8.530831452718134, + "F": 342.75501631999356, + "delta": 334.2241848672754 + }, + { + "seed": 46, + "B": 8.616775268138941, + "F": 82.57466802682492, + "delta": 73.95789275868599 + }, + { + "seed": 47, + "B": 8.399036294860874, + "F": 346.3568388160152, + "delta": 337.95780252115435 + }, + { + "seed": 48, + "B": 8.527887103971324, + "F": 342.9612816173764, + "delta": 334.4333945134051 + }, + { + "seed": 49, + "B": 8.517139018075015, + "F": 116.06720384039482, + "delta": 107.55006482231981 + }, + { + "seed": 50, + "B": 8.448905777753668, + "F": 171.36873558575093, + "delta": 162.91982980799725 + }, + { + "seed": 51, + "B": 8.57681750029125, + "F": 107.03821019089881, + "delta": 98.46139269060755 + }, + { + "seed": 52, + "B": 8.499147733543529, + "F": 407.27526018834806, + "delta": 398.77611245480455 + }, + { + "seed": 53, + "B": 8.580534329357635, + "F": 168.11576973272216, + "delta": 159.5352354033645 + }, + { + "seed": 54, + "B": 8.693173128612607, + "F": 112.27675191351292, + "delta": 103.58357878490031 + }, + { + "seed": 55, + "B": 8.503254728109283, + "F": 345.791405216921, + "delta": 337.28815048881177 + }, + { + "seed": 56, + "B": 8.617020606791158, + "F": 371.0030376043078, + "delta": 362.38601699751666 + }, + { + "seed": 57, + "B": 8.438460003247506, + "F": 409.7465920290405, + "delta": 401.30813202579304 + }, + { + "seed": 58, + "B": 8.423442879493416, + "F": 404.90933966988376, + "delta": 396.48589679039037 + }, + { + "seed": 59, + "B": 8.493276290159583, + "F": 345.40185098954646, + "delta": 336.90857469938686 + }, + { + "seed": 60, + "B": 8.707425662635234, + "F": 74.85872734440912, + "delta": 66.15130168177389 + }, + { + "seed": 61, + "B": 8.527467030151016, + "F": 168.3556605515446, + "delta": 159.8281935213936 + } + ], + "wins": 20, + "ties": 0, + "losses": 0, + "mean_difference": 233.99675773608848, + "F_step_interval_ms": { + "mean": 200.0077628149011, + "min": 198.96644609980285, + "max": 201.19602186605334 + }, + "F_revision": "d33af26b7f87357c1b7a2b75bae4923bd90b6cc1", + "F_checkpoint_sha256": "fe4fb89a88d9aab7f266a69921b32cfc00a0e2964664be3953988a8f4062d3ca", + "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/hopper/vla/starvla-qwengr00t-h1/hopper_profile_gr00t_h1_5fps_2h100_20260917/F-verification.json b/latency-aware/hopper/vla/starvla-qwengr00t-h1/hopper_profile_gr00t_h1_5fps_2h100_20260917/F-verification.json new file mode 100644 index 0000000000000000000000000000000000000000..e95a88c429df2afd87c8332c4d1de05b01722e58 --- /dev/null +++ b/latency-aware/hopper/vla/starvla-qwengr00t-h1/hopper_profile_gr00t_h1_5fps_2h100_20260917/F-verification.json @@ -0,0 +1,6 @@ +{ + "state": "F20_AND_MATCHED_COMPARISON_VERIFIED", + "summary": "/workspace/lzj/latency-sensitive-bench/runs/hopper/gr00t_h1_5fps_20260917/F-summary.json", + "steps": 2002, + "verified_at": "2026-09-17T21:34:33.830761+00:00" +} diff --git a/latency-aware/hopper/vla/starvla-qwengr00t-h1/hopper_profile_gr00t_h1_5fps_2h100_20260917/README.md b/latency-aware/hopper/vla/starvla-qwengr00t-h1/hopper_profile_gr00t_h1_5fps_2h100_20260917/README.md new file mode 100644 index 0000000000000000000000000000000000000000..146d34e28726d1ddca7b2ad2f0ce4d194e898187 --- /dev/null +++ b/latency-aware/hopper/vla/starvla-qwengr00t-h1/hopper_profile_gr00t_h1_5fps_2h100_20260917/README.md @@ -0,0 +1,31 @@ +# hopper / qwengr00t + +Training condition: `latency-aware`. Run: `hopper_profile_gr00t_h1_5fps_2h100_20260917`. + +Copied from [Standard-Pipeline](https://huggingface.co/latency-sensitive-bench/Standard-Pipeline/tree/2208875f92b21e3f8ec2363a30a6705dfe56cc04/hopper/profile_latency/GR00T) at `2208875f92b21e3f8ec2363a30a6705dfe56cc04`. + +- Checkpoint: `checkpoints/model.pt` +- Selection: only published VLA checkpoint +- Checkpoint SHA256: `fe4fb89a88d9aab7f266a69921b32cfc00a0e2964664be3953988a8f4062d3ca` +- 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/hopper/vla/starvla-qwengr00t-h1/hopper_profile_gr00t_h1_5fps_2h100_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/hopper/vla/starvla-qwengr00t-h1/hopper_profile_gr00t_h1_5fps_2h100_20260917/checkpoints/model.pt b/latency-aware/hopper/vla/starvla-qwengr00t-h1/hopper_profile_gr00t_h1_5fps_2h100_20260917/checkpoints/model.pt new file mode 100644 index 0000000000000000000000000000000000000000..089d8a475b85a7e0abad2677d08327d8237ce177 --- /dev/null +++ b/latency-aware/hopper/vla/starvla-qwengr00t-h1/hopper_profile_gr00t_h1_5fps_2h100_20260917/checkpoints/model.pt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fe4fb89a88d9aab7f266a69921b32cfc00a0e2964664be3953988a8f4062d3ca +size 9976837923 diff --git a/latency-aware/hopper/vla/starvla-qwengr00t-h1/hopper_profile_gr00t_h1_5fps_2h100_20260917/config.full.yaml b/latency-aware/hopper/vla/starvla-qwengr00t-h1/hopper_profile_gr00t_h1_5fps_2h100_20260917/config.full.yaml new file mode 100644 index 0000000000000000000000000000000000000000..5523c9163f5518dd5f39ab16b33813d7199cd331 --- /dev/null +++ b/latency-aware/hopper/vla/starvla-qwengr00t-h1/hopper_profile_gr00t_h1_5fps_2h100_20260917/config.full.yaml @@ -0,0 +1,242 @@ +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: 3 + state_dim: 11 + 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: 3 + 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/hopper/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: 8 + 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 + labels: + - thigh_torque + - leg_torque + - foot_torque + low: + - -1.0 + - -1.0 + - -1.0 + type: box + base_prompt: Move the Hopper robot forward while keeping its torso upright. + Predict three continuous torques in [-1, 1] ordered as thigh, leg, and foot. + env_fps: 5.0 + env_id: LatencyBench/Hopper-v0 + frame_stack: 1 + make_kwargs: + base_env_id: Hopper-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 + obs_fps: 5.0 + registration_imports: + - latency_bench.envs.gymnasium_hopper + state_space: + labels: + - torso_height + - torso_angle + - thigh_angle + - leg_angle + - foot_angle + - torso_x_velocity + - torso_z_velocity + - torso_angular_velocity + - thigh_angular_velocity + - leg_angular_velocity + - foot_angular_velocity + task_name: hopper + active_action_dim: 3 +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: 4 + 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: hopper_profile_gr00t_h1_5fps_2h100_20260917 +run_root_dir: /mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/hopper/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: hopper_profile_gr00t_h1_5fps_2h100_20260917 +wandb_group: gr00t-six-env-h1-5fps +wandb_tags: +- hopper +- 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/hopper/profile_latency/train.yaml +output_dir: /mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/hopper/profile_latency/vla/training/hopper_profile_gr00t_h1_5fps_2h100_20260917 diff --git a/latency-aware/hopper/vla/starvla-qwengr00t-h1/hopper_profile_gr00t_h1_5fps_2h100_20260917/config.yaml b/latency-aware/hopper/vla/starvla-qwengr00t-h1/hopper_profile_gr00t_h1_5fps_2h100_20260917/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..facb7ca9a0efe47a98a064e8482f40a6af911c2f --- /dev/null +++ b/latency-aware/hopper/vla/starvla-qwengr00t-h1/hopper_profile_gr00t_h1_5fps_2h100_20260917/config.yaml @@ -0,0 +1,119 @@ +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: 3 + state_dim: 11 + 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: 3 + 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 + labels: + - thigh_torque + - leg_torque + - foot_torque + low: + - -1.0 + - -1.0 + - -1.0 + type: box + base_prompt: Move the Hopper robot forward while keeping its torso upright. + Predict three continuous torques in [-1, 1] ordered as thigh, leg, and foot. + env_fps: 5.0 + env_id: LatencyBench/Hopper-v0 + frame_stack: 1 + make_kwargs: + base_env_id: Hopper-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 + obs_fps: 5.0 + registration_imports: + - latency_bench.envs.gymnasium_hopper + state_space: + labels: + - torso_height + - torso_angle + - thigh_angle + - leg_angle + - foot_angle + - torso_x_velocity + - torso_z_velocity + - torso_angular_velocity + - thigh_angular_velocity + - leg_angular_velocity + - foot_angular_velocity + task_name: hopper +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/hopper/vla/starvla-qwengr00t-h1/hopper_profile_gr00t_h1_5fps_2h100_20260917/dataset_statistics.json b/latency-aware/hopper/vla/starvla-qwengr00t-h1/hopper_profile_gr00t_h1_5fps_2h100_20260917/dataset_statistics.json new file mode 100644 index 0000000000000000000000000000000000000000..c3fb2fda90b074253215e23138b1db53de3835cf --- /dev/null +++ b/latency-aware/hopper/vla/starvla-qwengr00t-h1/hopper_profile_gr00t_h1_5fps_2h100_20260917/dataset_statistics.json @@ -0,0 +1,123 @@ +{ + "new_embodiment": { + "action": { + "mean": [ + 0.37568429112434387, + 0.12602593004703522, + 0.18286840617656708 + ], + "std": [ + 0.5004534125328064, + 0.5383408665657043, + 0.7935799360275269 + ], + "max": [ + 1.0, + 1.0, + 1.0 + ], + "min": [ + -1.0, + -1.0, + -1.0 + ], + "q01": [ + -1.0, + -1.0, + -1.0 + ], + "q99": [ + 1.0, + 1.0, + 1.0 + ], + "mask": [ + true, + true, + true + ] + }, + "state": { + "mean": [ + 0.18549983203411102, + -0.47602275013923645, + 0.6153591275215149, + -0.20731453597545624, + 0.3875531256198883, + 0.07606308162212372, + -0.010008123703300953, + -0.10106530040502548, + -0.10059873759746552, + -0.015130510553717613, + 0.010883210226893425 + ], + "std": [ + 0.5010573267936707, + 0.500958263874054, + 0.4302949011325836, + 0.5819286108016968, + 0.6522682905197144, + 0.29966363310813904, + 0.4919671416282654, + 0.32692041993141174, + 0.3827664256095886, + 0.48192644119262695, + 0.6321708559989929 + ], + "max": [ + 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 + ], + "q01": [ + -0.8911992394924164, + -0.9787489545345306, + -0.8409363627433777, + -0.9166498214006424, + -0.9754678171873092, + -0.779312139749527, + -0.8334154105186462, + -0.907303871512413, + -0.934087490439415, + -1.0, + -1.0 + ], + "q99": [ + 0.9115053570270537, + 0.8891597759723663, + 0.983530193567276, + 0.9664776229858397, + 0.983356647491455, + 0.921393676996231, + 0.949374679327011, + 0.8133167958259582, + 1.0, + 1.0, + 1.0 + ] + }, + "num_transitions": 90000, + "num_trajectories": 90 + } +} \ No newline at end of file diff --git a/latency-aware/hopper/vla/starvla-qwengr00t-h1/hopper_profile_gr00t_h1_5fps_2h100_20260917/latency_prompt_map.json b/latency-aware/hopper/vla/starvla-qwengr00t-h1/hopper_profile_gr00t_h1_5fps_2h100_20260917/latency_prompt_map.json new file mode 100644 index 0000000000000000000000000000000000000000..b8b9505554698daf03386329f526ba09ec3ac8c9 --- /dev/null +++ b/latency-aware/hopper/vla/starvla-qwengr00t-h1/hopper_profile_gr00t_h1_5fps_2h100_20260917/latency_prompt_map.json @@ -0,0 +1,7 @@ +{ + "1": { + "prompt": "Move the Hopper robot forward while keeping its torso upright. Predict three continuous torques in [-1, 1] ordered as thigh, leg, and 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/hopper/vla/starvla-qwengr00t-h1/hopper_profile_gr00t_h1_5fps_2h100_20260917/manifest.json b/latency-aware/hopper/vla/starvla-qwengr00t-h1/hopper_profile_gr00t_h1_5fps_2h100_20260917/manifest.json new file mode 100644 index 0000000000000000000000000000000000000000..7b94e7e21f78e79624e085ff33079580d8982c2c --- /dev/null +++ b/latency-aware/hopper/vla/starvla-qwengr00t-h1/hopper_profile_gr00t_h1_5fps_2h100_20260917/manifest.json @@ -0,0 +1,142 @@ +{ + "dataset_name": "hopper_h1_5fps_profile", + "env_name": "hopper", + "episodes": 90, + "frames": 90000, + "task_prompts": [ + "Move the Hopper robot forward while keeping its torso upright. Predict three continuous torques in [-1, 1] ordered as thigh, leg, and 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/hopper/profile_latency/data/raw_profile", + "integration_name": "gymnasium", + "task_name": "hopper", + "action_layout": "gymnasium_continuous_v1", + "action_labels": [ + "thigh_torque", + "leg_torque", + "foot_torque" + ], + "carrier_action_labels": [ + "thigh_torque", + "leg_torque", + "foot_torque" + ], + "action_dim": 3, + "active_action_dim": 3, + "action_carrier": "native", + "rows_unit": "decision_step", + "obs_fps": 5.0, + "obs_stride_raw_frames": 1, + "uses_state": true, + "state_dim": 11, + "state_labels": [ + "torso_height", + "torso_angle", + "thigh_angle", + "leg_angle", + "foot_angle", + "torso_x_velocity", + "torso_z_velocity", + "torso_angular_velocity", + "thigh_angular_velocity", + "leg_angular_velocity", + "foot_angular_velocity" + ], + "state_normalization": { + "type": "min_max", + "min": [ + 1.018517255783081, + -0.18582995235919952, + -0.8983789682388306, + -1.3514255285263062, + -0.9186444282531738, + -0.011477897875010967, + -3.1067798137664795, + -5.069515705108643, + -8.289618492126465, + -10.0, + -10.0 + ], + "max": [ + 1.693691611289978, + 0.09866146743297577, + 0.08487974107265472, + 0.12726113200187683, + 0.87968510389328, + 5.243149757385254, + 3.2616147994995117, + 6.1643218994140625, + 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/hopper/profile_latency/data/lerobot/hopper_h1_5fps_profile/latency_prompt_map.json", + "custom_mixtures_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/hopper/profile_latency/data/lerobot/_generated_mixtures/hopper_h1_5fps_profile.json", + "gymnasium_task": { + "action_space": { + "dtype": "float32", + "high": [ + 1.0, + 1.0, + 1.0 + ], + "labels": [ + "thigh_torque", + "leg_torque", + "foot_torque" + ], + "low": [ + -1.0, + -1.0, + -1.0 + ], + "type": "box" + }, + "base_prompt": "Move the Hopper robot forward while keeping its torso upright. Predict three continuous torques in [-1, 1] ordered as thigh, leg, and foot.", + "env_fps": 5.0, + "env_id": "LatencyBench/Hopper-v0", + "frame_stack": 1, + "make_kwargs": { + "base_env_id": "Hopper-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 + ], + "obs_fps": 5.0, + "registration_imports": [ + "latency_bench.envs.gymnasium_hopper" + ], + "state_space": { + "labels": [ + "torso_height", + "torso_angle", + "thigh_angle", + "leg_angle", + "foot_angle", + "torso_x_velocity", + "torso_z_velocity", + "torso_angular_velocity", + "thigh_angular_velocity", + "leg_angular_velocity", + "foot_angular_velocity" + ] + }, + "task_name": "hopper" + }, + "validation_dataset_name": "hopper_h1_5fps_profile__val", + "validation_episodes": 10, + "validation_frames": 10000 +} \ No newline at end of file diff --git a/latency-aware/hopper/vla/starvla-qwengr00t-h1/hopper_profile_gr00t_h1_5fps_2h100_20260917/provenance.json b/latency-aware/hopper/vla/starvla-qwengr00t-h1/hopper_profile_gr00t_h1_5fps_2h100_20260917/provenance.json new file mode 100644 index 0000000000000000000000000000000000000000..9d63f1ab7fed2372268668719b9fe6ed17952f94 --- /dev/null +++ b/latency-aware/hopper/vla/starvla-qwengr00t-h1/hopper_profile_gr00t_h1_5fps_2h100_20260917/provenance.json @@ -0,0 +1,34 @@ +{ + "task": "hopper", + "model": "qwengr00t", + "training_condition": "latency-aware", + "training_run_id": "hopper_profile_gr00t_h1_5fps_2h100_20260917", + "source": { + "repo_id": "latency-sensitive-bench/Standard-Pipeline", + "revision": "2208875f92b21e3f8ec2363a30a6705dfe56cc04", + "path": "hopper/profile_latency/GR00T", + "url": "https://huggingface.co/latency-sensitive-bench/Standard-Pipeline/tree/2208875f92b21e3f8ec2363a30a6705dfe56cc04/hopper/profile_latency/GR00T" + }, + "hf_repo_id": "latency-sensitive-bench/benchmark-models", + "hf_path": "latency-aware/hopper/vla/starvla-qwengr00t-h1/hopper_profile_gr00t_h1_5fps_2h100_20260917", + "checkpoint": { + "source_file": "hopper/profile_latency/GR00T/checkpoints/model.pt", + "source_sha256": "fe4fb89a88d9aab7f266a69921b32cfc00a0e2964664be3953988a8f4062d3ca", + "source_bytes": 9976837923, + "file": "checkpoints/model.pt", + "selection_rule": "only published VLA checkpoint", + "method": "copy", + "sha256": "fe4fb89a88d9aab7f266a69921b32cfc00a0e2964664be3953988a8f4062d3ca", + "bytes": 9976837923 + }, + "config_source": "hopper/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/hopper/vla/starvla-qwengr00t-h1/hopper_profile_gr00t_h1_5fps_2h100_20260917/reload-validation.json b/latency-aware/hopper/vla/starvla-qwengr00t-h1/hopper_profile_gr00t_h1_5fps_2h100_20260917/reload-validation.json new file mode 100644 index 0000000000000000000000000000000000000000..5df2e7bb409922ed87471e635e83a102dd0acdbd --- /dev/null +++ b/latency-aware/hopper/vla/starvla-qwengr00t-h1/hopper_profile_gr00t_h1_5fps_2h100_20260917/reload-validation.json @@ -0,0 +1,23 @@ +{ + "verified_at": "2026-09-17T18:41:37.060205+00:00", + "state": "TRAINING_AND_SAVED_BUNDLE_FORWARD_VERIFIED", + "training_updates": 5000, + "checkpoint_sha256": "fe4fb89a88d9aab7f266a69921b32cfc00a0e2964664be3953988a8f4062d3ca", + "action_horizon": 1, + "state_dim": 11, + "action_dim": 3, + "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, + 3 + ], + "forward_finite": true, + "F20_status": "not yet run", + "acceptance": "not established by this engineering check" +} diff --git a/latency-aware/hopper/vla/starvla-qwengr00t-h1/hopper_profile_gr00t_h1_5fps_2h100_20260917/source/README.md b/latency-aware/hopper/vla/starvla-qwengr00t-h1/hopper_profile_gr00t_h1_5fps_2h100_20260917/source/README.md new file mode 100644 index 0000000000000000000000000000000000000000..5e827a2c679a4f0ba444a4b162f3a5bb28f874c4 --- /dev/null +++ b/latency-aware/hopper/vla/starvla-qwengr00t-h1/hopper_profile_gr00t_h1_5fps_2h100_20260917/source/README.md @@ -0,0 +1,5 @@ +# Hopper GR00T profile H1 /5FPS + +Fresh5000update training, savedbundle verification and matchedRTX3090 measuredF20 are complete. B=8.553799 ± 0.091644; F=242.550557 ± 127.573697; 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 f572d99f0e14d858c8a21a8f3fd182ecb34f4795. Original reload-validation remains a historical snapshot. Evaluatedweights unchanged at d33af26b7f87357c1b7a2b75bae4923bd90b6cc1. diff --git a/latency-aware/hopper/vla/starvla-qwengr00t-h1/hopper_profile_gr00t_h1_5fps_2h100_20260917/source/provenance.json b/latency-aware/hopper/vla/starvla-qwengr00t-h1/hopper_profile_gr00t_h1_5fps_2h100_20260917/source/provenance.json new file mode 100644 index 0000000000000000000000000000000000000000..4918d30f97909940a497bb07f5f719e8d3ebe314 --- /dev/null +++ b/latency-aware/hopper/vla/starvla-qwengr00t-h1/hopper_profile_gr00t_h1_5fps_2h100_20260917/source/provenance.json @@ -0,0 +1,163 @@ +{ + "task": "hopper", + "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": "hopper_profile_gr00t_h1_5fps_2h100_20260917", + "condition": "profile_latency", + "source": { + "code_source": { + "repo": "latency-sensitive-bench/Standard-Pipeline", + "revision": "571cb5801720488ed6458b9cfe3f4c595b069cea", + "prefix": "hopper/zero_latency/shared/source_100ep_v1/demonstrations/raw", + "sha256": { + "metadata.json": "6f6ba3dc2556aa2c16f81defde6231458cbe051d082ba7e054150ba1b65e39cd", + "train.parquet": "fcd0e418fbc33c531235c5765c1b5d3288dacd5b0a7bdcc6b288f1d4e1009b75", + "val.parquet": "1e53ba71d9efa4dd8cb8839a2e40206a9d682480c753e2bcdab8c3f417836e20" + }, + "code": { + "root": "1681edb8d9e4f7d10b7dfa647670e01e0d5cce37", + "sample_factory": "4b11d8754ffc989cdd15a340cfc9be0277377d5b", + "starvla": "1d0d7b139d1725cab268cb9dae2007ef4c5d05d2" + } + }, + "teacher_selection": { + "selected": "training_best", + "candidates": { + "final": { + "checkpoint": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/hopper/profile_latency/teacher/checkpoints/hopper_profile_appo_h1_5fps_20260917/checkpoint_p0/checkpoint_000019536_10002432.pth", + "sha256": "d7bccf0797e5622069c1af0aa67796863bd68c40707cd96c7e7399fa27ca49e0", + "episodes": 20, + "returns": [ + 3813.902369620451, + 3814.717929846383, + 3815.678995110681, + 3813.5158104903485, + 3812.1721732868205, + 3822.869610539937, + 3816.213698287465, + 3819.3530723476006, + 3809.7239266677057, + 3819.2709564714014, + 3804.324024566971, + 3818.1522276093633, + 3829.262769910777, + 3815.7688288666577, + 3814.218115130112, + 3802.188637269843, + 3814.133230818948, + 3811.2177504405886, + 3803.348027121752, + 3819.7451215416377 + ], + "mean": 3814.488863797272, + "std": 6.329159120614397, + "strict_gt3000": 20 + }, + "training_best": { + "checkpoint": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/hopper/profile_latency/teacher/checkpoints/hopper_profile_appo_h1_5fps_20260917/checkpoint_p0/best_000019512_9990144_reward_3220.937.pth", + "sha256": "f8b6089e67f129a7c17504bc8d6146003177b260e927cc1281a178d26f404b74", + "episodes": 20, + "returns": [ + 3818.3299991080344, + 3801.949979062483, + 3799.39904827517, + 3806.0984441131777, + 3821.3468930346507, + 3818.6013992377857, + 3825.5948641715036, + 3817.364507562392, + 3814.7261006971084, + 3825.008281517956, + 3816.453969095774, + 3815.17650963725, + 3816.6675065195645, + 3810.7786398491785, + 3813.106914295106, + 3803.3492543466577, + 3815.0139371581436, + 3819.671617704402, + 3825.458467648868, + 3825.03671743075 + ], + "mean": 3815.4566525232976, + "std": 7.632802007802108, + "strict_gt3000": 20 + } + }, + "completed_at": "2026-09-17T14:44:34.549937+00:00" + }, + "profile": { + "state": "PROFILE_TEACHER_CONFIG_READY", + "profile": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/hopper/profile_latency/profile_3090_20260917T141828786022Z/profile.json", + "profile_sha256": "9d2e175b36914cc07fe93711db34ea89593c6c71e2b2a3d7fbe9176173ab7184", + "source_recipe": "/home/ubuntu/lzj/latency-sensitive-bench/configs/examples/gymnasium/hopper/small_model_train.yaml", + "source_recipe_sha256": "2e22333ef7e8a77e175654f9952e5c22f03af26590fb6d5bc543c2d399b63350", + "config": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/hopper/profile_latency/teacher.yaml", + "algo": "APPO", + "budget_env_steps": 10000000, + "seed": 3333, + "fps": 5, + "latency": "iid", + "initialization": "fresh", + "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": 3794.1470665335655, + "max_replay_return": 3829.4147548675537, + "sha256": "f527c237b3189b010e2e19639ada8ca2a03eaae28de00749559e61ff78c54684" + }, + "val": { + "episodes": 10, + "frames": 10000, + "min_replay_return": 3793.609395623207, + "max_replay_return": 3827.4473445415497, + "sha256": "746e4b49e9d5e3378af13b5a6efe04efd2b1ecaf10ca7ecfaeaad0e58a3e64e2" + } + }, + "dataset": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/hopper/profile_latency/data/lerobot/hopper_h1_5fps_profile", + "condition": "profile_latency", + "action_horizon": 1, + "state_normalization": { + "type": "min_max", + "min": [ + 1.018517255783081, + -0.18582995235919952, + -0.8983789682388306, + -1.3514255285263062, + -0.9186444282531738, + -0.011477897875010967, + -3.1067798137664795, + -5.069515705108643, + -8.289618492126465, + -10.0, + -10.0 + ], + "max": [ + 1.693691611289978, + 0.09866146743297577, + 0.08487974107265472, + 0.12726113200187683, + 0.87968510389328, + 5.243149757385254, + 3.2616147994995117, + 6.1643218994140625, + 10.0, + 10.0, + 10.0 + ] + }, + "prompt_key": 1, + "fresh_vla_initialization": true + }, + "training_config_sha256": "82d27c50c369031f3a84b92d20275452ba3961dc7f26c9d9af2c0ff82f8b5bf3", + "dataset_manifest_sha256": "124b98d6005f55f1637cb21e0fd7c64726026c1fcdbb34bd64c74c3436794985" +} diff --git a/latency-aware/hopper/vla/starvla-qwengr00t-h1/hopper_profile_gr00t_h1_5fps_2h100_20260917/task_contract.json b/latency-aware/hopper/vla/starvla-qwengr00t-h1/hopper_profile_gr00t_h1_5fps_2h100_20260917/task_contract.json new file mode 100644 index 0000000000000000000000000000000000000000..674ea753e4b61eb10e05434c9a1bdea8f4f28f61 --- /dev/null +++ b/latency-aware/hopper/vla/starvla-qwengr00t-h1/hopper_profile_gr00t_h1_5fps_2h100_20260917/task_contract.json @@ -0,0 +1,62 @@ +{ + "action_space": { + "dtype": "float32", + "high": [ + 1.0, + 1.0, + 1.0 + ], + "labels": [ + "thigh_torque", + "leg_torque", + "foot_torque" + ], + "low": [ + -1.0, + -1.0, + -1.0 + ], + "type": "box" + }, + "base_prompt": "Move the Hopper robot forward while keeping its torso upright. Predict three continuous torques in [-1, 1] ordered as thigh, leg, and foot.", + "env_fps": 5.0, + "env_id": "LatencyBench/Hopper-v0", + "frame_stack": 1, + "make_kwargs": { + "base_env_id": "Hopper-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 + ], + "obs_fps": 5.0, + "registration_imports": [ + "latency_bench.envs.gymnasium_hopper" + ], + "state_space": { + "labels": [ + "torso_height", + "torso_angle", + "thigh_angle", + "leg_angle", + "foot_angle", + "torso_x_velocity", + "torso_z_velocity", + "torso_angular_velocity", + "thigh_angular_velocity", + "leg_angular_velocity", + "foot_angular_velocity" + ] + }, + "task_name": "hopper" +} diff --git a/latency-aware/hopper/vla/starvla-qwenoft-h1/hopper_rgb_state_profile_20260911T180943Z_openvla_native_continuous_projector_sft_5k/README.md b/latency-aware/hopper/vla/starvla-qwenoft-h1/hopper_rgb_state_profile_20260911T180943Z_openvla_native_continuous_projector_sft_5k/README.md new file mode 100644 index 0000000000000000000000000000000000000000..6b7f7d353647c7c4ef93b0857679a3149257a4dc --- /dev/null +++ b/latency-aware/hopper/vla/starvla-qwenoft-h1/hopper_rgb_state_profile_20260911T180943Z_openvla_native_continuous_projector_sft_5k/README.md @@ -0,0 +1,29 @@ +# hopper / qwenoft + +Training condition: `latency-aware`. Run: `hopper_rgb_state_profile_20260911T180943Z_openvla_native_continuous_projector_sft_5k`. + +Copied from [Standard-Pipeline](https://huggingface.co/latency-sensitive-bench/Standard-Pipeline/tree/2208875f92b21e3f8ec2363a30a6705dfe56cc04/hopper/profile_latency/OpenVLA) at `2208875f92b21e3f8ec2363a30a6705dfe56cc04`. + +- Checkpoint: `checkpoints/model.pt` +- Selection: only published VLA checkpoint +- Checkpoint SHA256: `2c9456b87aaba7d89cd53481047badde32513375f31056c4464a72642cacbe42` +- 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/hopper/vla/starvla-qwenoft-h1/hopper_rgb_state_profile_20260911T180943Z_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/hopper/vla/starvla-qwenoft-h1/hopper_rgb_state_profile_20260911T180943Z_openvla_native_continuous_projector_sft_5k/checkpoints/model.pt b/latency-aware/hopper/vla/starvla-qwenoft-h1/hopper_rgb_state_profile_20260911T180943Z_openvla_native_continuous_projector_sft_5k/checkpoints/model.pt new file mode 100644 index 0000000000000000000000000000000000000000..6250389136e78a938f582fb53c6ddb698c7b496c --- /dev/null +++ b/latency-aware/hopper/vla/starvla-qwenoft-h1/hopper_rgb_state_profile_20260911T180943Z_openvla_native_continuous_projector_sft_5k/checkpoints/model.pt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2c9456b87aaba7d89cd53481047badde32513375f31056c4464a72642cacbe42 +size 9785081249 diff --git a/latency-aware/hopper/vla/starvla-qwenoft-h1/hopper_rgb_state_profile_20260911T180943Z_openvla_native_continuous_projector_sft_5k/config.full.yaml b/latency-aware/hopper/vla/starvla-qwenoft-h1/hopper_rgb_state_profile_20260911T180943Z_openvla_native_continuous_projector_sft_5k/config.full.yaml new file mode 100644 index 0000000000000000000000000000000000000000..ab49725208bc9ae23deed3e3c0413919637433b4 --- /dev/null +++ b/latency-aware/hopper/vla/starvla-qwenoft-h1/hopper_rgb_state_profile_20260911T180943Z_openvla_native_continuous_projector_sft_5k/config.full.yaml @@ -0,0 +1,358 @@ +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: 3 + state_dim: 11 + 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: 3 + 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/hopper/20260911T180943Z-hopper/converted + data_mix: hopper_rgb_state_profile_20260911T180943Z + eval_data_mix: hopper_rgb_state_profile_20260911T180943Z__val + custom_mixtures_path: /mnt/results/latency-sensitive-bench/profile-latency-new/hopper/20260911T180943Z-hopper/converted/_generated_mixtures/hopper_rgb_state_profile_20260911T180943Z.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 + labels: + - thigh_torque + - leg_torque + - foot_torque + low: + - -1.0 + - -1.0 + - -1.0 + type: box + base_prompt: Move the Hopper robot forward while keeping its torso upright. + Predict three continuous torques in [-1, 1] ordered as thigh, leg, and foot. + env_fps: 10.0 + env_id: LatencyBench/Hopper-v0 + frame_stack: 1 + make_kwargs: + base_env_id: Hopper-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 + obs_fps: 10.0 + registration_imports: + - latency_bench.envs.gymnasium_hopper + state_space: + labels: + - torso_height + - torso_angle + - thigh_angle + - leg_angle + - foot_angle + - torso_x_velocity + - torso_z_velocity + - torso_angular_velocity + - thigh_angular_velocity + - leg_angular_velocity + - foot_angular_velocity + task_name: hopper_rgb_state + active_action_dim: 3 +dataset: + source_hf: '' + config_name: null + source_subdir: null + converted_name: hopper_rgb_state_profile_20260911T180943Z + single_source_hf: '' + mixed_source_hf: '' + single_converted_name: hopper_rgb_state_profile_20260911T180943Z + mixed_converted_name: hopper_rgb_state_profile_20260911T180943Z + 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/hopper/20260911T180943Z-hopper/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/hopper/20260911T180943Z-hopper/vla + dataset_local_dir: /mnt/results/latency-sensitive-bench/profile-latency-new/hopper/20260911T180943Z-hopper/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/hopper/20260911T180943Z-hopper/converted/hopper_rgb_state_profile_20260911T180943Z/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 + labels: + - thigh_torque + - leg_torque + - foot_torque + low: + - -1.0 + - -1.0 + - -1.0 + type: box + base_prompt: Move the Hopper robot forward while keeping its torso upright. + Predict three continuous torques in [-1, 1] ordered as thigh, leg, and foot. + env_fps: 10.0 + env_id: LatencyBench/Hopper-v0 + frame_stack: 1 + make_kwargs: + base_env_id: Hopper-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 + obs_fps: 10.0 + registration_imports: + - latency_bench.envs.gymnasium_hopper + state_space: + labels: + - torso_height + - torso_angle + - thigh_angle + - leg_angle + - foot_angle + - torso_x_velocity + - torso_z_velocity + - torso_angular_velocity + - thigh_angular_velocity + - leg_angular_velocity + - foot_angular_velocity + task_name: hopper_rgb_state + task_name: hopper_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: hopper_rgb_state_profile_20260911T180943Z_openvla_native_continuous_projector_sft_5k +output_dir: /mnt/results/latency-sensitive-bench/profile-latency-new/hopper/20260911T180943Z-hopper/vla/hopper_rgb_state_profile_20260911T180943Z_openvla_native_continuous_projector_sft_5k +config_yaml: /mnt/results/latency-sensitive-bench/profile-latency-new/hopper/20260911T180943Z-hopper/h100/vla_train.yaml +is_debug: false +version_id: '0.21' diff --git a/latency-aware/hopper/vla/starvla-qwenoft-h1/hopper_rgb_state_profile_20260911T180943Z_openvla_native_continuous_projector_sft_5k/config.yaml b/latency-aware/hopper/vla/starvla-qwenoft-h1/hopper_rgb_state_profile_20260911T180943Z_openvla_native_continuous_projector_sft_5k/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..ab49725208bc9ae23deed3e3c0413919637433b4 --- /dev/null +++ b/latency-aware/hopper/vla/starvla-qwenoft-h1/hopper_rgb_state_profile_20260911T180943Z_openvla_native_continuous_projector_sft_5k/config.yaml @@ -0,0 +1,358 @@ +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: 3 + state_dim: 11 + 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: 3 + 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/hopper/20260911T180943Z-hopper/converted + data_mix: hopper_rgb_state_profile_20260911T180943Z + eval_data_mix: hopper_rgb_state_profile_20260911T180943Z__val + custom_mixtures_path: /mnt/results/latency-sensitive-bench/profile-latency-new/hopper/20260911T180943Z-hopper/converted/_generated_mixtures/hopper_rgb_state_profile_20260911T180943Z.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 + labels: + - thigh_torque + - leg_torque + - foot_torque + low: + - -1.0 + - -1.0 + - -1.0 + type: box + base_prompt: Move the Hopper robot forward while keeping its torso upright. + Predict three continuous torques in [-1, 1] ordered as thigh, leg, and foot. + env_fps: 10.0 + env_id: LatencyBench/Hopper-v0 + frame_stack: 1 + make_kwargs: + base_env_id: Hopper-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 + obs_fps: 10.0 + registration_imports: + - latency_bench.envs.gymnasium_hopper + state_space: + labels: + - torso_height + - torso_angle + - thigh_angle + - leg_angle + - foot_angle + - torso_x_velocity + - torso_z_velocity + - torso_angular_velocity + - thigh_angular_velocity + - leg_angular_velocity + - foot_angular_velocity + task_name: hopper_rgb_state + active_action_dim: 3 +dataset: + source_hf: '' + config_name: null + source_subdir: null + converted_name: hopper_rgb_state_profile_20260911T180943Z + single_source_hf: '' + mixed_source_hf: '' + single_converted_name: hopper_rgb_state_profile_20260911T180943Z + mixed_converted_name: hopper_rgb_state_profile_20260911T180943Z + 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/hopper/20260911T180943Z-hopper/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/hopper/20260911T180943Z-hopper/vla + dataset_local_dir: /mnt/results/latency-sensitive-bench/profile-latency-new/hopper/20260911T180943Z-hopper/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/hopper/20260911T180943Z-hopper/converted/hopper_rgb_state_profile_20260911T180943Z/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 + labels: + - thigh_torque + - leg_torque + - foot_torque + low: + - -1.0 + - -1.0 + - -1.0 + type: box + base_prompt: Move the Hopper robot forward while keeping its torso upright. + Predict three continuous torques in [-1, 1] ordered as thigh, leg, and foot. + env_fps: 10.0 + env_id: LatencyBench/Hopper-v0 + frame_stack: 1 + make_kwargs: + base_env_id: Hopper-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 + obs_fps: 10.0 + registration_imports: + - latency_bench.envs.gymnasium_hopper + state_space: + labels: + - torso_height + - torso_angle + - thigh_angle + - leg_angle + - foot_angle + - torso_x_velocity + - torso_z_velocity + - torso_angular_velocity + - thigh_angular_velocity + - leg_angular_velocity + - foot_angular_velocity + task_name: hopper_rgb_state + task_name: hopper_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: hopper_rgb_state_profile_20260911T180943Z_openvla_native_continuous_projector_sft_5k +output_dir: /mnt/results/latency-sensitive-bench/profile-latency-new/hopper/20260911T180943Z-hopper/vla/hopper_rgb_state_profile_20260911T180943Z_openvla_native_continuous_projector_sft_5k +config_yaml: /mnt/results/latency-sensitive-bench/profile-latency-new/hopper/20260911T180943Z-hopper/h100/vla_train.yaml +is_debug: false +version_id: '0.21' diff --git a/latency-aware/hopper/vla/starvla-qwenoft-h1/hopper_rgb_state_profile_20260911T180943Z_openvla_native_continuous_projector_sft_5k/dataset_statistics.json b/latency-aware/hopper/vla/starvla-qwenoft-h1/hopper_rgb_state_profile_20260911T180943Z_openvla_native_continuous_projector_sft_5k/dataset_statistics.json new file mode 100644 index 0000000000000000000000000000000000000000..322dca9fdc722e8b3350e29c3c9fbc18f823a3ad --- /dev/null +++ b/latency-aware/hopper/vla/starvla-qwenoft-h1/hopper_rgb_state_profile_20260911T180943Z_openvla_native_continuous_projector_sft_5k/dataset_statistics.json @@ -0,0 +1,123 @@ +{ + "new_embodiment": { + "action": { + "mean": [ + 0.022998139262199402, + 0.4831850230693817, + 0.016577335074543953 + ], + "std": [ + 0.2603372633457184, + 0.6833056211471558, + 0.7595406770706177 + ], + "max": [ + 0.8152548670768738, + 1.0, + 1.0 + ], + "min": [ + -0.8057039976119995, + -1.0, + -1.0 + ], + "q01": [ + -0.4942622232437134, + -1.0, + -1.0 + ], + "q99": [ + 0.733130006790161, + 1.0, + 1.0 + ], + "mask": [ + true, + true, + true + ] + }, + "state": { + "mean": [ + 0.33439695835113525, + -0.47676873207092285, + 0.47354835271835327, + 0.5554091930389404, + 0.17205235362052917, + -0.0025759802665561438, + 0.0016160731902346015, + 0.0011788663687184453, + 0.015976371243596077, + 0.002401757752522826, + 0.01624964363873005 + ], + "std": [ + 0.33200088143348694, + 0.2534998655319214, + 0.27482420206069946, + 0.4218027889728546, + 0.6764597296714783, + 0.23773561418056488, + 0.5150550603866577, + 0.18555213510990143, + 0.28494536876678467, + 0.45271560549736023, + 0.5971301794052124 + ], + "max": [ + 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 + ], + "q01": [ + -0.33900941133499146, + -0.8636440396308899, + -0.1049540376663208, + -0.45442226886749265, + -0.9769227361679077, + -0.8518737506866455, + -0.863742151260376, + -0.596899824142456, + -0.6311568546295167, + -1.0, + -1.0 + ], + "q99": [ + 0.8846877050399776, + 0.3958153772354122, + 0.9532253932952879, + 0.9580308675765992, + 0.9603813648223877, + 0.42569099426269497, + 0.9137406253814697, + 0.31418379783630346, + 0.7651577854156494, + 0.9602302312850948, + 1.0 + ] + }, + "num_transitions": 88817, + "num_trajectories": 90 + } +} \ No newline at end of file diff --git a/latency-aware/hopper/vla/starvla-qwenoft-h1/hopper_rgb_state_profile_20260911T180943Z_openvla_native_continuous_projector_sft_5k/dataset_statistics_eval.json b/latency-aware/hopper/vla/starvla-qwenoft-h1/hopper_rgb_state_profile_20260911T180943Z_openvla_native_continuous_projector_sft_5k/dataset_statistics_eval.json new file mode 100644 index 0000000000000000000000000000000000000000..c2e358933e36025e3aa2d53263798a45967368e9 --- /dev/null +++ b/latency-aware/hopper/vla/starvla-qwenoft-h1/hopper_rgb_state_profile_20260911T180943Z_openvla_native_continuous_projector_sft_5k/dataset_statistics_eval.json @@ -0,0 +1,123 @@ +{ + "new_embodiment": { + "action": { + "mean": [ + 0.02262929268181324, + 0.4847935140132904, + 0.02151433937251568 + ], + "std": [ + 0.2588505148887634, + 0.681768000125885, + 0.7605242133140564 + ], + "max": [ + 0.7959198355674744, + 1.0, + 1.0 + ], + "min": [ + -0.7861610651016235, + -1.0, + -1.0 + ], + "q01": [ + -0.48549440264701843, + -1.0, + -1.0 + ], + "q99": [ + 0.7299425488710404, + 1.0, + 1.0 + ], + "mask": [ + true, + true, + true + ] + }, + "state": { + "mean": [ + 0.34278014302253723, + -0.48208826780319214, + 0.48088178038597107, + 0.5591239929199219, + 0.17630372941493988, + -0.006248814985156059, + 0.005157782696187496, + 0.0004372408729977906, + 0.01563315838575363, + 0.003359832102432847, + 0.018366994336247444 + ], + "std": [ + 0.32974857091903687, + 0.24938634037971497, + 0.2650088667869568, + 0.4167846441268921, + 0.6752994656562805, + 0.23108573257923126, + 0.5158278942108154, + 0.1842319220304489, + 0.2808758616447449, + 0.4490443170070648, + 0.5966046452522278 + ], + "max": [ + 0.9829795360565186, + 0.57453453540802, + 0.9969793558120728, + 1.0, + 1.0, + 0.47722434997558594, + 0.9945762157440186, + 0.44636070728302, + 0.9959362745285034, + 1.0, + 1.0 + ], + "min": [ + -0.39860212802886963, + -0.9755522012710571, + -0.1488437056541443, + -0.7457688450813293, + -0.9961583018302917, + -0.9987747073173523, + -0.9836085438728333, + -0.9795570373535156, + -0.8917129635810852, + -1.0, + -1.0 + ], + "q01": [ + -0.3126556849479675, + -0.863114053606987, + -0.04237606108188629, + -0.43479173064231874, + -0.9768222767114639, + -0.7950224620103831, + -0.8634002166986465, + -0.5940668570995331, + -0.5733073079586029, + -1.0, + -1.0 + ], + "q99": [ + 0.9167150700092318, + 0.3906000924110415, + 0.9528681910037995, + 0.957846553325653, + 0.9608215284347534, + 0.3674080061912542, + 0.9136585640907289, + 0.30735854387283335, + 0.752859001159669, + 0.9454998314380646, + 1.0 + ] + }, + "num_transitions": 10000, + "num_trajectories": 10 + } +} \ No newline at end of file diff --git a/latency-aware/hopper/vla/starvla-qwenoft-h1/hopper_rgb_state_profile_20260911T180943Z_openvla_native_continuous_projector_sft_5k/latency_prompt_map.json b/latency-aware/hopper/vla/starvla-qwenoft-h1/hopper_rgb_state_profile_20260911T180943Z_openvla_native_continuous_projector_sft_5k/latency_prompt_map.json new file mode 100644 index 0000000000000000000000000000000000000000..484eee358c2e672001f40e4b1813bfa052ffe51f --- /dev/null +++ b/latency-aware/hopper/vla/starvla-qwenoft-h1/hopper_rgb_state_profile_20260911T180943Z_openvla_native_continuous_projector_sft_5k/latency_prompt_map.json @@ -0,0 +1,12 @@ +{ + "1": { + "prompt": "Move the Hopper robot forward while keeping its torso upright. Predict three continuous torques in [-1, 1] ordered as thigh, leg, and 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 Hopper robot forward while keeping its torso upright. Predict three continuous torques in [-1, 1] ordered as thigh, leg, and 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/hopper/vla/starvla-qwenoft-h1/hopper_rgb_state_profile_20260911T180943Z_openvla_native_continuous_projector_sft_5k/manifest.json b/latency-aware/hopper/vla/starvla-qwenoft-h1/hopper_rgb_state_profile_20260911T180943Z_openvla_native_continuous_projector_sft_5k/manifest.json new file mode 100644 index 0000000000000000000000000000000000000000..0ea0074d0af9cdca2135004f86f0409a21b3fbda --- /dev/null +++ b/latency-aware/hopper/vla/starvla-qwenoft-h1/hopper_rgb_state_profile_20260911T180943Z_openvla_native_continuous_projector_sft_5k/manifest.json @@ -0,0 +1,143 @@ +{ + "dataset_name": "hopper_rgb_state_profile_20260911T180943Z", + "env_name": "hopper_rgb_state", + "episodes": 90, + "frames": 88817, + "task_prompts": [ + "Move the Hopper robot forward while keeping its torso upright. Predict three continuous torques in [-1, 1] ordered as thigh, leg, and 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 Hopper robot forward while keeping its torso upright. Predict three continuous torques in [-1, 1] ordered as thigh, leg, and 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/hopper/20260911T180943Z-hopper/raw", + "integration_name": "gymnasium", + "task_name": "hopper_rgb_state", + "action_layout": "gymnasium_continuous_v1", + "action_labels": [ + "thigh_torque", + "leg_torque", + "foot_torque" + ], + "carrier_action_labels": [ + "thigh_torque", + "leg_torque", + "foot_torque" + ], + "action_dim": 3, + "active_action_dim": 3, + "action_carrier": "native", + "rows_unit": "decision_step", + "obs_fps": 10.0, + "obs_stride_raw_frames": 1, + "uses_state": true, + "state_dim": 11, + "state_labels": [ + "torso_height", + "torso_angle", + "thigh_angle", + "leg_angle", + "foot_angle", + "torso_x_velocity", + "torso_z_velocity", + "torso_angular_velocity", + "thigh_angular_velocity", + "leg_angular_velocity", + "foot_angular_velocity" + ], + "state_normalization": { + "type": "min_max", + "min": [ + 0.7003283500671387, + -0.18373166024684906, + -1.5099823474884033, + -1.6614010334014893, + -0.9186854362487793, + -0.3373064696788788, + -3.1125423908233643, + -4.973757743835449, + -9.478679656982422, + -10.0, + -10.0 + ], + "max": [ + 1.7507554292678833, + 0.19961893558502197, + 0.04018906503915787, + 0.1078878715634346, + 0.914368748664856, + 5.7077836990356445, + 3.0955286026000977, + 4.943385124206543, + 9.040213584899902, + 10.0, + 10.0 + ] + }, + "robot_type": "rl_games_gymnasium", + "latency_prompt_map_path": "/mnt/results/latency-sensitive-bench/profile-latency-new/hopper/20260911T180943Z-hopper/converted/hopper_rgb_state_profile_20260911T180943Z/latency_prompt_map.json", + "custom_mixtures_path": "/mnt/results/latency-sensitive-bench/profile-latency-new/hopper/20260911T180943Z-hopper/converted/_generated_mixtures/hopper_rgb_state_profile_20260911T180943Z.json", + "gymnasium_task": { + "action_space": { + "dtype": "float32", + "high": [ + 1.0, + 1.0, + 1.0 + ], + "labels": [ + "thigh_torque", + "leg_torque", + "foot_torque" + ], + "low": [ + -1.0, + -1.0, + -1.0 + ], + "type": "box" + }, + "base_prompt": "Move the Hopper robot forward while keeping its torso upright. Predict three continuous torques in [-1, 1] ordered as thigh, leg, and foot.", + "env_fps": 10.0, + "env_id": "LatencyBench/Hopper-v0", + "frame_stack": 1, + "make_kwargs": { + "base_env_id": "Hopper-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 + ], + "obs_fps": 10.0, + "registration_imports": [ + "latency_bench.envs.gymnasium_hopper" + ], + "state_space": { + "labels": [ + "torso_height", + "torso_angle", + "thigh_angle", + "leg_angle", + "foot_angle", + "torso_x_velocity", + "torso_z_velocity", + "torso_angular_velocity", + "thigh_angular_velocity", + "leg_angular_velocity", + "foot_angular_velocity" + ] + }, + "task_name": "hopper_rgb_state" + }, + "validation_dataset_name": "hopper_rgb_state_profile_20260911T180943Z__val", + "validation_episodes": 10, + "validation_frames": 10000 +} \ No newline at end of file diff --git a/latency-aware/hopper/vla/starvla-qwenoft-h1/hopper_rgb_state_profile_20260911T180943Z_openvla_native_continuous_projector_sft_5k/profile/DONE b/latency-aware/hopper/vla/starvla-qwenoft-h1/hopper_rgb_state_profile_20260911T180943Z_openvla_native_continuous_projector_sft_5k/profile/DONE new file mode 100644 index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391 diff --git a/latency-aware/hopper/vla/starvla-qwenoft-h1/hopper_rgb_state_profile_20260911T180943Z_openvla_native_continuous_projector_sft_5k/profile/hardware.json b/latency-aware/hopper/vla/starvla-qwenoft-h1/hopper_rgb_state_profile_20260911T180943Z_openvla_native_continuous_projector_sft_5k/profile/hardware.json new file mode 100644 index 0000000000000000000000000000000000000000..e8204e81807044975c0e0a0953d16d6835192f31 --- /dev/null +++ b/latency-aware/hopper/vla/starvla-qwenoft-h1/hopper_rgb_state_profile_20260911T180943Z_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/hopper/vla/starvla-qwenoft-h1/hopper_rgb_state_profile_20260911T180943Z_openvla_native_continuous_projector_sft_5k/profile/latency_burst_model.json b/latency-aware/hopper/vla/starvla-qwenoft-h1/hopper_rgb_state_profile_20260911T180943Z_openvla_native_continuous_projector_sft_5k/profile/latency_burst_model.json new file mode 100644 index 0000000000000000000000000000000000000000..1d402e53f0f814c206fb40d3f5366493189ae014 --- /dev/null +++ b/latency-aware/hopper/vla/starvla-qwenoft-h1/hopper_rgb_state_profile_20260911T180943Z_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": [ + 92.92644906044006, + 92.92644906044006, + 92.92644906044006, + 92.92644906044006, + 92.92644906044006, + 92.92644906044006, + 92.92644906044006, + 92.92644906044006, + 92.92644906044006, + 92.92644906044006, + 92.92644906044006, + 92.92644906044006, + 92.92644906044006, + 92.92644906044006, + 92.92644906044006, + 92.92644906044006, + 92.92644906044006, + 92.92644906044006, + 92.92644906044006, + 92.92644906044006, + 92.92644906044006, + 92.92644906044006, + 92.92644906044006, + 92.92644906044006, + 92.92644906044006, + 92.92644906044006, + 92.92644906044006, + 92.92644906044006, + 92.92818662166596, + 92.93339930534363, + 92.9386119890213, + 92.94382467269898, + 92.94903735637665, + 92.95425004005432, + 92.95946272373199, + 92.96467540740967, + 92.96988809108734, + 92.97510077476501, + 92.9803134584427, + 92.98552614212036, + 92.99073882579803, + 92.9959515094757, + 93.00116419315339, + 93.00637687683106, + 93.01158956050872, + 93.01680224418641, + 93.02201492786408, + 93.02722761154175, + 93.03244029521942, + 93.0376529788971, + 93.04286566257477, + 93.04807834625244, + 93.05329102993011, + 93.05850371360779, + 93.06371639728546, + 93.06892908096313, + 93.07414176464081, + 93.07935444831848, + 93.08456713199615, + 93.08977981567382, + 93.0949924993515, + 93.10020518302917, + 93.65123818159104, + 94.20227118015289, + 94.75330417871476, + 95.30433717727661, + 95.85537017583847, + 96.40640317440034, + 96.95743617296219, + 97.50846917152404, + 98.0595021700859, + 98.61053516864776, + 99.16156816720962, + 99.71260116577149, + 100.26363416433334, + 100.8146671628952, + 101.36570016145707, + 101.91673316001892, + 102.46776615858079, + 103.01879915714264, + 103.56983215570449, + 104.12086515426635, + 104.67189815282822, + 105.22293115139007, + 105.77396414995194, + 106.32499714851379, + 106.87603014707565, + 107.42706314563752, + 107.97809614419937, + 108.52912914276124, + 109.08016214132309, + 109.63119513988495, + 110.18222813844682, + 110.73326113700867, + 111.28429413557052, + 111.46797180175781, + 111.46797180175781, + 111.46797180175781, + 111.46797180175781, + 111.46797180175781, + 111.46797180175781, + 111.46797180175781, + 111.46797180175781, + 111.46797180175781, + 111.46797180175781, + 111.46797180175781, + 111.46797180175781, + 111.46797180175781, + 111.46797180175781, + 111.46797180175781, + 111.46797180175781, + 111.46797180175781, + 111.46797180175781, + 111.46797180175781, + 111.46797180175781, + 111.46797180175781, + 111.46797180175781, + 111.46797180175781, + 111.46797180175781, + 111.46797180175781, + 111.46797180175781, + 111.46797180175781, + 111.46797180175781 + ], + "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": [ + 0.31311678886413574, + 0.31311678886413574, + 0.3165763258934021, + 0.32325554490089414, + 0.3290378570556641, + 0.34071471095085143, + 0.3509213447570801, + 0.35257425904273987, + 0.3555735468864441, + 0.3601503908634186, + 0.3644925594329834, + 0.3818242847919464, + 0.3849796652793884, + 0.43671417236328125, + 0.4749835133552551, + 0.4945000410079956, + 0.5159845948219299, + 0.5300548076629639, + 0.5394640564918518, + 0.547340989112854, + 0.5548242330551147, + 0.5603091716766357, + 0.5666623711585999, + 0.5706260204315186, + 0.5731815695762634, + 0.5773470401763916, + 0.5807499289512634, + 0.5849419832229614, + 0.5887086987495422, + 0.5918290615081787, + 0.5939978957176208, + 0.5969480276107788, + 0.5994340777397156, + 0.6025159358978271, + 0.6040467619895935, + 0.6064813137054443, + 0.6088453531265259, + 0.6113431453704834, + 0.6135084629058838, + 0.6157195568084717, + 0.6180734634399414, + 0.6201481819152832, + 0.6233775615692139, + 0.6259479522705078, + 0.6277621388435364, + 0.6300039291381836, + 0.6324813365936279, + 0.6342650651931763, + 0.6360381841659546, + 0.6377451419830322, + 0.639049768447876, + 0.640360951423645, + 0.6418464183807373, + 0.6428799629211426, + 0.6441222429275513, + 0.6458930969238281, + 0.6482654809951782, + 0.6496939659118652, + 0.6526684761047363, + 0.6541144847869873, + 0.6560317873954773, + 0.658473014831543, + 0.659576952457428, + 0.6612474918365479, + 0.6628846526145935, + 0.6641979217529297, + 0.6656185388565063, + 0.6666920185089111, + 0.6680689454078674, + 0.6693181991577148, + 0.6703876256942749, + 0.6712313890457153, + 0.6727586388587952, + 0.6742727756500244, + 0.6758086681365967, + 0.6774241924285889, + 0.6792400479316711, + 0.6818399429321289, + 0.683896005153656, + 0.6866179704666138, + 0.6883576512336731, + 0.6901149749755859, + 0.692481517791748, + 0.6947081089019775, + 0.697051465511322, + 0.698936939239502, + 0.7017368674278259, + 0.7050299644470215, + 0.70822674036026, + 0.7101070880889893, + 0.7132005095481873, + 0.7152360677719116, + 0.7168729901313782, + 0.7200279235839844, + 0.7235506176948547, + 0.7269525527954102, + 0.7312224507331848, + 0.7360110282897949, + 0.7416163682937622, + 0.7470424175262451, + 0.7538335919380188, + 0.7591180801391602, + 0.7654633522033691, + 0.7710039615631104, + 0.7774527072906494, + 0.7881929874420166, + 0.8031460642814636, + 0.8192280530929565, + 0.8465723991394043, + 0.8769099712371826, + 0.9416267275810242, + 0.9441980957984923, + 0.9461143732070924, + 0.9521929323673245, + 0.9566469669342051, + 0.9747528433799744, + 0.979430532455442, + 1.0052420735359195, + 1.0194490909576401, + 1.0595047175884362, + 1.2158348947763677, + 1.3550479412078857, + 1.3550479412078857 + ], + "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": 2072 + }, + "regime_transition_counts": { + "burst": { + "burst": 0, + "calm": 3 + }, + "calm": { + "burst": 3, + "calm": 2064 + } + }, + "reset_scope": "session", + "schema_version": 12, + "spike_median_multiplier": 1.25, + "spike_threshold_ms_by_worker_slot": { + "0": 87.82212123274803 + }, + "worker_count": 1 +} diff --git a/latency-aware/hopper/vla/starvla-qwenoft-h1/hopper_rgb_state_profile_20260911T180943Z_openvla_native_continuous_projector_sft_5k/profile/latency_distribution.json b/latency-aware/hopper/vla/starvla-qwenoft-h1/hopper_rgb_state_profile_20260911T180943Z_openvla_native_continuous_projector_sft_5k/profile/latency_distribution.json new file mode 100644 index 0000000000000000000000000000000000000000..59cde1722b922a5e61b79b6e5c535c5fe9b10c7c --- /dev/null +++ b/latency-aware/hopper/vla/starvla-qwenoft-h1/hopper_rgb_state_profile_20260911T180943Z_openvla_native_continuous_projector_sft_5k/profile/latency_distribution.json @@ -0,0 +1,1027 @@ +{ + "schema_version": 3, + "worker_slots": { + "0": { + "all": { + "count": 2075, + "observation_to_action_latency_ms": { + "distribution_type": "inverse_cdf", + "latency_ms": [ + 65.19399881362915, + 65.19399881362915, + 65.21452818512917, + 65.37617138624191, + 65.54988111257553, + 65.60978311300278, + 65.85620288848877, + 66.11677795648575, + 66.1667419910431, + 66.25579528808593, + 66.3200498342514, + 66.37821425199509, + 66.44661968946457, + 67.40388679504395, + 67.9565160870552, + 68.30614387989044, + 68.86876726150513, + 69.23061203956604, + 69.40943151712418, + 69.708052277565, + 69.82890748977661, + 69.99191188812256, + 70.07996428012848, + 70.20106554031372, + 70.2888200879097, + 70.34249114990234, + 70.42751479148865, + 70.50662112236023, + 70.5581624507904, + 70.60912585258484, + 70.65479511022568, + 70.69437503814697, + 70.74371039867401, + 70.80917811393738, + 70.83664780855179, + 70.88321900367737, + 70.91019904613495, + 70.95217895507812, + 70.97798812389374, + 71.01112449169159, + 71.03684651851654, + 71.07607984542847, + 71.12114173173904, + 71.14498901367188, + 71.169668674469, + 71.19082713127136, + 71.21074295043945, + 71.24556457996368, + 71.27112966775894, + 71.28998637199402, + 71.31158643960953, + 71.34597957134247, + 71.36409616470337, + 71.39237093925476, + 71.41189247369766, + 71.44127345085144, + 71.47214812040329, + 71.49239921569824, + 71.51301550865173, + 71.53032338619232, + 71.5557844042778, + 71.5749728679657, + 71.59549480676651, + 71.6154580116272, + 71.63657933473587, + 71.6589138507843, + 71.68088454008102, + 71.70165157318115, + 71.73091351985931, + 71.75631403923035, + 71.77797484397888, + 71.82077741622925, + 71.85434252023697, + 71.88232922554016, + 71.9016506075859, + 71.91977643966675, + 71.95200443267822, + 71.98209404945374, + 72.0122202038765, + 72.03439712524414, + 72.06454348564148, + 72.09861707687378, + 72.12148833274841, + 72.1564769744873, + 72.1896619796753, + 72.22209119796753, + 72.27087277173996, + 72.31662225723267, + 72.35859662294388, + 72.4115800857544, + 72.43648117780685, + 72.47373795509338, + 72.49812859296799, + 72.53804278373718, + 72.58772403001785, + 72.64632666110992, + 72.70239788293839, + 72.79391694068909, + 72.83307409286499, + 72.89201402664185, + 72.95603406429291, + 73.05068588256836, + 73.21339404582977, + 73.34779798984528, + 73.54553669691086, + 73.78722310066223, + 74.11802303791046, + 74.58567237854004, + 75.28934186697006, + 76.44343304634094, + 77.00749260187149, + 77.320473575592, + 77.51085505485536, + 78.1938346505165, + 78.28879662752153, + 78.42176166176796, + 79.36353144645672, + 81.95102273225795, + 83.81126960515967, + 93.65813264846805, + 102.26191507876086, + 112.13530778884888, + 112.13530778884888 + ], + "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": 0.9980124420738008, + "worker_service_time_ms": { + "distribution_type": "inverse_cdf", + "latency_ms": [ + 64.81042885780334, + 64.81042885780334, + 64.8304044932127, + 64.86529658436775, + 65.09479154348374, + 65.18544340133667, + 65.45056700706482, + 65.71252375841141, + 65.7791820526123, + 65.83106318712234, + 65.96416923999786, + 65.99974170923232, + 66.05729752779007, + 66.85541200637817, + 67.4124995470047, + 67.70765841007233, + 68.38320344686508, + 68.66614484786987, + 68.86086118221283, + 69.11453056335449, + 69.2644213438034, + 69.38441181182861, + 69.49850225448608, + 69.59222090244293, + 69.66973400115967, + 69.74627423286438, + 69.80929481983185, + 69.88673007488251, + 69.92713761329651, + 69.97099304199219, + 70.01391452550888, + 70.07167446613312, + 70.1147900223732, + 70.16800880432129, + 70.20960211753845, + 70.24057853221893, + 70.27172857522964, + 70.30395483970642, + 70.33524876832962, + 70.36733794212341, + 70.41446840763092, + 70.43638014793396, + 70.47007262706757, + 70.49437403678894, + 70.52942305803299, + 70.54843187332153, + 70.56563359498978, + 70.58997797966003, + 70.61523360013962, + 70.63820695877075, + 70.65534472465515, + 70.68713617324829, + 70.72156208753586, + 70.74655199050903, + 70.77104866504669, + 70.794921875, + 70.81391477584839, + 70.83501505851746, + 70.85863208770752, + 70.87805342674255, + 70.89673417806625, + 70.91481804847717, + 70.93379074335098, + 70.95436489582062, + 70.9727611541748, + 70.99777722358704, + 71.01445686817169, + 71.04416513442993, + 71.07488167285919, + 71.09937310218811, + 71.1260045170784, + 71.14588451385498, + 71.17757761478424, + 71.20317506790161, + 71.22922629117966, + 71.26177847385406, + 71.28523004055023, + 71.31312823295593, + 71.33135783672333, + 71.35811042785645, + 71.38167172670364, + 71.40552186965942, + 71.44026362895966, + 71.4722410440445, + 71.50916689634323, + 71.55513215065002, + 71.59390467405319, + 71.63793969154358, + 71.6769991517067, + 71.70788717269897, + 71.74298250675201, + 71.7811050415039, + 71.81722241640091, + 71.84697794914246, + 71.89713567495346, + 71.96147060394287, + 72.00608330965042, + 72.07237100601196, + 72.13394355773926, + 72.19530701637268, + 72.25350522994995, + 72.34730696678162, + 72.50979733467102, + 72.63740396499634, + 72.84575146436691, + 73.07518124580383, + 73.36473077535629, + 73.83938801288605, + 74.32575643062592, + 75.57429695129395, + 76.11366033554077, + 76.45826106667515, + 76.83195266723635, + 77.16471412181853, + 77.66161592006684, + 77.73926329612732, + 78.54240913391091, + 81.13938184976583, + 83.1561977505683, + 93.00029541254047, + 101.59529724419284, + 111.46797180175781, + 111.46797180175781 + ], + "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": 2072, + "observation_to_action_latency_ms": { + "distribution_type": "inverse_cdf", + "latency_ms": [ + 65.19399881362915, + 65.19399881362915, + 65.21447089385987, + 65.37542019462586, + 65.54987103176117, + 65.60966022491455, + 65.85360446739197, + 66.113779296875, + 66.16652703666686, + 66.25572955894471, + 66.31964143466949, + 66.37805958938598, + 66.44557957172394, + 67.40347755432128, + 67.95384462833404, + 68.3040751028061, + 68.86173973083496, + 69.22473413944245, + 69.40377369403839, + 69.70454369068146, + 69.82808116436004, + 69.99165661334992, + 70.07831597328186, + 70.19712745666504, + 70.2876419210434, + 70.3415141248703, + 70.42468881607056, + 70.50353671073914, + 70.55655439853668, + 70.60617962837219, + 70.65110831737519, + 70.69143896102905, + 70.74109094619752, + 70.8089477443695, + 70.83344268321991, + 70.88257723331452, + 70.90947210788727, + 70.94953246593475, + 70.9757197189331, + 71.01045125961303, + 71.03570990562439, + 71.07419350147248, + 71.11981529712676, + 71.14466162681579, + 71.16853910923004, + 71.19018000125885, + 71.20882227420807, + 71.24291341304779, + 71.26965581893921, + 71.28910193443298, + 71.30903896331787, + 71.34541156291962, + 71.3635185432434, + 71.3916244506836, + 71.41094114780427, + 71.43986271858215, + 71.47048921585083, + 71.49185324668885, + 71.51238204956054, + 71.52781995773316, + 71.55354582309722, + 71.57390081882477, + 71.59174089431762, + 71.61493656635284, + 71.6342034482956, + 71.65483741283417, + 71.68058023452758, + 71.70002612113953, + 71.72443329334259, + 71.75478495121003, + 71.77438542366028, + 71.81701493263245, + 71.84875475883484, + 71.87945019245147, + 71.90045991420746, + 71.91874521255494, + 71.94669437408447, + 71.98143951892852, + 72.01119516849518, + 72.0328631067276, + 72.06257130146027, + 72.09504079818726, + 72.11566365242004, + 72.15310551643371, + 72.18850584506988, + 72.21814324855805, + 72.26514148712158, + 72.31500059127808, + 72.35579874992371, + 72.40864204883576, + 72.43190546512604, + 72.46693255901337, + 72.49548690319061, + 72.53481489181519, + 72.58008307933807, + 72.64314253807068, + 72.69793500900269, + 72.78377663612366, + 72.82674848079681, + 72.88684828281403, + 72.93725447654724, + 73.03033211231232, + 73.20079249858856, + 73.33228698730468, + 73.50521060466767, + 73.73466797351837, + 74.04130051136016, + 74.44832262992858, + 75.16131977558135, + 76.2820675754547, + 76.88091934204103, + 76.94890914344788, + 77.14981128501891, + 77.46755316638948, + 78.03063362503056, + 78.24113952636718, + 78.38724931049347, + 78.70385138702399, + 80.69423282146485, + 83.1145306940078, + 83.92277062034601, + 84.44702100753784, + 84.44702100753784 + ], + "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": 0.9980037963516655, + "worker_service_time_ms": { + "distribution_type": "inverse_cdf", + "latency_ms": [ + 64.81042885780334, + 64.81042885780334, + 64.83034874725342, + 64.8652042169571, + 65.09434249591827, + 65.18473916053772, + 65.44755630970002, + 65.70904842376709, + 65.7790016708374, + 65.82948631858825, + 65.96403185939789, + 65.99917395210267, + 66.05419579982758, + 66.85226134777069, + 67.41213180541992, + 67.70523978710175, + 68.37709033489227, + 68.6659171819687, + 68.85855054855347, + 69.11082427978516, + 69.26426580905914, + 69.38339400291443, + 69.49818974494934, + 69.59014869213104, + 69.66760065555573, + 69.74625440597534, + 69.80519919395446, + 69.88650359630584, + 69.92661521434783, + 69.96951825618744, + 70.01145081996918, + 70.07099788188934, + 70.11137570858001, + 70.16706781387329, + 70.20938957214355, + 70.24025132656098, + 70.26918590068817, + 70.30251583099366, + 70.33475484848023, + 70.36707436561585, + 70.4130216884613, + 70.43617479801178, + 70.4693127822876, + 70.49378530979156, + 70.52594029903412, + 70.54812473297119, + 70.5624710559845, + 70.58988870143891, + 70.61189743995666, + 70.63650825500488, + 70.65514147758483, + 70.68650047779083, + 70.72054006099701, + 70.74340092658997, + 70.77034149169921, + 70.79453890323639, + 70.81376779079437, + 70.83277872562408, + 70.85774645805358, + 70.87755789279937, + 70.8952799797058, + 70.91331553459167, + 70.93301747322083, + 70.95305408000947, + 70.97179078102111, + 70.99544564723969, + 71.01263489723206, + 71.04232339382172, + 71.07317966461181, + 71.09767675876617, + 71.12224251270294, + 71.14307947158814, + 71.1765646982193, + 71.20110731601716, + 71.22757455825806, + 71.26078465461731, + 71.28300664424896, + 71.31065947055816, + 71.32941711902619, + 71.35752510547638, + 71.3804721879959, + 71.40329875946045, + 71.4374256324768, + 71.47053267478942, + 71.50463972568512, + 71.55248319149017, + 71.58942151069641, + 71.63497022628785, + 71.67235461711884, + 71.70529507637023, + 71.73997480869293, + 71.77655177116394, + 71.81513391971588, + 71.84466456413269, + 71.89355994224549, + 71.95444693088531, + 72.0025963306427, + 72.05938720703125, + 72.1252225112915, + 72.19015914440155, + 72.24081826210022, + 72.335604429245, + 72.48756666660309, + 72.61886465549469, + 72.82262350082398, + 73.05337915420532, + 73.28593082427979, + 73.7462468290329, + 74.27689339637756, + 75.49606513023376, + 75.9689773273468, + 76.06074317169188, + 76.27554173469544, + 76.56049769020082, + 77.12394835662843, + 77.4304215621948, + 77.7176802444458, + 77.9243575134278, + 80.04920254230532, + 82.06012575626357, + 83.2668972358703, + 83.78737902641296, + 83.78737902641296 + ], + "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/hopper/vla/starvla-qwenoft-h1/hopper_rgb_state_profile_20260911T180943Z_openvla_native_continuous_projector_sft_5k/profile/latency_profile.png b/latency-aware/hopper/vla/starvla-qwenoft-h1/hopper_rgb_state_profile_20260911T180943Z_openvla_native_continuous_projector_sft_5k/profile/latency_profile.png new file mode 100644 index 0000000000000000000000000000000000000000..9e1413a55e254fa732a9544836153c7f3f798a8b Binary files /dev/null and b/latency-aware/hopper/vla/starvla-qwenoft-h1/hopper_rgb_state_profile_20260911T180943Z_openvla_native_continuous_projector_sft_5k/profile/latency_profile.png differ diff --git a/latency-aware/hopper/vla/starvla-qwenoft-h1/hopper_rgb_state_profile_20260911T180943Z_openvla_native_continuous_projector_sft_5k/profile/profile.json b/latency-aware/hopper/vla/starvla-qwenoft-h1/hopper_rgb_state_profile_20260911T180943Z_openvla_native_continuous_projector_sft_5k/profile/profile.json new file mode 100644 index 0000000000000000000000000000000000000000..f326bbb79f0cad25de15e0e64806a9ac00a43afc --- /dev/null +++ b/latency-aware/hopper/vla/starvla-qwenoft-h1/hopper_rgb_state_profile_20260911T180943Z_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": 2075, + "n_capacity_drops": 1, + "n_observation_attempts": 2076, + "per_slot_summary": { + "0": { + "admitted_count": 2075, + "mean_observation_to_action_latency_ms": 71.62769928219807, + "mean_worker_service_time_ms": 70.97048402751784, + "p95_observation_to_action_latency_ms": 74.10973887443542, + "p95_worker_service_time_ms": 73.34952919483185, + "p99_worker_service_time_ms": 76.0833806705475 + } + }, + "provenance": { + "base_config": "/workspace/tasks/20260911T023128Z-p-only4/hopper/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/hopper_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": "20260911T034939618615Z", + "summary": { + "frame_ms": 100.0, + "max_ms": 112.13530778884888, + "mean_effective_frames": 0.7162769928219807, + "mean_ms": 71.62769928219807, + "min_ms": 65.19399881362915, + "n_samples": 2075, + "p50_frames": 0.715749728679657, + "p50_ms": 71.5749728679657, + "p90_frames": 0.7304598674774171, + "p90_ms": 73.04598674774171, + "p95_frames": 0.7410973887443543, + "p95_ms": 74.10973887443542, + "p99_frames": 0.7696984185695647, + "p99_ms": 76.96984185695646, + "prob_latency_gt_1_frame": 0.00048192771084337347, + "prob_latency_gt_2_frames": 0.0, + "prob_latency_gt_3_frames": 0.0, + "std_ms": 2.032231876669252 + }, + "visualization_path": "latency_profile.png", + "workload_id": "hopper" +} diff --git a/latency-aware/hopper/vla/starvla-qwenoft-h1/hopper_rgb_state_profile_20260911T180943Z_openvla_native_continuous_projector_sft_5k/provenance.json b/latency-aware/hopper/vla/starvla-qwenoft-h1/hopper_rgb_state_profile_20260911T180943Z_openvla_native_continuous_projector_sft_5k/provenance.json new file mode 100644 index 0000000000000000000000000000000000000000..669a3410d2713b937cdf00bc2ea2b31c497caa3f --- /dev/null +++ b/latency-aware/hopper/vla/starvla-qwenoft-h1/hopper_rgb_state_profile_20260911T180943Z_openvla_native_continuous_projector_sft_5k/provenance.json @@ -0,0 +1,34 @@ +{ + "task": "hopper", + "model": "qwenoft", + "training_condition": "latency-aware", + "training_run_id": "hopper_rgb_state_profile_20260911T180943Z_openvla_native_continuous_projector_sft_5k", + "source": { + "repo_id": "latency-sensitive-bench/Standard-Pipeline", + "revision": "2208875f92b21e3f8ec2363a30a6705dfe56cc04", + "path": "hopper/profile_latency/OpenVLA", + "url": "https://huggingface.co/latency-sensitive-bench/Standard-Pipeline/tree/2208875f92b21e3f8ec2363a30a6705dfe56cc04/hopper/profile_latency/OpenVLA" + }, + "hf_repo_id": "latency-sensitive-bench/benchmark-models", + "hf_path": "latency-aware/hopper/vla/starvla-qwenoft-h1/hopper_rgb_state_profile_20260911T180943Z_openvla_native_continuous_projector_sft_5k", + "checkpoint": { + "source_file": "hopper/profile_latency/OpenVLA/checkpoints/steps_5000_pytorch_model.pt", + "source_sha256": "2c9456b87aaba7d89cd53481047badde32513375f31056c4464a72642cacbe42", + "source_bytes": 9785081249, + "file": "checkpoints/model.pt", + "selection_rule": "only published VLA checkpoint", + "method": "copy", + "sha256": "2c9456b87aaba7d89cd53481047badde32513375f31056c4464a72642cacbe42", + "bytes": 9785081249 + }, + "config_source": "hopper/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/hopper/vla/starvla-qwenoft-h1/hopper_rgb_state_profile_20260911T180943Z_openvla_native_continuous_projector_sft_5k/source/config.yaml b/latency-aware/hopper/vla/starvla-qwenoft-h1/hopper_rgb_state_profile_20260911T180943Z_openvla_native_continuous_projector_sft_5k/source/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..873a6ae04bdccc381e697eb769c5b5bbc329c1e3 --- /dev/null +++ b/latency-aware/hopper/vla/starvla-qwenoft-h1/hopper_rgb_state_profile_20260911T180943Z_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: hopper_rgb_state_profile_20260911T180943Z + dataset_py: lerobot_datasets + eval_data_mix: hopper_rgb_state_profile_20260911T180943Z__val + latency_curriculum: + enabled: false + per_device_batch_size: 16 +framework: + action_model: + action_dim: 3 + action_env_dim: 3 + action_hidden_dim: 2560 + action_horizon: 1 + action_model_type: MLP + loss_type: l1 + state_dim: 11 + 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/hopper/20260911T180943Z-hopper/vla/hopper_rgb_state_profile_20260911T180943Z_openvla_native_continuous_projector_sft_5k +rl_games: + env_eval: + enabled: false + task: gymnasium +run_id: hopper_rgb_state_profile_20260911T180943Z_openvla_native_continuous_projector_sft_5k +run_root_dir: /mnt/results/latency-sensitive-bench/profile-latency-new/hopper/20260911T180943Z-hopper/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/hopper/vla/starvla-qwenoft-h1/hopper_rgb_state_profile_20260911T180943Z_openvla_native_continuous_projector_sft_5k/source/provenance.json b/latency-aware/hopper/vla/starvla-qwenoft-h1/hopper_rgb_state_profile_20260911T180943Z_openvla_native_continuous_projector_sft_5k/source/provenance.json new file mode 100644 index 0000000000000000000000000000000000000000..53619a015bb31e1194c3c7d3b83abd60c8dc287e --- /dev/null +++ b/latency-aware/hopper/vla/starvla-qwenoft-h1/hopper_rgb_state_profile_20260911T180943Z_openvla_native_continuous_projector_sft_5k/source/provenance.json @@ -0,0 +1,820 @@ +{ + "run_id": "20260911T180943Z-hopper", + "final_step": 5000, + "exit_code": 0, + "checkpoint_name": "steps_5000_pytorch_model.pt", + "checkpoint_size": 9785081249, + "checkpoint_sha256": "2c9456b87aaba7d89cd53481047badde32513375f31056c4464a72642cacbe42", + "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": 12887.055035223, + "_step": 5000, + "_timestamp": 1789168886.562035, + "_wandb.runtime": 12887, + "batch/effective_tokens": 2640, + "batch/image_count": 16, + "batch/input_len_max": 165, + "batch/input_len_mean": 165, + "batch/padding_ratio": 0, + "batch/pixel_values_rows": 4096, + "batch/size": 16, + "epoch": 0.9, + "eval/action_loss/samples": 3200, + "eval/action_loss/seconds": 29.00665699999081, + "eval/hopper_rgb_state/latency_1/loss": 0.015956107527017593, + "eval/hopper_rgb_state/loss": 0.015956107527017593, + "eval/latency_1/loss": 0.015956107527017593, + "eval/loss": 0.015956107527017593, + "learning_rate/action_model": 5e-06, + "learning_rate/qwen_vl_interface": 5e-07, + "throughput/effective_tokens_per_sec": 1064.2968362444833, + "throughput/samples_per_sec": 6.450283856027171, + "timing/action_head_loss": 0.0016742919979151338, + "timing/backward": 0.7752675079973415, + "timing/checkpoint_total": 13.517078855002184, + "timing/data": 0.00044024801172781736, + "timing/dataloader_next": 0.0006373850046657026, + "timing/eval_action_loss_total": 29.009374604996992, + "timing/forward": 0.15355812601046637, + "timing/log_metrics_total": 0.005541367994737811, + "timing/lr_scheduler": 8.582200098317116e-05, + "timing/model": 2.480336508000619, + "timing/optimizer_step": 1.5480549599888036, + "timing/qwen_h2d": 0.0031774929957464337, + "timing/qwen_input_build_total": 0.020895038003800437, + "timing/qwen_processor": 0.01715597600559704, + "timing/train_step_total": 2.4805109910084866, + "timing/vlm_forward": 0.13031690299976617, + "train/grad_norm_pre_clip": 15.139167785644531, + "train/loss": 0.014269337058067322 + }, + "url": "https://wandb.ai/dongqianyu99-zhejiang-university/latency-sensitive-bench/runs/biefsdo4", + "exit_code": 0, + "completed_utc": "2026-09-11T23:21:31Z" + }, + "dataset_upload": [ + { + "repo_id": "latency-sensitive-bench/extra-envs-rollouts", + "repo_type": "dataset", + "prefix": "profile-latency-new/hopper/20260911T180943Z-hopper/raw", + "commit": "f2cd7441fea202de16d341a6c30d7610f4ad0838", + "url": "https://huggingface.co/datasets/latency-sensitive-bench/extra-envs-rollouts/commit/f2cd7441fea202de16d341a6c30d7610f4ad0838", + "verified_files": 6, + "bytes": 2542898636, + "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/hopper/20260911T180943Z-hopper/converted", + "commit": "6f1ec3a2d26adf88ec4281f6bdbe7be48303fe64", + "url": "https://huggingface.co/datasets/latency-sensitive-bench/extra-envs-rollouts/commit/6f1ec3a2d26adf88ec4281f6bdbe7be48303fe64", + "verified_files": 117, + "bytes": 2541581158, + "verification": "Every file size; Git blob SHA1 for regular files and SHA256 for LFS files" + } + ], + "data_validation": { + "episodes": [ + { + "split": "train", + "episode_idx": 0, + "seed": 3333, + "actual_rows": 1000, + "actual_return": 3666.6915742754936 + }, + { + "split": "train", + "episode_idx": 2, + "seed": 3335, + "actual_rows": 1000, + "actual_return": 3651.1633432507515 + }, + { + "split": "train", + "episode_idx": 4, + "seed": 3337, + "actual_rows": 1000, + "actual_return": 3659.8270612955093 + }, + { + "split": "train", + "episode_idx": 5, + "seed": 3338, + "actual_rows": 1000, + "actual_return": 3663.8320927023888 + }, + { + "split": "train", + "episode_idx": 6, + "seed": 3339, + "actual_rows": 1000, + "actual_return": 3648.6879789233208 + }, + { + "split": "train", + "episode_idx": 8, + "seed": 3343, + "actual_rows": 1000, + "actual_return": 3753.7137526273727 + }, + { + "split": "train", + "episode_idx": 9, + "seed": 3344, + "actual_rows": 1000, + "actual_return": 3668.0288415551186 + }, + { + "split": "train", + "episode_idx": 10, + "seed": 3345, + "actual_rows": 1000, + "actual_return": 3659.7238799333572 + }, + { + "split": "train", + "episode_idx": 11, + "seed": 3346, + "actual_rows": 1000, + "actual_return": 3661.552826344967 + }, + { + "split": "train", + "episode_idx": 12, + "seed": 3349, + "actual_rows": 877, + "actual_return": 3353.1152487397194 + }, + { + "split": "train", + "episode_idx": 13, + "seed": 3352, + "actual_rows": 920, + "actual_return": 3483.5199750065804 + }, + { + "split": "train", + "episode_idx": 14, + "seed": 3347, + "actual_rows": 1000, + "actual_return": 3769.1338381767273 + }, + { + "split": "train", + "episode_idx": 15, + "seed": 3348, + "actual_rows": 1000, + "actual_return": 3664.1757635474205 + }, + { + "split": "train", + "episode_idx": 16, + "seed": 3350, + "actual_rows": 1000, + "actual_return": 3635.382632493973 + }, + { + "split": "train", + "episode_idx": 18, + "seed": 3354, + "actual_rows": 1000, + "actual_return": 3668.5161008238792 + }, + { + "split": "train", + "episode_idx": 19, + "seed": 3357, + "actual_rows": 1000, + "actual_return": 3667.794293284416 + }, + { + "split": "train", + "episode_idx": 20, + "seed": 3359, + "actual_rows": 1000, + "actual_return": 3644.1916761398315 + }, + { + "split": "train", + "episode_idx": 21, + "seed": 3361, + "actual_rows": 1000, + "actual_return": 3630.3504772782326 + }, + { + "split": "train", + "episode_idx": 22, + "seed": 3362, + "actual_rows": 1000, + "actual_return": 3666.6928303837776 + }, + { + "split": "train", + "episode_idx": 23, + "seed": 3365, + "actual_rows": 1000, + "actual_return": 3723.548827588558 + }, + { + "split": "train", + "episode_idx": 24, + "seed": 3367, + "actual_rows": 1000, + "actual_return": 3646.3406707644463 + }, + { + "split": "train", + "episode_idx": 26, + "seed": 3369, + "actual_rows": 1000, + "actual_return": 3680.6361914277077 + }, + { + "split": "train", + "episode_idx": 27, + "seed": 3370, + "actual_rows": 1000, + "actual_return": 3655.40633392334 + }, + { + "split": "train", + "episode_idx": 28, + "seed": 3374, + "actual_rows": 817, + "actual_return": 3076.763543844223 + }, + { + "split": "train", + "episode_idx": 30, + "seed": 3373, + "actual_rows": 1000, + "actual_return": 3680.408108472824 + }, + { + "split": "train", + "episode_idx": 31, + "seed": 3375, + "actual_rows": 1000, + "actual_return": 3719.736001729965 + }, + { + "split": "train", + "episode_idx": 32, + "seed": 3376, + "actual_rows": 1000, + "actual_return": 3625.328989505768 + }, + { + "split": "train", + "episode_idx": 33, + "seed": 3377, + "actual_rows": 1000, + "actual_return": 3658.801306784153 + }, + { + "split": "train", + "episode_idx": 34, + "seed": 3378, + "actual_rows": 1000, + "actual_return": 3648.415136575699 + }, + { + "split": "train", + "episode_idx": 35, + "seed": 3379, + "actual_rows": 1000, + "actual_return": 3654.3165352344513 + }, + { + "split": "train", + "episode_idx": 36, + "seed": 3380, + "actual_rows": 1000, + "actual_return": 3653.875044286251 + }, + { + "split": "train", + "episode_idx": 37, + "seed": 3381, + "actual_rows": 1000, + "actual_return": 3655.5150515437126 + }, + { + "split": "train", + "episode_idx": 38, + "seed": 3382, + "actual_rows": 1000, + "actual_return": 3673.6515297293663 + }, + { + "split": "train", + "episode_idx": 39, + "seed": 3383, + "actual_rows": 1000, + "actual_return": 3658.504702627659 + }, + { + "split": "train", + "episode_idx": 40, + "seed": 3384, + "actual_rows": 1000, + "actual_return": 3663.9864615797997 + }, + { + "split": "train", + "episode_idx": 41, + "seed": 3385, + "actual_rows": 1000, + "actual_return": 3661.3572192788124 + }, + { + "split": "train", + "episode_idx": 42, + "seed": 3387, + "actual_rows": 1000, + "actual_return": 3633.1511325240135 + }, + { + "split": "train", + "episode_idx": 43, + "seed": 3388, + "actual_rows": 1000, + "actual_return": 3664.4560843110085 + }, + { + "split": "train", + "episode_idx": 44, + "seed": 3391, + "actual_rows": 1000, + "actual_return": 3648.8136288523674 + }, + { + "split": "train", + "episode_idx": 45, + "seed": 3393, + "actual_rows": 901, + "actual_return": 3388.605288386345 + }, + { + "split": "train", + "episode_idx": 46, + "seed": 3395, + "actual_rows": 926, + "actual_return": 3507.439621448517 + }, + { + "split": "train", + "episode_idx": 48, + "seed": 3397, + "actual_rows": 1000, + "actual_return": 3641.3666764497757 + }, + { + "split": "train", + "episode_idx": 49, + "seed": 3399, + "actual_rows": 1000, + "actual_return": 3681.385838329792 + }, + { + "split": "train", + "episode_idx": 50, + "seed": 3401, + "actual_rows": 1000, + "actual_return": 3627.1116198301315 + }, + { + "split": "train", + "episode_idx": 51, + "seed": 3403, + "actual_rows": 1000, + "actual_return": 3755.0067950487137 + }, + { + "split": "train", + "episode_idx": 52, + "seed": 3405, + "actual_rows": 1000, + "actual_return": 3654.9195495843887 + }, + { + "split": "train", + "episode_idx": 53, + "seed": 3408, + "actual_rows": 822, + "actual_return": 3103.7388365268707 + }, + { + "split": "train", + "episode_idx": 54, + "seed": 3406, + "actual_rows": 1000, + "actual_return": 3645.0947138667107 + }, + { + "split": "train", + "episode_idx": 55, + "seed": 3407, + "actual_rows": 1000, + "actual_return": 3655.4239656329155 + }, + { + "split": "train", + "episode_idx": 56, + "seed": 3410, + "actual_rows": 1000, + "actual_return": 3648.959351658821 + }, + { + "split": "train", + "episode_idx": 57, + "seed": 3411, + "actual_rows": 1000, + "actual_return": 3674.377905368805 + }, + { + "split": "train", + "episode_idx": 59, + "seed": 3413, + "actual_rows": 1000, + "actual_return": 3650.8543192744255 + }, + { + "split": "train", + "episode_idx": 60, + "seed": 3415, + "actual_rows": 928, + "actual_return": 3497.252098798752 + }, + { + "split": "train", + "episode_idx": 61, + "seed": 3416, + "actual_rows": 1000, + "actual_return": 3652.530795931816 + }, + { + "split": "train", + "episode_idx": 62, + "seed": 3417, + "actual_rows": 1000, + "actual_return": 3637.32221609354 + }, + { + "split": "train", + "episode_idx": 63, + "seed": 3419, + "actual_rows": 1000, + "actual_return": 3779.3595331907272 + }, + { + "split": "train", + "episode_idx": 64, + "seed": 3420, + "actual_rows": 1000, + "actual_return": 3657.5068569779396 + }, + { + "split": "train", + "episode_idx": 65, + "seed": 3421, + "actual_rows": 1000, + "actual_return": 3777.8512984514236 + }, + { + "split": "train", + "episode_idx": 66, + "seed": 3422, + "actual_rows": 1000, + "actual_return": 3668.410031735897 + }, + { + "split": "train", + "episode_idx": 67, + "seed": 3423, + "actual_rows": 1000, + "actual_return": 3663.460659146309 + }, + { + "split": "train", + "episode_idx": 68, + "seed": 3424, + "actual_rows": 1000, + "actual_return": 3635.0319840312004 + }, + { + "split": "train", + "episode_idx": 69, + "seed": 3425, + "actual_rows": 998, + "actual_return": 3758.966288924217 + }, + { + "split": "train", + "episode_idx": 70, + "seed": 3427, + "actual_rows": 1000, + "actual_return": 3643.151776909828 + }, + { + "split": "train", + "episode_idx": 71, + "seed": 3428, + "actual_rows": 1000, + "actual_return": 3691.0575286746025 + }, + { + "split": "train", + "episode_idx": 72, + "seed": 3429, + "actual_rows": 1000, + "actual_return": 3646.7782967686653 + }, + { + "split": "train", + "episode_idx": 73, + "seed": 3430, + "actual_rows": 1000, + "actual_return": 3650.3808391690254 + }, + { + "split": "train", + "episode_idx": 74, + "seed": 3431, + "actual_rows": 1000, + "actual_return": 3681.0871458649635 + }, + { + "split": "train", + "episode_idx": 75, + "seed": 3432, + "actual_rows": 1000, + "actual_return": 3716.7892948389053 + }, + { + "split": "train", + "episode_idx": 76, + "seed": 3433, + "actual_rows": 1000, + "actual_return": 3680.078442156315 + }, + { + "split": "train", + "episode_idx": 78, + "seed": 3436, + "actual_rows": 1000, + "actual_return": 3682.9771634936333 + }, + { + "split": "train", + "episode_idx": 79, + "seed": 3437, + "actual_rows": 1000, + "actual_return": 3649.6008907556534 + }, + { + "split": "train", + "episode_idx": 80, + "seed": 3439, + "actual_rows": 1000, + "actual_return": 3662.393703520298 + }, + { + "split": "train", + "episode_idx": 82, + "seed": 3441, + "actual_rows": 1000, + "actual_return": 3652.5504442453384 + }, + { + "split": "train", + "episode_idx": 83, + "seed": 3443, + "actual_rows": 1000, + "actual_return": 3632.5786336660385 + }, + { + "split": "train", + "episode_idx": 84, + "seed": 3444, + "actual_rows": 1000, + "actual_return": 3654.980935573578 + }, + { + "split": "train", + "episode_idx": 85, + "seed": 3445, + "actual_rows": 1000, + "actual_return": 3660.7770435214043 + }, + { + "split": "train", + "episode_idx": 86, + "seed": 3446, + "actual_rows": 1000, + "actual_return": 3666.435356259346 + }, + { + "split": "train", + "episode_idx": 87, + "seed": 3447, + "actual_rows": 1000, + "actual_return": 3651.312874853611 + }, + { + "split": "train", + "episode_idx": 88, + "seed": 3448, + "actual_rows": 1000, + "actual_return": 3662.263320147991 + }, + { + "split": "train", + "episode_idx": 89, + "seed": 3449, + "actual_rows": 1000, + "actual_return": 3726.8336040973663 + }, + { + "split": "train", + "episode_idx": 90, + "seed": 3450, + "actual_rows": 1000, + "actual_return": 3651.4421622157097 + }, + { + "split": "train", + "episode_idx": 91, + "seed": 3451, + "actual_rows": 1000, + "actual_return": 3676.0585535764694 + }, + { + "split": "train", + "episode_idx": 92, + "seed": 3453, + "actual_rows": 817, + "actual_return": 3079.6246042251587 + }, + { + "split": "train", + "episode_idx": 93, + "seed": 3454, + "actual_rows": 1000, + "actual_return": 3617.6154170036316 + }, + { + "split": "train", + "episode_idx": 94, + "seed": 3461, + "actual_rows": 811, + "actual_return": 3057.7498267292976 + }, + { + "split": "train", + "episode_idx": 95, + "seed": 3455, + "actual_rows": 1000, + "actual_return": 3652.219022333622 + }, + { + "split": "train", + "episode_idx": 96, + "seed": 3456, + "actual_rows": 1000, + "actual_return": 3639.2247934937477 + }, + { + "split": "train", + "episode_idx": 97, + "seed": 3457, + "actual_rows": 1000, + "actual_return": 3675.2209025621414 + }, + { + "split": "train", + "episode_idx": 98, + "seed": 3459, + "actual_rows": 1000, + "actual_return": 3676.3608739972115 + }, + { + "split": "train", + "episode_idx": 99, + "seed": 3460, + "actual_rows": 1000, + "actual_return": 3655.0186158418655 + }, + { + "split": "val", + "episode_idx": 1, + "seed": 3334, + "actual_rows": 1000, + "actual_return": 3646.8913483023643 + }, + { + "split": "val", + "episode_idx": 3, + "seed": 3336, + "actual_rows": 1000, + "actual_return": 3678.674166083336 + }, + { + "split": "val", + "episode_idx": 7, + "seed": 3340, + "actual_rows": 1000, + "actual_return": 3680.9379561543465 + }, + { + "split": "val", + "episode_idx": 17, + "seed": 3351, + "actual_rows": 1000, + "actual_return": 3649.780366063118 + }, + { + "split": "val", + "episode_idx": 25, + "seed": 3368, + "actual_rows": 1000, + "actual_return": 3660.5823563933372 + }, + { + "split": "val", + "episode_idx": 29, + "seed": 3372, + "actual_rows": 1000, + "actual_return": 3708.77580422163 + }, + { + "split": "val", + "episode_idx": 47, + "seed": 3396, + "actual_rows": 1000, + "actual_return": 3656.931324362755 + }, + { + "split": "val", + "episode_idx": 58, + "seed": 3412, + "actual_rows": 1000, + "actual_return": 3665.4246680140495 + }, + { + "split": "val", + "episode_idx": 77, + "seed": 3434, + "actual_rows": 1000, + "actual_return": 3636.464349627495 + }, + { + "split": "val", + "episode_idx": 81, + "seed": 3440, + "actual_rows": 1000, + "actual_return": 3681.014655470848 + } + ], + "total_episodes": 100, + "total_rows": 98817, + "pass_gate_count": 100, + "mean_return": 3633.191219932437, + "minimum_return": 3057.7498267292976, + "probe_train_rows": 88817, + "probe_val_rows": 10000, + "actual_train_rows": 88817, + "actual_val_rows": 10000, + "per_episode_probe_replay_match": true + }, + "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_gate": 3000, + "effective_gate": 3000, + "operator": ">", + "field": "episode_raw_return", + "changed": false, + "basis": "Bounded original-condition action-IID probe accepted10/12 at original3000, so preserve source quality threshold.", + "accepted_episodes": 100 + }, + "normalization_source": "new Hopper converted dataset and actual VLA training dataset_statistics.json" +} diff --git a/latency-aware/hopper/vla/starvla-qwenoft-h1/hopper_rgb_state_profile_20260911T180943Z_openvla_native_continuous_projector_sft_5k/task_contract.json b/latency-aware/hopper/vla/starvla-qwenoft-h1/hopper_rgb_state_profile_20260911T180943Z_openvla_native_continuous_projector_sft_5k/task_contract.json new file mode 100644 index 0000000000000000000000000000000000000000..7637a7f4aa1370962370250374850b9fa4dc2373 --- /dev/null +++ b/latency-aware/hopper/vla/starvla-qwenoft-h1/hopper_rgb_state_profile_20260911T180943Z_openvla_native_continuous_projector_sft_5k/task_contract.json @@ -0,0 +1,62 @@ +{ + "action_space": { + "dtype": "float32", + "high": [ + 1.0, + 1.0, + 1.0 + ], + "labels": [ + "thigh_torque", + "leg_torque", + "foot_torque" + ], + "low": [ + -1.0, + -1.0, + -1.0 + ], + "type": "box" + }, + "base_prompt": "Move the Hopper robot forward while keeping its torso upright. Predict three continuous torques in [-1, 1] ordered as thigh, leg, and foot.", + "env_fps": 10.0, + "env_id": "LatencyBench/Hopper-v0", + "frame_stack": 1, + "make_kwargs": { + "base_env_id": "Hopper-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 + ], + "obs_fps": 10.0, + "registration_imports": [ + "latency_bench.envs.gymnasium_hopper" + ], + "state_space": { + "labels": [ + "torso_height", + "torso_angle", + "thigh_angle", + "leg_angle", + "foot_angle", + "torso_x_velocity", + "torso_z_velocity", + "torso_angular_velocity", + "thigh_angular_velocity", + "leg_angular_velocity", + "foot_angular_velocity" + ] + }, + "task_name": "hopper_rgb_state" +} diff --git a/latency-aware/hopper/vla/starvla-qwenpi_v3-h1/hopper_pi05_profile_gt2500_h1_5fps_g128_20260922/README.md b/latency-aware/hopper/vla/starvla-qwenpi_v3-h1/hopper_pi05_profile_gt2500_h1_5fps_g128_20260922/README.md new file mode 100644 index 0000000000000000000000000000000000000000..f0231e9e4bfd839beed3537e49a53ee06a41e3c7 --- /dev/null +++ b/latency-aware/hopper/vla/starvla-qwenpi_v3-h1/hopper_pi05_profile_gt2500_h1_5fps_g128_20260922/README.md @@ -0,0 +1,31 @@ +# hopper / qwenpi_v3 + +Training condition: `latency-aware`. Run: `hopper_pi05_profile_gt2500_h1_5fps_g128_20260922`. + +Copied from [Standard-Pipeline](https://huggingface.co/latency-sensitive-bench/Standard-Pipeline/tree/2208875f92b21e3f8ec2363a30a6705dfe56cc04/hopper/profile_latency/Pi05) at `2208875f92b21e3f8ec2363a30a6705dfe56cc04`. + +- Checkpoint: `checkpoints/model.pt` +- Selection: only published VLA checkpoint +- Checkpoint SHA256: `93f66960f8eae69b52350e2d15271ef037dc8826332268fac69ce7c018d654c8` +- 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/hopper/vla/starvla-qwenpi_v3-h1/hopper_pi05_profile_gt2500_h1_5fps_g128_20260922" +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/hopper/vla/starvla-qwenpi_v3-h1/hopper_pi05_profile_gt2500_h1_5fps_g128_20260922/checkpoints/model.pt b/latency-aware/hopper/vla/starvla-qwenpi_v3-h1/hopper_pi05_profile_gt2500_h1_5fps_g128_20260922/checkpoints/model.pt new file mode 100644 index 0000000000000000000000000000000000000000..e7d7d25cde3e4f80211075478052337a3aa0886e --- /dev/null +++ b/latency-aware/hopper/vla/starvla-qwenpi_v3-h1/hopper_pi05_profile_gt2500_h1_5fps_g128_20260922/checkpoints/model.pt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:93f66960f8eae69b52350e2d15271ef037dc8826332268fac69ce7c018d654c8 +size 10922629277 diff --git a/latency-aware/hopper/vla/starvla-qwenpi_v3-h1/hopper_pi05_profile_gt2500_h1_5fps_g128_20260922/config.full.yaml b/latency-aware/hopper/vla/starvla-qwenpi_v3-h1/hopper_pi05_profile_gt2500_h1_5fps_g128_20260922/config.full.yaml new file mode 100644 index 0000000000000000000000000000000000000000..03da84df401fec997985ea6c707f624950a69561 --- /dev/null +++ b/latency-aware/hopper/vla/starvla-qwenpi_v3-h1/hopper_pi05_profile_gt2500_h1_5fps_g128_20260922/config.full.yaml @@ -0,0 +1,246 @@ +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: 3 + state_dim: 11 + 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: 3 + 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_gt2500_20260922/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 + labels: + - thigh_torque + - leg_torque + - foot_torque + low: + - -1.0 + - -1.0 + - -1.0 + type: box + base_prompt: Move the Hopper robot forward while keeping its torso upright. + Predict three continuous torques in [-1, 1] ordered as thigh, leg, and foot. + env_fps: 5.0 + env_id: LatencyBench/Hopper-v0 + frame_stack: 1 + make_kwargs: + base_env_id: Hopper-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 + obs_fps: 5.0 + registration_imports: + - latency_bench.envs.gymnasium_hopper + state_space: + labels: + - torso_height + - torso_angle + - thigh_angle + - leg_angle + - foot_angle + - torso_x_velocity + - torso_z_velocity + - torso_angular_velocity + - thigh_angular_velocity + - leg_angular_velocity + - foot_angular_velocity + task_name: hopper + active_action_dim: 3 +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: hopper_pi05_profile_gt2500_h1_5fps_g128_20260922 +run_root_dir: ${PI05_RUN_DIR}/profile_latency/vla_gt2500_20260922/training +seed: 42 +is_debug: false +version_id: '0.21' +wandb_project: latency-sensitive-bench +wandb_entity: dongqianyu99-zhejiang-university +wandb_name: hopper_pi05_profile_gt2500_h1_5fps_g128_20260922 +wandb_group: pi05-seven-env +wandb_tags: +- hopper +- 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 +config_yaml: ${PI05_RUN_DIR}/profile_latency/vla_gt2500_20260922/train.yaml +output_dir: ${PI05_RUN_DIR}/profile_latency/vla_gt2500_20260922/training/hopper_pi05_profile_gt2500_h1_5fps_g128_20260922 diff --git a/latency-aware/hopper/vla/starvla-qwenpi_v3-h1/hopper_pi05_profile_gt2500_h1_5fps_g128_20260922/config.yaml b/latency-aware/hopper/vla/starvla-qwenpi_v3-h1/hopper_pi05_profile_gt2500_h1_5fps_g128_20260922/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..82ffbbde4b690326f35826fe42a0c9921237a13f --- /dev/null +++ b/latency-aware/hopper/vla/starvla-qwenpi_v3-h1/hopper_pi05_profile_gt2500_h1_5fps_g128_20260922/config.yaml @@ -0,0 +1,123 @@ +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: 3 + state_dim: 11 + 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: 3 + 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 + labels: + - thigh_torque + - leg_torque + - foot_torque + low: + - -1.0 + - -1.0 + - -1.0 + type: box + base_prompt: Move the Hopper robot forward while keeping its torso upright. + Predict three continuous torques in [-1, 1] ordered as thigh, leg, and foot. + env_fps: 5.0 + env_id: LatencyBench/Hopper-v0 + frame_stack: 1 + make_kwargs: + base_env_id: Hopper-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 + obs_fps: 5.0 + registration_imports: + - latency_bench.envs.gymnasium_hopper + state_space: + labels: + - torso_height + - torso_angle + - thigh_angle + - leg_angle + - foot_angle + - torso_x_velocity + - torso_z_velocity + - torso_angular_velocity + - thigh_angular_velocity + - leg_angular_velocity + - foot_angular_velocity + task_name: hopper +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/hopper/vla/starvla-qwenpi_v3-h1/hopper_pi05_profile_gt2500_h1_5fps_g128_20260922/dataset_statistics.json b/latency-aware/hopper/vla/starvla-qwenpi_v3-h1/hopper_pi05_profile_gt2500_h1_5fps_g128_20260922/dataset_statistics.json new file mode 100644 index 0000000000000000000000000000000000000000..b93096bbdc9cb9ebccedded01bcb2ccae138309c --- /dev/null +++ b/latency-aware/hopper/vla/starvla-qwenpi_v3-h1/hopper_pi05_profile_gt2500_h1_5fps_g128_20260922/dataset_statistics.json @@ -0,0 +1,123 @@ +{ + "new_embodiment": { + "action": { + "mean": [ + 0.004830775782465935, + 0.6506988406181335, + -0.16816289722919464 + ], + "std": [ + 0.2572890818119049, + 0.6471112370491028, + 0.8130112886428833 + ], + "max": [ + 0.92793208360672, + 1.0, + 1.0 + ], + "min": [ + -0.6471852660179138, + -1.0, + -1.0 + ], + "q01": [ + -0.535171184539795, + -1.0, + -1.0 + ], + "q99": [ + 0.6878180891275406, + 1.0, + 1.0 + ], + "mask": [ + true, + true, + true + ] + }, + "state": { + "mean": [ + 0.256259024143219, + -0.2865329384803772, + 0.40575700998306274, + 0.7239107489585876, + -0.09208492189645767, + -0.27265146374702454, + 0.23160217702388763, + -0.001304655452258885, + 0.15476475656032562, + -0.0018315694760531187, + -0.0054734814912080765 + ], + "std": [ + 0.32620856165885925, + 0.27870386838912964, + 0.22778616845607758, + 0.2515462040901184, + 0.6812174916267395, + 0.2129003256559372, + 0.389985591173172, + 0.1715511977672577, + 0.2775718867778778, + 0.40339428186416626, + 0.58819580078125 + ], + "max": [ + 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 + ], + "q01": [ + -0.4246590751409531, + -0.8132983136177063, + -0.05858398914337158, + 0.08747239470481873, + -0.9689728409051895, + -0.8721107774972916, + -0.4223246592283249, + -0.42529766380786893, + -0.36811228275299074, + -0.9744406068325042, + -1.0 + ], + "q99": [ + 0.8820206296443941, + 0.38169503450393677, + 0.9299243867397309, + 0.9230736374855042, + 0.935717363357544, + 0.10131574153900186, + 0.9169844460487366, + 0.36585895061492923, + 0.8261520147323609, + 0.9779282450675965, + 1.0 + ] + }, + "num_transitions": 87404, + "num_trajectories": 90 + } +} \ No newline at end of file diff --git a/latency-aware/hopper/vla/starvla-qwenpi_v3-h1/hopper_pi05_profile_gt2500_h1_5fps_g128_20260922/latency_prompt_map.json b/latency-aware/hopper/vla/starvla-qwenpi_v3-h1/hopper_pi05_profile_gt2500_h1_5fps_g128_20260922/latency_prompt_map.json new file mode 100644 index 0000000000000000000000000000000000000000..b307ec7bb2ec5ee7998d49478e96b24ae3f2c84a --- /dev/null +++ b/latency-aware/hopper/vla/starvla-qwenpi_v3-h1/hopper_pi05_profile_gt2500_h1_5fps_g128_20260922/latency_prompt_map.json @@ -0,0 +1,12 @@ +{ + "1": { + "prompt": "Move the Hopper robot forward while keeping its torso upright. Predict three continuous torques in [-1, 1] ordered as thigh, leg, and 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 Hopper robot forward while keeping its torso upright. Predict three continuous torques in [-1, 1] ordered as thigh, leg, and 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 + } +} \ No newline at end of file diff --git a/latency-aware/hopper/vla/starvla-qwenpi_v3-h1/hopper_pi05_profile_gt2500_h1_5fps_g128_20260922/manifest.json b/latency-aware/hopper/vla/starvla-qwenpi_v3-h1/hopper_pi05_profile_gt2500_h1_5fps_g128_20260922/manifest.json new file mode 100644 index 0000000000000000000000000000000000000000..58dd565706228890110fe328b49c48e3bf8b99cf --- /dev/null +++ b/latency-aware/hopper/vla/starvla-qwenpi_v3-h1/hopper_pi05_profile_gt2500_h1_5fps_g128_20260922/manifest.json @@ -0,0 +1,143 @@ +{ + "dataset_name": "hopper_h1_5fps_profile_gt2500", + "env_name": "hopper", + "episodes": 90, + "frames": 87404, + "task_prompts": [ + "Move the Hopper robot forward while keeping its torso upright. Predict three continuous torques in [-1, 1] ordered as thigh, leg, and 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 Hopper robot forward while keeping its torso upright. Predict three continuous torques in [-1, 1] ordered as thigh, leg, and 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." + ], + "format": "lerobot_v2_image_parquet", + "source": "${PI05_RUN_DIR}/profile_latency/data/raw_profile_gt2500", + "integration_name": "gymnasium", + "task_name": "hopper", + "action_layout": "gymnasium_continuous_v1", + "action_labels": [ + "thigh_torque", + "leg_torque", + "foot_torque" + ], + "carrier_action_labels": [ + "thigh_torque", + "leg_torque", + "foot_torque" + ], + "action_dim": 3, + "active_action_dim": 3, + "action_carrier": "native", + "rows_unit": "decision_step", + "obs_fps": 5.0, + "obs_stride_raw_frames": 1, + "uses_state": true, + "state_dim": 11, + "state_labels": [ + "torso_height", + "torso_angle", + "thigh_angle", + "leg_angle", + "foot_angle", + "torso_x_velocity", + "torso_z_velocity", + "torso_angular_velocity", + "thigh_angular_velocity", + "leg_angular_velocity", + "foot_angular_velocity" + ], + "state_normalization": { + "type": "min_max", + "min": [ + 0.7025294303894043, + -0.18769855797290802, + -1.4441453218460083, + -1.9619364738464355, + -0.9252399802207947, + -0.014216306619346142, + -5.1731038093566895, + -6.168967247009277, + -10.0, + -10.0, + -10.0 + ], + "max": [ + 1.819760799407959, + 0.19951485097408295, + 0.0425863191485405, + 0.15077050030231476, + 0.9662268757820129, + 7.153366565704346, + 3.250192880630493, + 6.1806535720825195, + 7.204126834869385, + 10.0, + 10.0 + ] + }, + "robot_type": "rl_games_gymnasium", + "latency_prompt_map_path": "${PI05_RUN_DIR}/profile_latency/data/lerobot/hopper_h1_5fps_profile_gt2500/latency_prompt_map.json", + "custom_mixtures_path": "${PI05_RUN_DIR}/profile_latency/data/lerobot/_generated_mixtures/hopper_h1_5fps_profile_gt2500.json", + "gymnasium_task": { + "action_space": { + "dtype": "float32", + "high": [ + 1.0, + 1.0, + 1.0 + ], + "labels": [ + "thigh_torque", + "leg_torque", + "foot_torque" + ], + "low": [ + -1.0, + -1.0, + -1.0 + ], + "type": "box" + }, + "base_prompt": "Move the Hopper robot forward while keeping its torso upright. Predict three continuous torques in [-1, 1] ordered as thigh, leg, and foot.", + "env_fps": 5.0, + "env_id": "LatencyBench/Hopper-v0", + "frame_stack": 1, + "make_kwargs": { + "base_env_id": "Hopper-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 + ], + "obs_fps": 5.0, + "registration_imports": [ + "latency_bench.envs.gymnasium_hopper" + ], + "state_space": { + "labels": [ + "torso_height", + "torso_angle", + "thigh_angle", + "leg_angle", + "foot_angle", + "torso_x_velocity", + "torso_z_velocity", + "torso_angular_velocity", + "thigh_angular_velocity", + "leg_angular_velocity", + "foot_angular_velocity" + ] + }, + "task_name": "hopper" + }, + "validation_dataset_name": "hopper_h1_5fps_profile_gt2500__val", + "validation_episodes": 10, + "validation_frames": 9776 +} \ No newline at end of file diff --git a/latency-aware/hopper/vla/starvla-qwenpi_v3-h1/hopper_pi05_profile_gt2500_h1_5fps_g128_20260922/provenance.json b/latency-aware/hopper/vla/starvla-qwenpi_v3-h1/hopper_pi05_profile_gt2500_h1_5fps_g128_20260922/provenance.json new file mode 100644 index 0000000000000000000000000000000000000000..2a96191a492d73163418bcbe0b612be5403c2067 --- /dev/null +++ b/latency-aware/hopper/vla/starvla-qwenpi_v3-h1/hopper_pi05_profile_gt2500_h1_5fps_g128_20260922/provenance.json @@ -0,0 +1,34 @@ +{ + "task": "hopper", + "model": "qwenpi_v3", + "training_condition": "latency-aware", + "training_run_id": "hopper_pi05_profile_gt2500_h1_5fps_g128_20260922", + "source": { + "repo_id": "latency-sensitive-bench/Standard-Pipeline", + "revision": "2208875f92b21e3f8ec2363a30a6705dfe56cc04", + "path": "hopper/profile_latency/Pi05", + "url": "https://huggingface.co/latency-sensitive-bench/Standard-Pipeline/tree/2208875f92b21e3f8ec2363a30a6705dfe56cc04/hopper/profile_latency/Pi05" + }, + "hf_repo_id": "latency-sensitive-bench/benchmark-models", + "hf_path": "latency-aware/hopper/vla/starvla-qwenpi_v3-h1/hopper_pi05_profile_gt2500_h1_5fps_g128_20260922", + "checkpoint": { + "source_file": "hopper/profile_latency/Pi05/checkpoints/model.pt", + "source_sha256": "93f66960f8eae69b52350e2d15271ef037dc8826332268fac69ce7c018d654c8", + "source_bytes": 10922629277, + "file": "checkpoints/model.pt", + "selection_rule": "only published VLA checkpoint", + "method": "copy", + "sha256": "93f66960f8eae69b52350e2d15271ef037dc8826332268fac69ce7c018d654c8", + "bytes": 10922629277 + }, + "config_source": "hopper/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/hopper/vla/starvla-qwenpi_v3-h1/hopper_pi05_profile_gt2500_h1_5fps_g128_20260922/source/README.md b/latency-aware/hopper/vla/starvla-qwenpi_v3-h1/hopper_pi05_profile_gt2500_h1_5fps_g128_20260922/source/README.md new file mode 100644 index 0000000000000000000000000000000000000000..3cab54929dcc01ebcea30fe5a0d6c623b4356c3d --- /dev/null +++ b/latency-aware/hopper/vla/starvla-qwenpi_v3-h1/hopper_pi05_profile_gt2500_h1_5fps_g128_20260922/source/README.md @@ -0,0 +1,5 @@ +# hopper Pi0.5 profile-latency VLA + +Fresh QwenPI_v3,5000 updates,seed42,global128 (micro64 × accumulation1 ×2 GPUs),GC off,ZeRO2,H1,RGB224,5/5FPS,native state/action and train-only state min-max once. Dataset used the explicitly approved >2500 gate without an attempt ceiling: 218 total attempts, 53 earlier >3000 accepted records retained plus 47 new >2500 records. + +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/hopper/vla/starvla-qwenpi_v3-h1/hopper_pi05_profile_gt2500_h1_5fps_g128_20260922/source/provenance.json b/latency-aware/hopper/vla/starvla-qwenpi_v3-h1/hopper_pi05_profile_gt2500_h1_5fps_g128_20260922/source/provenance.json new file mode 100644 index 0000000000000000000000000000000000000000..d76025946f5e98df4da9e9c2865e1ce6600e4c25 --- /dev/null +++ b/latency-aware/hopper/vla/starvla-qwenpi_v3-h1/hopper_pi05_profile_gt2500_h1_5fps_g128_20260922/source/provenance.json @@ -0,0 +1,512 @@ +{ + "task": "hopper", + "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": "hopper_pi05_profile_gt2500_h1_5fps_g128_20260922", + "condition": "profile_latency", + "source": { + "task": "hopper", + "condition": "profile_latency", + "teacher_checkpoint_path": "${PI05_RUN_DIR}/profile_latency/teacher/checkpoints/pi05_hopper_profile_appo_h1_5fps_g128_20260921/checkpoint_p0/checkpoint_000019536_10002432.pth", + "teacher_checkpoint_sha256": "9b78ae32c795cf0eecbd2b66e04f2775091a9756a40e5f5635791ac2b2ca4b86", + "teacher": { + "state": "TEACHER_FULL_TRACE_AUDIT_PASSED", + "verified_at": "2026-09-21T14:09:31.112834+00:00", + "task": "hopper", + "model_revision": "408930d8a85f0cb316dc3bdebfb8142526570355", + "P_profile_sha256": "1289653bf6e8591cdd5cf805ad539ddd1cb041cdad08a9f2833597fe72271e1e", + "P_warnings": [ + { + "session": 1, + "warmup_status": "unstable", + "warmup_stability_reason": "drifting", + "samples_excluded": 100 + } + ], + "actual_final_env_steps": 10002432, + "actual_final_train_step": 19536, + "selection": { + "selected": "final", + "candidates": { + "final": { + "checkpoint": "${PI05_RUN_DIR}/profile_latency/teacher/checkpoints/pi05_hopper_profile_appo_h1_5fps_g128_20260921/checkpoint_p0/checkpoint_000019536_10002432.pth", + "sha256": "9b78ae32c795cf0eecbd2b66e04f2775091a9756a40e5f5635791ac2b2ca4b86", + "episodes": 20, + "seeds": [ + 104771, + 104772, + 104773, + 104774, + 104775, + 104776, + 104777, + 104778, + 104779, + 104780, + 104781, + 104782, + 104783, + 104784, + 104785, + 104786, + 104787, + 104788, + 104789, + 104790 + ], + "returns": [ + 2224.1403163381674, + 2819.128114077857, + 3489.646074260514, + 2009.7554991069278, + 2113.0389978032463, + 2530.889359173486, + 3528.0257119878297, + 3480.216905777773, + 1947.2163728885328, + 2019.6803983866998, + 271.37905453360946, + 2083.796228363727, + 3447.6906733909796, + 2591.8151761056497, + 265.78937272271213, + 1892.4932081793627, + 1981.7056002712736, + 3621.6136520722443, + 3614.0425221471028, + 3542.033646827985 + ], + "mean": 2473.704844220784, + "population_std": 983.5990191484575, + "strict_gt3000": 7 + }, + "training_best": { + "checkpoint": "${PI05_RUN_DIR}/profile_latency/teacher/checkpoints/pi05_hopper_profile_appo_h1_5fps_g128_20260921/checkpoint_p0/best_000019536_10002432_reward_2535.456.pth", + "sha256": "f0d71c11d6f468510a6b1e98d528060bfe213e094ae81a13a7c9129af2f5b2ec", + "episodes": 20, + "seeds": [ + 104771, + 104772, + 104773, + 104774, + 104775, + 104776, + 104777, + 104778, + 104779, + 104780, + 104781, + 104782, + 104783, + 104784, + 104785, + 104786, + 104787, + 104788, + 104789, + 104790 + ], + "returns": [ + 2224.1403163381674, + 2819.128114077857, + 3489.646074260514, + 2009.7554991069278, + 2113.0389978032463, + 2530.889359173486, + 3528.0257119878297, + 3480.216905777773, + 1947.2163728885328, + 2019.6803983866998, + 271.37905453360946, + 2083.796228363727, + 3447.6906733909796, + 2591.8151761056497, + 265.78937272271213, + 1892.4932081793627, + 1981.7056002712736, + 3621.6136520722443, + 3614.0425221471028, + 3542.033646827985 + ], + "mean": 2473.704844220784, + "population_std": 983.5990191484575, + "strict_gt3000": 7 + } + }, + "completed_at": "2026-09-21T14:06:00.799410+00:00", + "final_checkpoint_env_steps": 10002432, + "final_checkpoint_train_step": 19536 + }, + "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": [ + 2224.1403163381674, + 2819.128114077857, + 3489.646074260514, + 2009.7554991069278, + 2113.0389978032463, + 2530.889359173486, + 3528.0257119878297, + 3480.216905777773, + 1947.2163728885328, + 2019.6803983866998, + 271.37905453360946, + 2083.796228363727, + 3447.6906733909796, + 2591.8151761056497, + 265.78937272271213, + 1892.4932081793627, + 1981.7056002712736, + 3621.6136520722443, + 3614.0425221471028, + 3542.033646827985 + ], + "lengths": [ + 594, + 757, + 1000, + 544, + 593, + 682, + 1000, + 1000, + 528, + 539, + 134, + 566, + 919, + 698, + 132, + 508, + 539, + 1000, + 983, + 1000 + ], + "mean_return": 2473.704844220784, + "population_std_return": 983.5990191484575, + "mean_length": 685.8, + "raw_steps": 13716, + "strict_gt3000": 7, + "actions": 10606, + "action_latency_mean_ms": 199.08640804375509, + "action_latency_p95_ms": 256.6691669408085, + "commands_requiring_native_clip": 0, + "invalid_actions": 0, + "dropped_actions": 0, + "submitted_observations": 10606, + "dropped_observations": 3110, + "observation_drop_fraction": 0.2267424905220181, + "config_sha256": "39166d114a7d4b46e64da14382bd644f4095e7eb36fc6eae52479c2545093f81", + "state_contract": "Native state11 is bound to manifest/loader QA/saved reload; 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": [ + 2224.1403163381674, + 2819.128114077857, + 3489.646074260514, + 2009.7554991069278, + 2113.0389978032463, + 2530.889359173486, + 3528.0257119878297, + 3480.216905777773, + 1947.2163728885328, + 2019.6803983866998, + 271.37905453360946, + 2083.796228363727, + 3447.6906733909796, + 2591.8151761056497, + 265.78937272271213, + 1892.4932081793627, + 1981.7056002712736, + 3621.6136520722443, + 3614.0425221471028, + 3542.033646827985 + ], + "lengths": [ + 594, + 757, + 1000, + 544, + 593, + 682, + 1000, + 1000, + 528, + 539, + 134, + 566, + 919, + 698, + 132, + 508, + 539, + 1000, + 983, + 1000 + ], + "mean_return": 2473.704844220784, + "population_std_return": 983.5990191484575, + "mean_length": 685.8, + "raw_steps": 13716, + "strict_gt3000": 7, + "actions": 10606, + "action_latency_mean_ms": 199.08640804375509, + "action_latency_p95_ms": 256.6691669408085, + "commands_requiring_native_clip": 0, + "invalid_actions": 0, + "dropped_actions": 0, + "submitted_observations": 10606, + "dropped_observations": 3110, + "observation_drop_fraction": 0.2267424905220181, + "config_sha256": "8447bbc671595a369933dff026480db4f94846a7970e4e8ac6731abdf21d0138", + "state_contract": "Native state11 is bound to manifest/loader QA/saved reload; no per-step state vector is recorded." + } + }, + "E10": { + "episodes": 10, + "seeds": [ + 104771, + 104772, + 104773, + 104774, + 104775, + 104776, + 104777, + 104778, + 104779, + 104780 + ], + "returns": [ + 2224.1403163381674, + 2819.128114077857, + 3489.646074260514, + 2009.7554991069278, + 2113.0389978032463, + 2530.889359173486, + 3528.0257119878297, + 3480.216905777773, + 1947.2163728885328, + 2019.6803983866998 + ], + "lengths": [ + 594, + 757, + 1000, + 544, + 593, + 682, + 1000, + 1000, + 528, + 539 + ], + "mean_return": 2616.173774980103, + "population_std_return": 629.868919379035, + "mean_length": 723.7, + "raw_steps": 7237, + "strict_gt3000": 3, + "actions": 5577, + "action_latency_mean_ms": 199.29742636488007, + "action_latency_p95_ms": 257.7097693485515, + "commands_requiring_native_clip": 0, + "invalid_actions": 0, + "dropped_actions": 0, + "submitted_observations": 5577, + "dropped_observations": 1660, + "observation_drop_fraction": 0.22937681359679427, + "config_sha256": "8a5246d083413a735ed88e2e3b9a5c4f7b49896abfd3153488d9fcd1edd62176", + "state_contract": "Native state11 is bound to manifest/loader QA/saved reload; no per-step state vector is recorded." + }, + "probe": { + "attempts": 12, + "accepted": 2, + "rejected": 10, + "source_state_sha256": "fad846336f923f7f0bf755c2d5cdbbcc2a8c4f955556c2299ef088cbd20d1f72", + "accepted_specs": [ + { + "attempt_idx": 4, + "episode_idx": 0, + "episode_length": 1000, + "episode_raw_frames": 1000, + "episode_raw_return": 3573.1784443855286, + "seed": 3337, + "split": "train" + }, + { + "attempt_idx": 9, + "episode_idx": 1, + "episode_length": 1000, + "episode_raw_frames": 1000, + "episode_raw_return": 3552.9572883844376, + "seed": 3342, + "split": "val" + } + ] + }, + "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" + }, + "profile_source": { + "model_revision": "408930d8a85f0cb316dc3bdebfb8142526570355", + "checkpoint_sha256": "c35ddc57fef8fd096adca3df9ecf94662824c26b311c3884508f07646231f05b", + "profile_sha256": "1289653bf6e8591cdd5cf805ad539ddd1cb041cdad08a9f2833597fe72271e1e", + "source_run_id": "20260921T094220318783Z", + "P_publication_revision": "e1a9bdd566b96d5da4703f769d40eab656af491e", + "warnings": [ + { + "session": 1, + "warmup_status": "unstable", + "warmup_stability_reason": "drifting", + "samples_excluded": 100 + } + ] + }, + "dataset": { + "attempts": 218, + "accepted": 100, + "old_history": { + "attempts": 120, + "gate": "episode_raw_return > 3000", + "accepted_reused": 53, + "rejected": 67, + "rejections_reclassified": false + }, + "continuation": { + "start_attempt_idx": 120, + "attempts": 98, + "accepted": 47, + "rejected": 51, + "gate": "episode_raw_return > 2500", + "attempt_limit": null + }, + "max_total_attempts": null, + "splits": { + "train": { + "rows": 87404, + "episodes": 90, + "return_min": 2501.0231088399887, + "action_min": -1.0, + "action_max": 1.0 + }, + "val": { + "rows": 9776, + "episodes": 10, + "return_min": 2859.5413383841515, + "action_min": -1.0, + "action_max": 1.0 + } + }, + "sha256": { + "metadata.json": "6a0e1ef5b903e923c37f5091eeece485d4a00dfc7a8c63a95c523cf59399531a", + "train.parquet": "47b4bd70080aa670c285f207aca9cc2f3c803d1f07c58fa0faa3a585e325939a", + "val.parquet": "af5561f9912e9a580bc8112875cf11a55e7014e418677293428e32fb911962e0" + } + }, + "probe_state": { + "path": "${PI05_RUN_DIR}/profile_latency/data/probe_state_gt2500.json", + "sha256": "eae3de415d0cced764c95c9ed12ae9b70a2a02ae30f7bbbc4bb461cd7f3da675", + "attempted": 218, + "accepted": 100, + "rejected": 118, + "original_12_sha256": "fad846336f923f7f0bf755c2d5cdbbcc2a8c4f955556c2299ef088cbd20d1f72", + "original_120_sha256": "8427adc4e968df0ea8467a8848462e51a481a9d31a2dbbc7f5458e3700da3a9b" + }, + "evaluation_prompt_keys": [ + "1", + "2" + ], + "state_dim": 11, + "action_dim": 3, + "action_units": "native continuous; environment Box clips pre-clip values outside [-1,1]", + "preclip_action_range": [ + -1.0, + 1.0 + ], + "commands_requiring_native_clip": 0, + "teacher_seed": 3333, + "actual_final_env_steps": 10002432, + "selected_teacher": "final", + "selection_mean": 2473.704844220784, + "selection_strict_gt3000": 7, + "E10_repeats_first10_selection_seeds": true, + "loader_checks": { + "train": { + "rows": 87404, + "indices": [ + 0, + 43702, + 87403 + ], + "state_train_minmax_exact_once": true, + "action_image_prompt_exact": true + }, + "val": { + "rows": 9776, + "indices": [ + 0, + 4888, + 9775 + ], + "state_train_minmax_exact_once": true, + "action_image_prompt_exact": true + } + } + }, + "training_config_sha256": "1f30ce61cfd0ee83b15c2b34ca7f37ce2de836d7693aa7b4f37d4e921befb17c", + "dataset_manifest_sha256": "32324928c5d8f933ae59f76c0c9ac4121d631568a552348449fe7555d1f30e5e" +} diff --git a/latency-aware/hopper/vla/starvla-qwenpi_v3-h1/hopper_pi05_profile_gt2500_h1_5fps_g128_20260922/task_contract.json b/latency-aware/hopper/vla/starvla-qwenpi_v3-h1/hopper_pi05_profile_gt2500_h1_5fps_g128_20260922/task_contract.json new file mode 100644 index 0000000000000000000000000000000000000000..674ea753e4b61eb10e05434c9a1bdea8f4f28f61 --- /dev/null +++ b/latency-aware/hopper/vla/starvla-qwenpi_v3-h1/hopper_pi05_profile_gt2500_h1_5fps_g128_20260922/task_contract.json @@ -0,0 +1,62 @@ +{ + "action_space": { + "dtype": "float32", + "high": [ + 1.0, + 1.0, + 1.0 + ], + "labels": [ + "thigh_torque", + "leg_torque", + "foot_torque" + ], + "low": [ + -1.0, + -1.0, + -1.0 + ], + "type": "box" + }, + "base_prompt": "Move the Hopper robot forward while keeping its torso upright. Predict three continuous torques in [-1, 1] ordered as thigh, leg, and foot.", + "env_fps": 5.0, + "env_id": "LatencyBench/Hopper-v0", + "frame_stack": 1, + "make_kwargs": { + "base_env_id": "Hopper-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 + ], + "obs_fps": 5.0, + "registration_imports": [ + "latency_bench.envs.gymnasium_hopper" + ], + "state_space": { + "labels": [ + "torso_height", + "torso_angle", + "thigh_angle", + "leg_angle", + "foot_angle", + "torso_x_velocity", + "torso_z_velocity", + "torso_angular_velocity", + "thigh_angular_velocity", + "leg_angular_velocity", + "foot_angular_velocity" + ] + }, + "task_name": "hopper" +} diff --git a/latency-aware/hopper/vla/starvla-qwenpi_v3-h1/hopper_pi05_profile_gt2500_h1_5fps_g128_20260922/validation.json b/latency-aware/hopper/vla/starvla-qwenpi_v3-h1/hopper_pi05_profile_gt2500_h1_5fps_g128_20260922/validation.json new file mode 100644 index 0000000000000000000000000000000000000000..3d6c70dbb0f4fa6533558e2ef56705328f747b99 --- /dev/null +++ b/latency-aware/hopper/vla/starvla-qwenpi_v3-h1/hopper_pi05_profile_gt2500_h1_5fps_g128_20260922/validation.json @@ -0,0 +1,10 @@ +{ + "state": "CPU_STATE_DICT_AND_BUNDLE_HASH_VERIFIED", + "checkpoint_sha256": "93f66960f8eae69b52350e2d15271ef037dc8826332268fac69ce7c018d654c8", + "tensor_count": 1386, + "all_floating_tensors_finite": true, + "training_steps": 5000, + "global_batch": 128, + "gpu_forward": "NOT_RUN_AT_PUBLICATION", + "verified_at": "2026-09-22T01:31:19.235344+00:00" +} diff --git a/zero-latency/hopper/small-policy/sample-factory-v1/README.md b/zero-latency/hopper/small-policy/sample-factory-v1/README.md new file mode 100644 index 0000000000000000000000000000000000000000..002e41d1860bcd6a89aec785e6830bb34608e03b --- /dev/null +++ b/zero-latency/hopper/small-policy/sample-factory-v1/README.md @@ -0,0 +1,26 @@ +# hopper / sample-factory-appo + +Training condition: `zero-latency`. Run: `gymnasium_hopper_rgb_state_zero_latency_10m`. + +Copied from [Standard-Pipeline](https://huggingface.co/latency-sensitive-bench/Standard-Pipeline/tree/2208875f92b21e3f8ec2363a30a6705dfe56cc04/hopper/zero_latency/small_model) at `2208875f92b21e3f8ec2363a30a6705dfe56cc04`. + +- Checkpoint: `checkpoint.pth` +- Selection: existing zero-latency training best +- Checkpoint SHA256: `c003949b0f44ce2bd249aed8dc25cde028b1ee8ce9e9921f023fbdd3da6389ed` +- 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/hopper/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/hopper/small-policy/sample-factory-v1/checkpoint.pth b/zero-latency/hopper/small-policy/sample-factory-v1/checkpoint.pth new file mode 100644 index 0000000000000000000000000000000000000000..2737eb6af8b76cb3020d99184fcd4e1b0ad77146 --- /dev/null +++ b/zero-latency/hopper/small-policy/sample-factory-v1/checkpoint.pth @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c003949b0f44ce2bd249aed8dc25cde028b1ee8ce9e9921f023fbdd3da6389ed +size 27445 diff --git a/zero-latency/hopper/small-policy/sample-factory-v1/config.json b/zero-latency/hopper/small-policy/sample-factory-v1/config.json new file mode 100644 index 0000000000000000000000000000000000000000..63cb37b9a105bbb4afc4a28dadf805ef8538e365 --- /dev/null +++ b/zero-latency/hopper/small-policy/sample-factory-v1/config.json @@ -0,0 +1,254 @@ +{ + "help": false, + "algo": "APPO", + "env": "latency_gymnasium", + "experiment": "gymnasium_hopper_rgb_state_zero_latency_10m", + "train_dir": "/mnt/checkpoints/latency-sensitive-bench/small_models/hopper_rgb_state", + "restart_behavior": "overwrite", + "device": "gpu", + "seed": 3333, + "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": 10000, + "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": false, + "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": 1.0, + "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": 10, + "decorrelate_envs_on_one_worker": true, + "actor_worker_gpus": [], + "set_workers_cpu_affinity": true, + "force_envs_single_thread": true, + "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": 3, + "load_checkpoint_kind": "latest", + "save_milestones_sec": -1, + "save_best_every_sec": 60, + "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": 4, + "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": "hopper_rgb_state", + "gym_env_id": "LatencyBench/HopperRgbState-v0", + "gym_make_kwargs_json": "{\"base_env_id\": \"Hopper-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_hopper_rgb_state\"]", + "gym_action_space_json": "{\"type\": \"box\", \"labels\": [\"thigh_torque\", \"leg_torque\", \"foot_torque\"], \"low\": [-1.0, -1.0, -1.0], \"high\": [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]", + "gym_base_prompt": "Move the Hopper robot forward while keeping its torso upright. Predict three continuous torques in [-1, 1] ordered as thigh, leg, and foot.", + "gym_state_labels_json": "[\"torso_height\", \"torso_angle\", \"thigh_angle\", \"leg_angle\", \"foot_angle\", \"torso_x_velocity\", \"torso_z_velocity\", \"torso_angular_velocity\", \"thigh_angular_velocity\", \"leg_angular_velocity\", \"foot_angular_velocity\"]", + "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": 10, + "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/small_models/hopper_rgb_state/gymnasium_hopper_rgb_state_zero_latency_10m/episode_metrics.jsonl", + "command_line": "--mode train --algo APPO --env latency_gymnasium --experiment gymnasium_hopper_rgb_state_zero_latency_10m --train_dir /mnt/checkpoints/latency-sensitive-bench/small_models/hopper_rgb_state --restart_behavior overwrite --device gpu --seed 3333 --episode_metrics_path /mnt/checkpoints/latency-sensitive-bench/small_models/hopper_rgb_state/gymnasium_hopper_rgb_state_zero_latency_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 10000 --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 --initial_stddev 1.0 --env_framestack 4 --gamma 0.99 --gae_lambda 0.95 --ppo_clip_ratio 0.2 --ppo_clip_value 1.0 --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 --decorrelate_experience_max_seconds 10 --save_every_sec 600 --keep_checkpoints 3 --save_best_every_sec 60 --save_best_after 100000 --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 False --with_vtrace False --decorrelate_envs_on_one_worker True --set_workers_cpu_affinity True --force_envs_single_thread True --actor_critic_share_weights True --encoder_mlp_layers 64 64 --latency-type zero --add-latency-info False --eval-episodes 10 --eval-parallel-envs 1 --eval-max-steps 1000 --eval-deterministic True --gym-task-name hopper_rgb_state --gym-env-id LatencyBench/HopperRgbState-v0 --gym-make-kwargs-json {\"base_env_id\": \"Hopper-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_hopper_rgb_state\"] --gym-action-space-json {\"type\": \"box\", \"labels\": [\"thigh_torque\", \"leg_torque\", \"foot_torque\"], \"low\": [-1.0, -1.0, -1.0], \"high\": [1.0, 1.0, 1.0], \"dtype\": \"float32\"} --gym-noop-action-json [0.0, 0.0, 0.0] --gym-base-prompt Move the Hopper robot forward while keeping its torso upright. Predict three continuous torques in [-1, 1] ordered as thigh, leg, and foot. --gym-state-labels-json [\"torso_height\", \"torso_angle\", \"thigh_angle\", \"leg_angle\", \"foot_angle\", \"torso_x_velocity\", \"torso_z_velocity\", \"torso_angular_velocity\", \"thigh_angular_velocity\", \"leg_angular_velocity\", \"foot_angular_velocity\"] --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_hopper_rgb_state_zero_latency_10m", + "train_dir": "/mnt/checkpoints/latency-sensitive-bench/small_models/hopper_rgb_state", + "restart_behavior": "overwrite", + "device": "gpu", + "seed": 3333, + "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": 10000, + "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": false, + "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": 1.0, + "with_vtrace": false, + "max_grad_norm": 3.5, + "learning_rate": 0.00295, + "lr_schedule_kl_threshold": 0.008, + "normalize_input": true, + "decorrelate_experience_max_seconds": 10, + "decorrelate_envs_on_one_worker": true, + "set_workers_cpu_affinity": true, + "force_envs_single_thread": true, + "train_for_env_steps": 10000000, + "save_every_sec": 600, + "keep_checkpoints": 3, + "save_best_every_sec": 60, + "save_best_after": 100000, + "encoder_mlp_layers": [ + 64, + 64 + ], + "use_rnn": false, + "nonlinearity": "tanh", + "policy_initialization": "torch_default", + "actor_critic_share_weights": true, + "adaptive_stddev": false, + "initial_stddev": 1.0, + "use_env_info_cache": false, + "env_framestack": 4, + "gym_task_name": "hopper_rgb_state", + "gym_env_id": "LatencyBench/HopperRgbState-v0", + "gym_make_kwargs_json": "{\"base_env_id\": \"Hopper-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_hopper_rgb_state\"]", + "gym_action_space_json": "{\"type\": \"box\", \"labels\": [\"thigh_torque\", \"leg_torque\", \"foot_torque\"], \"low\": [-1.0, -1.0, -1.0], \"high\": [1.0, 1.0, 1.0], \"dtype\": \"float32\"}", + "gym_noop_action_json": "[0.0, 0.0, 0.0]", + "gym_base_prompt": "Move the Hopper robot forward while keeping its torso upright. Predict three continuous torques in [-1, 1] ordered as thigh, leg, and foot.", + "gym_state_labels_json": "[\"torso_height\", \"torso_angle\", \"thigh_angle\", \"leg_angle\", \"foot_angle\", \"torso_x_velocity\", \"torso_z_velocity\", \"torso_angular_velocity\", \"thigh_angular_velocity\", \"leg_angular_velocity\", \"foot_angular_velocity\"]", + "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": 10, + "eval_parallel_envs": 1, + "eval_max_steps": 1000, + "eval_deterministic": true, + "episode_metrics_path": "/mnt/checkpoints/latency-sensitive-bench/small_models/hopper_rgb_state/gymnasium_hopper_rgb_state_zero_latency_10m/episode_metrics.jsonl" + }, + "git_hash": "08b9d5ff39a7a62781d6e76c649251c2cdb225e8", + "git_repo_name": "https://github.com/ZihanWang314/latency-sensitive-bench.git", + "eval_env_frameskip": 1, + "output_dir": "/mnt/results/latency-sensitive-bench/tasks/hopper_rgb_state/small_model_train" +} \ No newline at end of file diff --git a/zero-latency/hopper/small-policy/sample-factory-v1/provenance.json b/zero-latency/hopper/small-policy/sample-factory-v1/provenance.json new file mode 100644 index 0000000000000000000000000000000000000000..2053318dab43f7c47ef7e3c84acaf95ef1516524 --- /dev/null +++ b/zero-latency/hopper/small-policy/sample-factory-v1/provenance.json @@ -0,0 +1,44 @@ +{ + "task": "hopper", + "model": "sample-factory-appo", + "training_condition": "zero-latency", + "training_run_id": "gymnasium_hopper_rgb_state_zero_latency_10m", + "source": { + "repo_id": "latency-sensitive-bench/Standard-Pipeline", + "revision": "2208875f92b21e3f8ec2363a30a6705dfe56cc04", + "path": "hopper/zero_latency/small_model", + "url": "https://huggingface.co/latency-sensitive-bench/Standard-Pipeline/tree/2208875f92b21e3f8ec2363a30a6705dfe56cc04/hopper/zero_latency/small_model" + }, + "hf_repo_id": "latency-sensitive-bench/benchmark-models", + "hf_path": "zero-latency/hopper/small-policy/sample-factory-v1", + "checkpoint": { + "source_file": "hopper/zero_latency/small_model/checkpoint_p0/best_000011992_6139904_reward_3206.405.pth", + "source_sha256": "147c25f152d6a97d0a2231006f736f2c576f1d4f2b0ef4b38a4540e9a4cb383a", + "source_bytes": 76933, + "file": "checkpoint.pth", + "selection_rule": "existing zero-latency training best", + "method": "inference_export", + "sha256": "c003949b0f44ce2bd249aed8dc25cde028b1ee8ce9e9921f023fbdd3da6389ed", + "bytes": 27445, + "train_step": 11992, + "env_steps": 6139904, + "tensor_count": 15, + "tensor_equality_verified": true, + "removed_fields": [ + "best_performance", + "curr_lr", + "optimizer" + ] + }, + "config_source": "hopper/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/hopper/vla/starvla-qwengr00t-h1/hopper_l0_gr00t_h1_5fps_2h100_20260917/README.md b/zero-latency/hopper/vla/starvla-qwengr00t-h1/hopper_l0_gr00t_h1_5fps_2h100_20260917/README.md new file mode 100644 index 0000000000000000000000000000000000000000..12491fe3f8927bf3a4c1e97707a1a2d7483efb10 --- /dev/null +++ b/zero-latency/hopper/vla/starvla-qwengr00t-h1/hopper_l0_gr00t_h1_5fps_2h100_20260917/README.md @@ -0,0 +1,29 @@ +# hopper / qwengr00t + +Training condition: `zero-latency`. Run: `hopper_l0_gr00t_h1_5fps_2h100_20260917`. + +Copied from [Standard-Pipeline](https://huggingface.co/latency-sensitive-bench/Standard-Pipeline/tree/2208875f92b21e3f8ec2363a30a6705dfe56cc04/hopper/zero_latency/GR00T) at `2208875f92b21e3f8ec2363a30a6705dfe56cc04`. + +- Checkpoint: `checkpoints/model.pt` +- Selection: only published VLA checkpoint +- Checkpoint SHA256: `5ec4a5fbdfb7fac819a51b523f3735c6702b90a2440a2957c5cc699941ffbf1e` +- 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 = "zero-latency/hopper/vla/starvla-qwengr00t-h1/hopper_l0_gr00t_h1_5fps_2h100_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/hopper/vla/starvla-qwengr00t-h1/hopper_l0_gr00t_h1_5fps_2h100_20260917/checkpoints/model.pt b/zero-latency/hopper/vla/starvla-qwengr00t-h1/hopper_l0_gr00t_h1_5fps_2h100_20260917/checkpoints/model.pt new file mode 100644 index 0000000000000000000000000000000000000000..002327b3f2358861cc481e2292581962c8620131 --- /dev/null +++ b/zero-latency/hopper/vla/starvla-qwengr00t-h1/hopper_l0_gr00t_h1_5fps_2h100_20260917/checkpoints/model.pt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5ec4a5fbdfb7fac819a51b523f3735c6702b90a2440a2957c5cc699941ffbf1e +size 9976837923 diff --git a/zero-latency/hopper/vla/starvla-qwengr00t-h1/hopper_l0_gr00t_h1_5fps_2h100_20260917/config.full.yaml b/zero-latency/hopper/vla/starvla-qwengr00t-h1/hopper_l0_gr00t_h1_5fps_2h100_20260917/config.full.yaml new file mode 100644 index 0000000000000000000000000000000000000000..e017b840e09447506759e455f724e6e4a016b315 --- /dev/null +++ b/zero-latency/hopper/vla/starvla-qwengr00t-h1/hopper_l0_gr00t_h1_5fps_2h100_20260917/config.full.yaml @@ -0,0 +1,242 @@ +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: 3 + state_dim: 11 + 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: 3 + 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/hopper/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: 8 + 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 + labels: + - thigh_torque + - leg_torque + - foot_torque + low: + - -1.0 + - -1.0 + - -1.0 + type: box + base_prompt: Move the Hopper robot forward while keeping its torso upright. + Predict three continuous torques in [-1, 1] ordered as thigh, leg, and foot. + env_fps: 5 + env_id: LatencyBench/HopperRgbState-v0 + frame_stack: 1 + make_kwargs: + base_env_id: Hopper-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 + obs_fps: 5 + registration_imports: + - latency_bench.envs.gymnasium_hopper_rgb_state + state_space: + labels: + - torso_height + - torso_angle + - thigh_angle + - leg_angle + - foot_angle + - torso_x_velocity + - torso_z_velocity + - torso_angular_velocity + - thigh_angular_velocity + - leg_angular_velocity + - foot_angular_velocity + task_name: hopper_rgb_state + active_action_dim: 3 +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: 4 + 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: hopper_l0_gr00t_h1_5fps_2h100_20260917 +run_root_dir: /mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/hopper/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: hopper_l0_gr00t_h1_5fps_2h100_20260917 +wandb_group: gr00t-six-env-h1-5fps +wandb_tags: +- hopper +- 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/hopper/zero_latency/train.yaml +output_dir: /mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/hopper/zero_latency/vla/training/hopper_l0_gr00t_h1_5fps_2h100_20260917 diff --git a/zero-latency/hopper/vla/starvla-qwengr00t-h1/hopper_l0_gr00t_h1_5fps_2h100_20260917/config.yaml b/zero-latency/hopper/vla/starvla-qwengr00t-h1/hopper_l0_gr00t_h1_5fps_2h100_20260917/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..aa1851995e18a68100449551929a9f8c71446914 --- /dev/null +++ b/zero-latency/hopper/vla/starvla-qwengr00t-h1/hopper_l0_gr00t_h1_5fps_2h100_20260917/config.yaml @@ -0,0 +1,119 @@ +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: 3 + state_dim: 11 + 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: 3 + 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 + labels: + - thigh_torque + - leg_torque + - foot_torque + low: + - -1.0 + - -1.0 + - -1.0 + type: box + base_prompt: Move the Hopper robot forward while keeping its torso upright. + Predict three continuous torques in [-1, 1] ordered as thigh, leg, and foot. + env_fps: 5 + env_id: LatencyBench/HopperRgbState-v0 + frame_stack: 1 + make_kwargs: + base_env_id: Hopper-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 + obs_fps: 5 + registration_imports: + - latency_bench.envs.gymnasium_hopper_rgb_state + state_space: + labels: + - torso_height + - torso_angle + - thigh_angle + - leg_angle + - foot_angle + - torso_x_velocity + - torso_z_velocity + - torso_angular_velocity + - thigh_angular_velocity + - leg_angular_velocity + - foot_angular_velocity + task_name: hopper_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/hopper/vla/starvla-qwengr00t-h1/hopper_l0_gr00t_h1_5fps_2h100_20260917/dataset_statistics.json b/zero-latency/hopper/vla/starvla-qwengr00t-h1/hopper_l0_gr00t_h1_5fps_2h100_20260917/dataset_statistics.json new file mode 100644 index 0000000000000000000000000000000000000000..bb92bf6761b388f4009b3ee0beb94cee8054651c --- /dev/null +++ b/zero-latency/hopper/vla/starvla-qwengr00t-h1/hopper_l0_gr00t_h1_5fps_2h100_20260917/dataset_statistics.json @@ -0,0 +1,123 @@ +{ + "new_embodiment": { + "action": { + "mean": [ + 0.03865884989500046, + 0.5777319669723511, + -0.13737158477306366 + ], + "std": [ + 0.37739884853363037, + 0.689887523651123, + 0.6762200593948364 + ], + "max": [ + 1.0, + 1.0, + 1.0 + ], + "min": [ + -1.0, + -1.0, + -1.0 + ], + "q01": [ + -0.9043600207567215, + -1.0, + -1.0 + ], + "q99": [ + 1.0, + 1.0, + 1.0 + ], + "mask": [ + true, + true, + true + ] + }, + "state": { + "mean": [ + 0.40871402621269226, + -0.008348838426172733, + 0.45469287037849426, + 0.6798303723335266, + 0.269776850938797, + -0.1342218965291977, + 0.05999795347452164, + 0.3777448832988739, + -0.007698146626353264, + -0.002836690517142415, + 0.015370956622064114 + ], + "std": [ + 0.3633615970611572, + 0.37869560718536377, + 0.25348955392837524, + 0.30314165353775024, + 0.6127533316612244, + 0.21872857213020325, + 0.4510791301727295, + 0.21530857682228088, + 0.3457658588886261, + 0.4274502992630005, + 0.6040666103363037 + ], + "max": [ + 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 + ], + "q01": [ + -0.5281156021356582, + -0.9567392879724502, + -0.2079367846250534, + -0.12157819628715515, + -0.9323781633377075, + -0.7369779413938522, + -0.6852116698026657, + -0.10524752676486969, + -0.5511617350578308, + -1.0, + -1.0 + ], + "q99": [ + 0.9306531548500061, + 0.7862786006927489, + 0.9816559219360351, + 0.9127335548400879, + 0.9102470433712005, + 0.2971139311790463, + 0.8896300184726715, + 0.9198877382278442, + 1.0, + 0.981240049600601, + 1.0 + ] + }, + "num_transitions": 88212, + "num_trajectories": 90 + } +} \ No newline at end of file diff --git a/zero-latency/hopper/vla/starvla-qwengr00t-h1/hopper_l0_gr00t_h1_5fps_2h100_20260917/latency_prompt_map.json b/zero-latency/hopper/vla/starvla-qwengr00t-h1/hopper_l0_gr00t_h1_5fps_2h100_20260917/latency_prompt_map.json new file mode 100644 index 0000000000000000000000000000000000000000..f2b7129bb06cf782fa08ced52a646bb6c42dfaec --- /dev/null +++ b/zero-latency/hopper/vla/starvla-qwengr00t-h1/hopper_l0_gr00t_h1_5fps_2h100_20260917/latency_prompt_map.json @@ -0,0 +1,7 @@ +{ + "0": { + "prompt": "Move the Hopper robot forward while keeping its torso upright. Predict three continuous torques in [-1, 1] ordered as thigh, leg, and 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/hopper/vla/starvla-qwengr00t-h1/hopper_l0_gr00t_h1_5fps_2h100_20260917/manifest.json b/zero-latency/hopper/vla/starvla-qwengr00t-h1/hopper_l0_gr00t_h1_5fps_2h100_20260917/manifest.json new file mode 100644 index 0000000000000000000000000000000000000000..5577ffab211644cf2e06426bc0a731c1b7f4f26a --- /dev/null +++ b/zero-latency/hopper/vla/starvla-qwengr00t-h1/hopper_l0_gr00t_h1_5fps_2h100_20260917/manifest.json @@ -0,0 +1,142 @@ +{ + "dataset_name": "hopper_h1_5fps_l0", + "env_name": "hopper_rgb_state", + "episodes": 90, + "frames": 88212, + "task_prompts": [ + "Move the Hopper robot forward while keeping its torso upright. Predict three continuous torques in [-1, 1] ordered as thigh, leg, and 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/hopper/zero_latency/data/raw_5fps", + "integration_name": "gymnasium", + "task_name": "hopper_rgb_state", + "action_layout": "gymnasium_continuous_v1", + "action_labels": [ + "thigh_torque", + "leg_torque", + "foot_torque" + ], + "carrier_action_labels": [ + "thigh_torque", + "leg_torque", + "foot_torque" + ], + "action_dim": 3, + "active_action_dim": 3, + "action_carrier": "native", + "rows_unit": "decision_step", + "obs_fps": 5, + "obs_stride_raw_frames": 1, + "uses_state": true, + "state_dim": 11, + "state_labels": [ + "torso_height", + "torso_angle", + "thigh_angle", + "leg_angle", + "foot_angle", + "torso_x_velocity", + "torso_z_velocity", + "torso_angular_velocity", + "thigh_angular_velocity", + "leg_angular_velocity", + "foot_angular_velocity" + ], + "state_normalization": { + "type": "min_max", + "min": [ + 0.7007287740707397, + -0.19482052326202393, + -1.6461776494979858, + -1.9154568910598755, + -0.9631047248840332, + -0.1792079657316208, + -3.757478713989258, + -6.126420021057129, + -10.0, + -10.0, + -10.0 + ], + "max": [ + 1.6431776285171509, + 0.17731712758541107, + 0.059119515120983124, + 0.14385102689266205, + 0.9273068308830261, + 6.297187805175781, + 3.283433198928833, + 2.7663865089416504, + 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/hopper/zero_latency/data/lerobot/hopper_h1_5fps_l0/latency_prompt_map.json", + "custom_mixtures_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/hopper/zero_latency/data/lerobot/_generated_mixtures/hopper_h1_5fps_l0.json", + "gymnasium_task": { + "action_space": { + "dtype": "float32", + "high": [ + 1.0, + 1.0, + 1.0 + ], + "labels": [ + "thigh_torque", + "leg_torque", + "foot_torque" + ], + "low": [ + -1.0, + -1.0, + -1.0 + ], + "type": "box" + }, + "base_prompt": "Move the Hopper robot forward while keeping its torso upright. Predict three continuous torques in [-1, 1] ordered as thigh, leg, and foot.", + "env_fps": 5, + "env_id": "LatencyBench/HopperRgbState-v0", + "frame_stack": 1, + "make_kwargs": { + "base_env_id": "Hopper-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 + ], + "obs_fps": 5, + "registration_imports": [ + "latency_bench.envs.gymnasium_hopper_rgb_state" + ], + "state_space": { + "labels": [ + "torso_height", + "torso_angle", + "thigh_angle", + "leg_angle", + "foot_angle", + "torso_x_velocity", + "torso_z_velocity", + "torso_angular_velocity", + "thigh_angular_velocity", + "leg_angular_velocity", + "foot_angular_velocity" + ] + }, + "task_name": "hopper_rgb_state" + }, + "validation_dataset_name": "hopper_h1_5fps_l0__val", + "validation_episodes": 10, + "validation_frames": 9917 +} \ No newline at end of file diff --git a/zero-latency/hopper/vla/starvla-qwengr00t-h1/hopper_l0_gr00t_h1_5fps_2h100_20260917/provenance.json b/zero-latency/hopper/vla/starvla-qwengr00t-h1/hopper_l0_gr00t_h1_5fps_2h100_20260917/provenance.json new file mode 100644 index 0000000000000000000000000000000000000000..6d347bddfe7d00479265a86a3521843b5d0d7979 --- /dev/null +++ b/zero-latency/hopper/vla/starvla-qwengr00t-h1/hopper_l0_gr00t_h1_5fps_2h100_20260917/provenance.json @@ -0,0 +1,34 @@ +{ + "task": "hopper", + "model": "qwengr00t", + "training_condition": "zero-latency", + "training_run_id": "hopper_l0_gr00t_h1_5fps_2h100_20260917", + "source": { + "repo_id": "latency-sensitive-bench/Standard-Pipeline", + "revision": "2208875f92b21e3f8ec2363a30a6705dfe56cc04", + "path": "hopper/zero_latency/GR00T", + "url": "https://huggingface.co/latency-sensitive-bench/Standard-Pipeline/tree/2208875f92b21e3f8ec2363a30a6705dfe56cc04/hopper/zero_latency/GR00T" + }, + "hf_repo_id": "latency-sensitive-bench/benchmark-models", + "hf_path": "zero-latency/hopper/vla/starvla-qwengr00t-h1/hopper_l0_gr00t_h1_5fps_2h100_20260917", + "checkpoint": { + "source_file": "hopper/zero_latency/GR00T/checkpoints/model.pt", + "source_sha256": "5ec4a5fbdfb7fac819a51b523f3735c6702b90a2440a2957c5cc699941ffbf1e", + "source_bytes": 9976837923, + "file": "checkpoints/model.pt", + "selection_rule": "only published VLA checkpoint", + "method": "copy", + "sha256": "5ec4a5fbdfb7fac819a51b523f3735c6702b90a2440a2957c5cc699941ffbf1e", + "bytes": 9976837923 + }, + "config_source": "hopper/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/hopper/vla/starvla-qwengr00t-h1/hopper_l0_gr00t_h1_5fps_2h100_20260917/source/provenance.json b/zero-latency/hopper/vla/starvla-qwengr00t-h1/hopper_l0_gr00t_h1_5fps_2h100_20260917/source/provenance.json new file mode 100644 index 0000000000000000000000000000000000000000..eff12f43bcf6a47f9ea45f9dc0ab99c136a6cfc8 --- /dev/null +++ b/zero-latency/hopper/vla/starvla-qwengr00t-h1/hopper_l0_gr00t_h1_5fps_2h100_20260917/source/provenance.json @@ -0,0 +1,30 @@ +{ + "task": "hopper", + "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": "hopper_l0_gr00t_h1_5fps_2h100_20260917", + "condition": "zero_latency", + "source": { + "repo": "latency-sensitive-bench/Standard-Pipeline", + "revision": "571cb5801720488ed6458b9cfe3f4c595b069cea", + "prefix": "hopper/zero_latency/shared/source_100ep_v1/demonstrations/raw", + "sha256": { + "metadata.json": "6f6ba3dc2556aa2c16f81defde6231458cbe051d082ba7e054150ba1b65e39cd", + "train.parquet": "fcd0e418fbc33c531235c5765c1b5d3288dacd5b0a7bdcc6b288f1d4e1009b75", + "val.parquet": "1e53ba71d9efa4dd8cb8839a2e40206a9d682480c753e2bcdab8c3f417836e20" + }, + "code": { + "root": "1681edb8d9e4f7d10b7dfa647670e01e0d5cce37", + "sample_factory": "4b11d8754ffc989cdd15a340cfc9be0277377d5b", + "starvla": "1d0d7b139d1725cab268cb9dae2007ef4c5d05d2" + } + }, + "training_config_sha256": "93ef15ef2f2e8432cb8e81595c7bcce02c969cb961ed775b6c92054ba0581329", + "dataset_manifest_sha256": "4c45ebf784a7e606efc55711c65cf0525c0261d5d32a1963b02d23728a67a815" +} diff --git a/zero-latency/hopper/vla/starvla-qwengr00t-h1/hopper_l0_gr00t_h1_5fps_2h100_20260917/task_contract.json b/zero-latency/hopper/vla/starvla-qwengr00t-h1/hopper_l0_gr00t_h1_5fps_2h100_20260917/task_contract.json new file mode 100644 index 0000000000000000000000000000000000000000..9f46ce4391f55da768343fd4aca8cf9c2d035daf --- /dev/null +++ b/zero-latency/hopper/vla/starvla-qwengr00t-h1/hopper_l0_gr00t_h1_5fps_2h100_20260917/task_contract.json @@ -0,0 +1,62 @@ +{ + "action_space": { + "dtype": "float32", + "high": [ + 1.0, + 1.0, + 1.0 + ], + "labels": [ + "thigh_torque", + "leg_torque", + "foot_torque" + ], + "low": [ + -1.0, + -1.0, + -1.0 + ], + "type": "box" + }, + "base_prompt": "Move the Hopper robot forward while keeping its torso upright. Predict three continuous torques in [-1, 1] ordered as thigh, leg, and foot.", + "env_fps": 5, + "env_id": "LatencyBench/HopperRgbState-v0", + "frame_stack": 1, + "make_kwargs": { + "base_env_id": "Hopper-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 + ], + "obs_fps": 5, + "registration_imports": [ + "latency_bench.envs.gymnasium_hopper_rgb_state" + ], + "state_space": { + "labels": [ + "torso_height", + "torso_angle", + "thigh_angle", + "leg_angle", + "foot_angle", + "torso_x_velocity", + "torso_z_velocity", + "torso_angular_velocity", + "thigh_angular_velocity", + "leg_angular_velocity", + "foot_angular_velocity" + ] + }, + "task_name": "hopper_rgb_state" +} diff --git a/zero-latency/hopper/vla/starvla-qwengr00t-h1/hopper_l0_gr00t_h1_5fps_2h100_20260917/validation/completion.json b/zero-latency/hopper/vla/starvla-qwengr00t-h1/hopper_l0_gr00t_h1_5fps_2h100_20260917/validation/completion.json new file mode 100644 index 0000000000000000000000000000000000000000..fe543ea656547888881b1cde888f075b603bbf2c --- /dev/null +++ b/zero-latency/hopper/vla/starvla-qwengr00t-h1/hopper_l0_gr00t_h1_5fps_2h100_20260917/validation/completion.json @@ -0,0 +1,7 @@ +{ + "state": "L0_TRAIN_AND_SIM_EVAL_COMPLETED", + "completed_at": "2026-09-17T13:59:03.908633+00:00", + "checkpoint": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/hopper/zero_latency/vla/training/hopper_l0_gr00t_h1_5fps_2h100_20260917/checkpoints/steps_5000_pytorch_model.pt", + "checkpoint_sha256": "5ec4a5fbdfb7fac819a51b523f3735c6702b90a2440a2957c5cc699941ffbf1e", + "bundle": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/hopper/zero_latency/bundle" +} diff --git a/zero-latency/hopper/vla/starvla-qwengr00t-h1/hopper_l0_gr00t_h1_5fps_2h100_20260917/validation/l0_sim20_audit.json b/zero-latency/hopper/vla/starvla-qwengr00t-h1/hopper_l0_gr00t_h1_5fps_2h100_20260917/validation/l0_sim20_audit.json new file mode 100644 index 0000000000000000000000000000000000000000..b02ca7952a5c93dcfc73bb3612df8b9bf10248e1 --- /dev/null +++ b/zero-latency/hopper/vla/starvla-qwengr00t-h1/hopper_l0_gr00t_h1_5fps_2h100_20260917/validation/l0_sim20_audit.json @@ -0,0 +1,105 @@ +{ + "verified_at": "2026-09-17T14:13:17.029371+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 + ], + "returns": [ + 1282.1555676491873, + 366.05242016047794, + 913.858877998007, + 1191.3960892024, + 921.2143774371431, + 283.5313980536821, + 1459.5742400242764, + 1255.278374269173, + 277.90720736037014, + 935.1710716954052, + 1290.9800537752794, + 277.45689777280415, + 982.4008424617601, + 289.80142580106235, + 1220.3819055826225, + 1402.9969813054104, + 278.20307535073226, + 1815.2644508418452, + 709.778048448092, + 303.28291056031935 + ], + "lengths": [ + 346, + 165, + 258, + 335, + 261, + 126, + 400, + 344, + 124, + 265, + 354, + 124, + 286, + 128, + 338, + 385, + 124, + 511, + 228, + 133 + ], + "mean_return": 872.8343107875025, + "std_return_population": 479.65390591614806, + "mean_length": 261.75, + "raw_steps": 5235, + "actions": 5235, + "latency_records": 5235, + "env_fps": 5, + "obs_fps": 5, + "latency_mode": "simulated zero", + "latency_ms": 0, + "dropped_observations": 0, + "dropped_actions": 0, + "invalid_actions": 0, + "logged_preclip_action_out_of_bounds_steps": 3704, + "logged_preclip_action_range": [ + -1.1191151142120361, + 1.1177167892456055 + ], + "action_clipping_evidence": "HopperEnv.step clips to env.action_space bounds before env.step; GymnasiumEnvAdapter.step overwrites applied_action with preclip action.value, so postclip physical actions are not directly logged", + "action_clipping_source_sha256": "616473b67fa33c759bed07e28d4736613b6f4600674da966d754dd18c27669f7", + "action_logging_source_sha256": "e5351a4d70520350c8bcbdcaf958ab57d8eca83e251adfd6f520ac34e40254f5", + "raw_step_time_reward_consistency": true, + "bundle_files": 8, + "bundle_sha256": { + "checkpoints/model.pt": "5ec4a5fbdfb7fac819a51b523f3735c6702b90a2440a2957c5cc699941ffbf1e", + "config.full.yaml": "ea0131fb531e70f404a252344589dd8f71c56f3d897699a20df32dbc2a80a371", + "config.yaml": "f40ecb9bb5f1450edd45f2bd473b3e6fc9ac65f91154629d3f80ea58673cb4c0", + "dataset_statistics.json": "bb23771f87f039a7fa41a682933aac9099f45c4c26b103fe03a51cb84b82c2c6", + "latency_prompt_map.json": "9fd22cc880afbd45589baa0db9ca0e237c3f89b4811156f6ed964600f82b186e", + "manifest.json": "4c45ebf784a7e606efc55711c65cf0525c0261d5d32a1963b02d23728a67a815", + "provenance.json": "835befe92adb1b99cefe0534d90dd93484f79870bf1f4e8cea6aa27047ef74f3" + }, + "checkpoint_steps": 5000, + "reload_evidence": "existing successful L0sim20 executed this final bundle; no repeated evaluation", + "acceptance": "L0evaluationcomplete; downstream3090P/Bandprofilepipelinepending" +} diff --git a/zero-latency/hopper/vla/starvla-qwengr00t-h1/hopper_l0_gr00t_h1_5fps_2h100_20260917/validation/loader_check.json b/zero-latency/hopper/vla/starvla-qwengr00t-h1/hopper_l0_gr00t_h1_5fps_2h100_20260917/validation/loader_check.json new file mode 100644 index 0000000000000000000000000000000000000000..bdbef9670f06baf7cb664e5aeebd07b688738ad4 --- /dev/null +++ b/zero-latency/hopper/vla/starvla-qwengr00t-h1/hopper_l0_gr00t_h1_5fps_2h100_20260917/validation/loader_check.json @@ -0,0 +1,22 @@ +{ + "loader_action_shape": [ + 1, + 3 + ], + "loader_state_shape": [ + 1, + 11 + ], + "native_action_roundtrip": true, + "state_minmax_once": true, + "image_pixels_identical": true, + "robot_type": "rl_games_gymnasium", + "sample_action": [ + [ + -1.0, + 1.0, + 1.0 + ] + ], + "native_decoder": "box identity slice; unnorm_key unused" +} diff --git a/zero-latency/hopper/vla/starvla-qwengr00t-h1/hopper_l0_gr00t_h1_5fps_2h100_20260917/validation/preparation.json b/zero-latency/hopper/vla/starvla-qwengr00t-h1/hopper_l0_gr00t_h1_5fps_2h100_20260917/validation/preparation.json new file mode 100644 index 0000000000000000000000000000000000000000..cfdfb29893de5bdf98e4b08ed562d4bcec211f79 --- /dev/null +++ b/zero-latency/hopper/vla/starvla-qwengr00t-h1/hopper_l0_gr00t_h1_5fps_2h100_20260917/validation/preparation.json @@ -0,0 +1,46 @@ +{ + "repo": "latency-sensitive-bench/Standard-Pipeline", + "revision": "571cb5801720488ed6458b9cfe3f4c595b069cea", + "prefix": "hopper/zero_latency/shared/source_100ep_v1/demonstrations/raw", + "sha256": { + "metadata.json": "6f6ba3dc2556aa2c16f81defde6231458cbe051d082ba7e054150ba1b65e39cd", + "train.parquet": "fcd0e418fbc33c531235c5765c1b5d3288dacd5b0a7bdcc6b288f1d4e1009b75", + "val.parquet": "1e53ba71d9efa4dd8cb8839a2e40206a9d682480c753e2bcdab8c3f417836e20" + }, + "code": { + "root": "1681edb8d9e4f7d10b7dfa647670e01e0d5cce37", + "sample_factory": "4b11d8754ffc989cdd15a340cfc9be0277377d5b", + "starvla": "1d0d7b139d1725cab268cb9dae2007ef4c5d05d2" + }, + "frames": { + "train": 88212, + "val": 9917 + }, + "episodes": { + "train": 90, + "val": 10 + }, + "return_ranges": { + "train": [ + 3036.393681228161, + 3711.7399080991745 + ], + "val": [ + 3405.0092537999153, + 3713.225490093231 + ] + }, + "image_size": [ + 224, + 224 + ], + "state_dim": 11, + "action_dim": 3, + "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/hopper/zero_latency/train.yaml", + "dataset": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/hopper/zero_latency/data/lerobot/hopper_h1_5fps_l0" +} diff --git a/zero-latency/hopper/vla/starvla-qwengr00t-h1/hopper_l0_gr00t_h1_5fps_2h100_20260917/vla/evaluation/l0_sim20.yaml b/zero-latency/hopper/vla/starvla-qwengr00t-h1/hopper_l0_gr00t_h1_5fps_2h100_20260917/vla/evaluation/l0_sim20.yaml new file mode 100644 index 0000000000000000000000000000000000000000..fbbb7fad47c9a38e538daa82a70ad7ceefd0b579 --- /dev/null +++ b/zero-latency/hopper/vla/starvla-qwengr00t-h1/hopper_l0_gr00t_h1_5fps_2h100_20260917/vla/evaluation/l0_sim20.yaml @@ -0,0 +1,157 @@ +env: + state_space: + labels: + - torso_height + - torso_angle + - thigh_angle + - leg_angle + - foot_angle + - torso_x_velocity + - torso_z_velocity + - torso_angular_velocity + - thigh_angular_velocity + - leg_angular_velocity + - foot_angular_velocity + name: gymnasium + task_name: hopper + env_id: LatencyBench/Hopper-v0 + registration_imports: + - latency_bench.envs.gymnasium_hopper + make_kwargs: + base_env_id: Hopper-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 + labels: + - thigh_torque + - leg_torque + - foot_torque + low: + - -1.0 + - -1.0 + - -1.0 + type: box + noop_action: + - 0.0 + - 0.0 + - 0.0 + base_prompt: Move the Hopper robot forward while keeping its torso upright. Predict + three continuous torques in [-1, 1] ordered as thigh, leg, and foot. + obs_resize: + - 224 + - 224 +experiment: + name: gr00t_hopper_zero + seed: 42 +backend: + type: sample_factory + algo: APPO + device: cuda + train_dir: /mnt/checkpoints/latency-sensitive-bench/small_models/hopper + 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 +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/hopper/zero_latency/bundle/checkpoints/model.pt + model_config_path: /mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/hopper/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/hopper/zero_latency/bundle/manifest.json + latency_prompt_map_path: /mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/hopper/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 + num_batches_to_accumulate: 2 + policy_workers_per_policy: 1 + max_policy_lag: 10000 + learning_rate: 0.00295 + lr_schedule: linear_decay + lr_schedule_kl_threshold: 0.008 + gamma: 0.99 + gae_lambda: 0.95 + ppo_clip_ratio: 0.2 + ppo_clip_value: 1.0 + value_loss_coeff: 1.3 + max_grad_norm: 3.5 + exploration_loss: entropy + exploration_loss_coeff: 0.0 + kl_loss_coeff: 0.1 + reward_scale: 1.0 + reward_clip: 1000.0 + async_rl: false + serial_mode: false + batched_sampling: false + with_vtrace: false + use_rnn: false + env_framestack: 4 + encoder_mlp_layers: + - 64 + - 64 + nonlinearity: tanh + adaptive_stddev: false + policy_initialization: torch_default + initial_stddev: 1.0 + actor_critic_share_weights: true + shuffle_minibatches: false + value_bootstrap: false + normalize_input: true + normalize_returns: true + decorrelate_experience_max_seconds: 10 + decorrelate_envs_on_one_worker: true + set_workers_cpu_affinity: true + force_envs_single_thread: true + save_every_sec: 600 + keep_checkpoints: 3 + save_best_every_sec: 60 + save_best_after: 100000 +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/hopper/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/hopper/vla/starvla-qwengr00t-h1/hopper_l0_gr00t_h1_5fps_2h100_20260917/vla/evaluation/l0_sim20/episode_metrics.jsonl b/zero-latency/hopper/vla/starvla-qwengr00t-h1/hopper_l0_gr00t_h1_5fps_2h100_20260917/vla/evaluation/l0_sim20/episode_metrics.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..a5cce8ba010bd31648541fbfb783ba8df56a8dc7 --- /dev/null +++ b/zero-latency/hopper/vla/starvla-qwengr00t-h1/hopper_l0_gr00t_h1_5fps_2h100_20260917/vla/evaluation/l0_sim20/episode_metrics.jsonl @@ -0,0 +1,20 @@ +{"episode_id": 0, "episode_return": 1282.1555676491873, "episode_return_env": 1282.1555676491873, "game_score": null, "mean_latency_ms": 0.0, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/hopper/zero_latency/bundle/checkpoints/model.pt", "config_name": "gr00t_hopper_zero", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Hopper-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/hopper/zero_latency/l0_sim20", "policy_id": "starvla", "profile_ref": null, "run_name": "gr00t_hopper_zero", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_run_id": null, "submitted_observation_frames": 346, "workload_id": null}, "num_actions": 346, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 0.0, "p99_latency_ms": 0.0, "return_raw": null, "survival_steps": 346} +{"episode_id": 1, "episode_return": 366.05242016047794, "episode_return_env": 366.05242016047794, "game_score": null, "mean_latency_ms": 0.0, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/hopper/zero_latency/bundle/checkpoints/model.pt", "config_name": "gr00t_hopper_zero", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Hopper-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/hopper/zero_latency/l0_sim20", "policy_id": "starvla", "profile_ref": null, "run_name": "gr00t_hopper_zero", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_run_id": null, "submitted_observation_frames": 165, "workload_id": null}, "num_actions": 165, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 0.0, "p99_latency_ms": 0.0, "return_raw": null, "survival_steps": 165} +{"episode_id": 2, "episode_return": 913.858877998007, "episode_return_env": 913.858877998007, "game_score": null, "mean_latency_ms": 0.0, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/hopper/zero_latency/bundle/checkpoints/model.pt", "config_name": "gr00t_hopper_zero", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Hopper-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/hopper/zero_latency/l0_sim20", "policy_id": "starvla", "profile_ref": null, "run_name": "gr00t_hopper_zero", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_run_id": null, "submitted_observation_frames": 258, "workload_id": null}, "num_actions": 258, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 0.0, "p99_latency_ms": 0.0, "return_raw": null, "survival_steps": 258} +{"episode_id": 3, "episode_return": 1191.3960892024, "episode_return_env": 1191.3960892024, "game_score": null, "mean_latency_ms": 0.0, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/hopper/zero_latency/bundle/checkpoints/model.pt", "config_name": "gr00t_hopper_zero", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Hopper-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/hopper/zero_latency/l0_sim20", "policy_id": "starvla", "profile_ref": null, "run_name": "gr00t_hopper_zero", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_run_id": null, "submitted_observation_frames": 335, "workload_id": null}, "num_actions": 335, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 0.0, "p99_latency_ms": 0.0, "return_raw": null, "survival_steps": 335} +{"episode_id": 4, "episode_return": 921.2143774371431, "episode_return_env": 921.2143774371431, "game_score": null, "mean_latency_ms": 0.0, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/hopper/zero_latency/bundle/checkpoints/model.pt", "config_name": "gr00t_hopper_zero", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Hopper-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/hopper/zero_latency/l0_sim20", "policy_id": "starvla", "profile_ref": null, "run_name": "gr00t_hopper_zero", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_run_id": null, "submitted_observation_frames": 261, "workload_id": null}, "num_actions": 261, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 0.0, "p99_latency_ms": 0.0, "return_raw": null, "survival_steps": 261} +{"episode_id": 5, "episode_return": 283.5313980536821, "episode_return_env": 283.5313980536821, "game_score": null, "mean_latency_ms": 0.0, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/hopper/zero_latency/bundle/checkpoints/model.pt", "config_name": "gr00t_hopper_zero", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Hopper-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/hopper/zero_latency/l0_sim20", "policy_id": "starvla", "profile_ref": null, "run_name": "gr00t_hopper_zero", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_run_id": null, "submitted_observation_frames": 126, "workload_id": null}, "num_actions": 126, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 0.0, "p99_latency_ms": 0.0, "return_raw": null, "survival_steps": 126} +{"episode_id": 6, "episode_return": 1459.5742400242764, "episode_return_env": 1459.5742400242764, "game_score": null, "mean_latency_ms": 0.0, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/hopper/zero_latency/bundle/checkpoints/model.pt", "config_name": "gr00t_hopper_zero", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Hopper-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/hopper/zero_latency/l0_sim20", "policy_id": "starvla", "profile_ref": null, "run_name": "gr00t_hopper_zero", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_run_id": null, "submitted_observation_frames": 400, "workload_id": null}, "num_actions": 400, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 0.0, "p99_latency_ms": 0.0, "return_raw": null, "survival_steps": 400} +{"episode_id": 7, "episode_return": 1255.278374269173, "episode_return_env": 1255.278374269173, "game_score": null, "mean_latency_ms": 0.0, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/hopper/zero_latency/bundle/checkpoints/model.pt", "config_name": "gr00t_hopper_zero", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Hopper-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/hopper/zero_latency/l0_sim20", "policy_id": "starvla", "profile_ref": null, "run_name": "gr00t_hopper_zero", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_run_id": null, "submitted_observation_frames": 344, "workload_id": null}, "num_actions": 344, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 0.0, "p99_latency_ms": 0.0, "return_raw": null, "survival_steps": 344} +{"episode_id": 8, "episode_return": 277.90720736037014, "episode_return_env": 277.90720736037014, "game_score": null, "mean_latency_ms": 0.0, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/hopper/zero_latency/bundle/checkpoints/model.pt", "config_name": "gr00t_hopper_zero", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Hopper-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/hopper/zero_latency/l0_sim20", "policy_id": "starvla", "profile_ref": null, "run_name": "gr00t_hopper_zero", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_run_id": null, "submitted_observation_frames": 124, "workload_id": null}, "num_actions": 124, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 0.0, "p99_latency_ms": 0.0, "return_raw": null, "survival_steps": 124} +{"episode_id": 9, "episode_return": 935.1710716954052, "episode_return_env": 935.1710716954052, "game_score": null, "mean_latency_ms": 0.0, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/hopper/zero_latency/bundle/checkpoints/model.pt", "config_name": "gr00t_hopper_zero", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Hopper-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/hopper/zero_latency/l0_sim20", "policy_id": "starvla", "profile_ref": null, "run_name": "gr00t_hopper_zero", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_run_id": null, "submitted_observation_frames": 265, "workload_id": null}, "num_actions": 265, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 0.0, "p99_latency_ms": 0.0, "return_raw": null, "survival_steps": 265} +{"episode_id": 10, "episode_return": 1290.9800537752794, "episode_return_env": 1290.9800537752794, "game_score": null, "mean_latency_ms": 0.0, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/hopper/zero_latency/bundle/checkpoints/model.pt", "config_name": "gr00t_hopper_zero", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Hopper-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/hopper/zero_latency/l0_sim20", "policy_id": "starvla", "profile_ref": null, "run_name": "gr00t_hopper_zero", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_run_id": null, "submitted_observation_frames": 354, "workload_id": null}, "num_actions": 354, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 0.0, "p99_latency_ms": 0.0, "return_raw": null, "survival_steps": 354} +{"episode_id": 11, "episode_return": 277.45689777280415, "episode_return_env": 277.45689777280415, "game_score": null, "mean_latency_ms": 0.0, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/hopper/zero_latency/bundle/checkpoints/model.pt", "config_name": "gr00t_hopper_zero", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Hopper-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/hopper/zero_latency/l0_sim20", "policy_id": "starvla", "profile_ref": null, "run_name": "gr00t_hopper_zero", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_run_id": null, "submitted_observation_frames": 124, "workload_id": null}, "num_actions": 124, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 0.0, "p99_latency_ms": 0.0, "return_raw": null, "survival_steps": 124} +{"episode_id": 12, "episode_return": 982.4008424617601, "episode_return_env": 982.4008424617601, "game_score": null, "mean_latency_ms": 0.0, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/hopper/zero_latency/bundle/checkpoints/model.pt", "config_name": "gr00t_hopper_zero", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Hopper-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/hopper/zero_latency/l0_sim20", "policy_id": "starvla", "profile_ref": null, "run_name": "gr00t_hopper_zero", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_run_id": null, "submitted_observation_frames": 286, "workload_id": null}, "num_actions": 286, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 0.0, "p99_latency_ms": 0.0, "return_raw": null, "survival_steps": 286} +{"episode_id": 13, "episode_return": 289.80142580106235, "episode_return_env": 289.80142580106235, "game_score": null, "mean_latency_ms": 0.0, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/hopper/zero_latency/bundle/checkpoints/model.pt", "config_name": "gr00t_hopper_zero", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Hopper-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/hopper/zero_latency/l0_sim20", "policy_id": "starvla", "profile_ref": null, "run_name": "gr00t_hopper_zero", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_run_id": null, "submitted_observation_frames": 128, "workload_id": null}, "num_actions": 128, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 0.0, "p99_latency_ms": 0.0, "return_raw": null, "survival_steps": 128} +{"episode_id": 14, "episode_return": 1220.3819055826225, "episode_return_env": 1220.3819055826225, "game_score": null, "mean_latency_ms": 0.0, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/hopper/zero_latency/bundle/checkpoints/model.pt", "config_name": "gr00t_hopper_zero", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Hopper-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/hopper/zero_latency/l0_sim20", "policy_id": "starvla", "profile_ref": null, "run_name": "gr00t_hopper_zero", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_run_id": null, "submitted_observation_frames": 338, "workload_id": null}, "num_actions": 338, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 0.0, "p99_latency_ms": 0.0, "return_raw": null, "survival_steps": 338} +{"episode_id": 15, "episode_return": 1402.9969813054104, "episode_return_env": 1402.9969813054104, "game_score": null, "mean_latency_ms": 0.0, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/hopper/zero_latency/bundle/checkpoints/model.pt", "config_name": "gr00t_hopper_zero", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Hopper-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/hopper/zero_latency/l0_sim20", "policy_id": "starvla", "profile_ref": null, "run_name": "gr00t_hopper_zero", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_run_id": null, "submitted_observation_frames": 385, "workload_id": null}, "num_actions": 385, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 0.0, "p99_latency_ms": 0.0, "return_raw": null, "survival_steps": 385} +{"episode_id": 16, "episode_return": 278.20307535073226, "episode_return_env": 278.20307535073226, "game_score": null, "mean_latency_ms": 0.0, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/hopper/zero_latency/bundle/checkpoints/model.pt", "config_name": "gr00t_hopper_zero", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Hopper-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/hopper/zero_latency/l0_sim20", "policy_id": "starvla", "profile_ref": null, "run_name": "gr00t_hopper_zero", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_run_id": null, "submitted_observation_frames": 124, "workload_id": null}, "num_actions": 124, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 0.0, "p99_latency_ms": 0.0, "return_raw": null, "survival_steps": 124} +{"episode_id": 17, "episode_return": 1815.2644508418452, "episode_return_env": 1815.2644508418452, "game_score": null, "mean_latency_ms": 0.0, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/hopper/zero_latency/bundle/checkpoints/model.pt", "config_name": "gr00t_hopper_zero", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Hopper-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/hopper/zero_latency/l0_sim20", "policy_id": "starvla", "profile_ref": null, "run_name": "gr00t_hopper_zero", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_run_id": null, "submitted_observation_frames": 511, "workload_id": null}, "num_actions": 511, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 0.0, "p99_latency_ms": 0.0, "return_raw": null, "survival_steps": 511} +{"episode_id": 18, "episode_return": 709.778048448092, "episode_return_env": 709.778048448092, "game_score": null, "mean_latency_ms": 0.0, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/hopper/zero_latency/bundle/checkpoints/model.pt", "config_name": "gr00t_hopper_zero", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Hopper-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/hopper/zero_latency/l0_sim20", "policy_id": "starvla", "profile_ref": null, "run_name": "gr00t_hopper_zero", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_run_id": null, "submitted_observation_frames": 228, "workload_id": null}, "num_actions": 228, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 0.0, "p99_latency_ms": 0.0, "return_raw": null, "survival_steps": 228} +{"episode_id": 19, "episode_return": 303.28291056031935, "episode_return_env": 303.28291056031935, "game_score": null, "mean_latency_ms": 0.0, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/hopper/zero_latency/bundle/checkpoints/model.pt", "config_name": "gr00t_hopper_zero", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/Hopper-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/hopper/zero_latency/l0_sim20", "policy_id": "starvla", "profile_ref": null, "run_name": "gr00t_hopper_zero", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_run_id": null, "submitted_observation_frames": 133, "workload_id": null}, "num_actions": 133, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 0.0, "p99_latency_ms": 0.0, "return_raw": null, "survival_steps": 133} diff --git a/zero-latency/hopper/vla/starvla-qwengr00t-h1/hopper_l0_gr00t_h1_5fps_2h100_20260917/vla/evaluation/l0_sim20/queue_eval_results.jsonl b/zero-latency/hopper/vla/starvla-qwengr00t-h1/hopper_l0_gr00t_h1_5fps_2h100_20260917/vla/evaluation/l0_sim20/queue_eval_results.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..dcffb5f64b6022946c2de5e06c0b05b3b1e0104c --- /dev/null +++ b/zero-latency/hopper/vla/starvla-qwengr00t-h1/hopper_l0_gr00t_h1_5fps_2h100_20260917/vla/evaluation/l0_sim20/queue_eval_results.jsonl @@ -0,0 +1 @@ +{"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/hopper/zero_latency/bundle/checkpoints/model.pt", "experiment_name": "gr00t_hopper_zero", "latency": 0, "latency_type": "zero", "lengths": [346, 165, 258, 335, 261, 126, 400, 344, 124, 265, 354, 124, 286, 128, 338, 385, 124, 511, 228, 133], "mean_length": 261.75, "mean_return": 872.8343107875025, "returns": [1282.1555676491873, 366.05242016047794, 913.858877998007, 1191.3960892024, 921.2143774371431, 283.5313980536821, 1459.5742400242764, 1255.278374269173, 277.90720736037014, 935.1710716954052, 1290.9800537752794, 277.45689777280415, 982.4008424617601, 289.80142580106235, 1220.3819055826225, 1402.9969813054104, 278.20307535073226, 1815.2644508418452, 709.778048448092, 303.28291056031935], "seed": 42, "std_return": 479.6539059161481, "suite_name": "fixed_0", "timestamp_utc": "2026-09-17T13:58:55.254682+00:00"} diff --git a/zero-latency/hopper/vla/starvla-qwenoft-h1/hopper_rgb_state_l0_return_gt3000_100ep_openvla_native_continuous_projector_sft_5k/README.md b/zero-latency/hopper/vla/starvla-qwenoft-h1/hopper_rgb_state_l0_return_gt3000_100ep_openvla_native_continuous_projector_sft_5k/README.md new file mode 100644 index 0000000000000000000000000000000000000000..c816aaf257a2251889a55fc28f8bff3de4aab487 --- /dev/null +++ b/zero-latency/hopper/vla/starvla-qwenoft-h1/hopper_rgb_state_l0_return_gt3000_100ep_openvla_native_continuous_projector_sft_5k/README.md @@ -0,0 +1,30 @@ +# hopper / qwenoft + +Training condition: `zero-latency`. Run: `hopper_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/hopper/zero_latency/OpenVLA) at `2208875f92b21e3f8ec2363a30a6705dfe56cc04`. + +- Checkpoint: `checkpoints/model.pt` +- Selection: only published VLA checkpoint +- Checkpoint SHA256: `87e4cb2bf569982a181bb7a7642a64499012542ab2cbf1545cd1c654b0a3d7cd` +- 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/hopper/vla/starvla-qwenoft-h1/hopper_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/hopper/vla/starvla-qwenoft-h1/hopper_rgb_state_l0_return_gt3000_100ep_openvla_native_continuous_projector_sft_5k/checkpoints/model.pt b/zero-latency/hopper/vla/starvla-qwenoft-h1/hopper_rgb_state_l0_return_gt3000_100ep_openvla_native_continuous_projector_sft_5k/checkpoints/model.pt new file mode 100644 index 0000000000000000000000000000000000000000..f3d6039620f4444e8b48f42855f8fdace7e73076 --- /dev/null +++ b/zero-latency/hopper/vla/starvla-qwenoft-h1/hopper_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 sha256:87e4cb2bf569982a181bb7a7642a64499012542ab2cbf1545cd1c654b0a3d7cd +size 9785081249 diff --git a/zero-latency/hopper/vla/starvla-qwenoft-h1/hopper_rgb_state_l0_return_gt3000_100ep_openvla_native_continuous_projector_sft_5k/config.full.yaml b/zero-latency/hopper/vla/starvla-qwenoft-h1/hopper_rgb_state_l0_return_gt3000_100ep_openvla_native_continuous_projector_sft_5k/config.full.yaml new file mode 100644 index 0000000000000000000000000000000000000000..a66c0d632f0646e1594c321b649f75ab4e0002a4 --- /dev/null +++ b/zero-latency/hopper/vla/starvla-qwenoft-h1/hopper_rgb_state_l0_return_gt3000_100ep_openvla_native_continuous_projector_sft_5k/config.full.yaml @@ -0,0 +1,356 @@ +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: 3 + action_hidden_dim: 2560 + future_action_window_size: 0 + past_action_window_size: 0 + loss_type: l1 + state_encoding: continuous_projector + state_dim: 11 + action_horizon: 1 + action_env_dim: 3 + 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: hopper_rgb_state_l0_return_gt3000_100ep + eval_data_mix: hopper_rgb_state_l0_return_gt3000_100ep__val + custom_mixtures_path: /mnt/data/latency-sensitive-bench/vla_datasets/rl_games/_generated_mixtures/hopper_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 + labels: + - thigh_torque + - leg_torque + - foot_torque + low: + - -1.0 + - -1.0 + - -1.0 + type: box + base_prompt: Move the Hopper robot forward while keeping its torso upright. + Predict three continuous torques in [-1, 1] ordered as thigh, leg, and foot. + env_fps: 125.0 + env_id: LatencyBench/HopperRgbState-v0 + frame_stack: 1 + make_kwargs: + base_env_id: Hopper-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 + obs_fps: 125.0 + registration_imports: + - latency_bench.envs.gymnasium_hopper_rgb_state + state_space: + labels: + - torso_height + - torso_angle + - thigh_angle + - leg_angle + - foot_angle + - torso_x_velocity + - torso_z_velocity + - torso_angular_velocity + - thigh_angular_velocity + - leg_angular_velocity + - foot_angular_velocity + task_name: hopper_rgb_state + active_action_dim: 3 +dataset: + source_hf: '' + config_name: null + source_subdir: null + converted_name: hopper_rgb_state_l0_return_gt3000_100ep + single_source_hf: '' + mixed_source_hf: '' + single_converted_name: hopper_rgb_state_l0_return_gt3000_100ep + mixed_converted_name: hopper_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 + labels: + - thigh_torque + - leg_torque + - foot_torque + low: + - -1.0 + - -1.0 + - -1.0 + type: box + base_prompt: Move the Hopper robot forward while keeping its torso upright. + Predict three continuous torques in [-1, 1] ordered as thigh, leg, and foot. + env_fps: 125.0 + env_id: LatencyBench/HopperRgbState-v0 + frame_stack: 1 + make_kwargs: + base_env_id: Hopper-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 + obs_fps: 125.0 + registration_imports: + - latency_bench.envs.gymnasium_hopper_rgb_state + state_space: + labels: + - torso_height + - torso_angle + - thigh_angle + - leg_angle + - foot_angle + - torso_x_velocity + - torso_z_velocity + - torso_angular_velocity + - thigh_angular_velocity + - leg_angular_velocity + - foot_angular_velocity + task_name: hopper_rgb_state + task_name: hopper_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: hopper_rgb_state_l0_return_gt3000_100ep_openvla_native_continuous_projector_sft_5k +output_dir: /mnt/checkpoints/latency-sensitive-bench/openvla_runs/hopper_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/hopper/vla/starvla-qwenoft-h1/hopper_rgb_state_l0_return_gt3000_100ep_openvla_native_continuous_projector_sft_5k/config.yaml b/zero-latency/hopper/vla/starvla-qwenoft-h1/hopper_rgb_state_l0_return_gt3000_100ep_openvla_native_continuous_projector_sft_5k/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..a66c0d632f0646e1594c321b649f75ab4e0002a4 --- /dev/null +++ b/zero-latency/hopper/vla/starvla-qwenoft-h1/hopper_rgb_state_l0_return_gt3000_100ep_openvla_native_continuous_projector_sft_5k/config.yaml @@ -0,0 +1,356 @@ +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: 3 + action_hidden_dim: 2560 + future_action_window_size: 0 + past_action_window_size: 0 + loss_type: l1 + state_encoding: continuous_projector + state_dim: 11 + action_horizon: 1 + action_env_dim: 3 + 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: hopper_rgb_state_l0_return_gt3000_100ep + eval_data_mix: hopper_rgb_state_l0_return_gt3000_100ep__val + custom_mixtures_path: /mnt/data/latency-sensitive-bench/vla_datasets/rl_games/_generated_mixtures/hopper_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 + labels: + - thigh_torque + - leg_torque + - foot_torque + low: + - -1.0 + - -1.0 + - -1.0 + type: box + base_prompt: Move the Hopper robot forward while keeping its torso upright. + Predict three continuous torques in [-1, 1] ordered as thigh, leg, and foot. + env_fps: 125.0 + env_id: LatencyBench/HopperRgbState-v0 + frame_stack: 1 + make_kwargs: + base_env_id: Hopper-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 + obs_fps: 125.0 + registration_imports: + - latency_bench.envs.gymnasium_hopper_rgb_state + state_space: + labels: + - torso_height + - torso_angle + - thigh_angle + - leg_angle + - foot_angle + - torso_x_velocity + - torso_z_velocity + - torso_angular_velocity + - thigh_angular_velocity + - leg_angular_velocity + - foot_angular_velocity + task_name: hopper_rgb_state + active_action_dim: 3 +dataset: + source_hf: '' + config_name: null + source_subdir: null + converted_name: hopper_rgb_state_l0_return_gt3000_100ep + single_source_hf: '' + mixed_source_hf: '' + single_converted_name: hopper_rgb_state_l0_return_gt3000_100ep + mixed_converted_name: hopper_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 + labels: + - thigh_torque + - leg_torque + - foot_torque + low: + - -1.0 + - -1.0 + - -1.0 + type: box + base_prompt: Move the Hopper robot forward while keeping its torso upright. + Predict three continuous torques in [-1, 1] ordered as thigh, leg, and foot. + env_fps: 125.0 + env_id: LatencyBench/HopperRgbState-v0 + frame_stack: 1 + make_kwargs: + base_env_id: Hopper-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 + obs_fps: 125.0 + registration_imports: + - latency_bench.envs.gymnasium_hopper_rgb_state + state_space: + labels: + - torso_height + - torso_angle + - thigh_angle + - leg_angle + - foot_angle + - torso_x_velocity + - torso_z_velocity + - torso_angular_velocity + - thigh_angular_velocity + - leg_angular_velocity + - foot_angular_velocity + task_name: hopper_rgb_state + task_name: hopper_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: hopper_rgb_state_l0_return_gt3000_100ep_openvla_native_continuous_projector_sft_5k +output_dir: /mnt/checkpoints/latency-sensitive-bench/openvla_runs/hopper_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/hopper/vla/starvla-qwenoft-h1/hopper_rgb_state_l0_return_gt3000_100ep_openvla_native_continuous_projector_sft_5k/dataset_statistics.json b/zero-latency/hopper/vla/starvla-qwenoft-h1/hopper_rgb_state_l0_return_gt3000_100ep_openvla_native_continuous_projector_sft_5k/dataset_statistics.json new file mode 100644 index 0000000000000000000000000000000000000000..bb92bf6761b388f4009b3ee0beb94cee8054651c --- /dev/null +++ b/zero-latency/hopper/vla/starvla-qwenoft-h1/hopper_rgb_state_l0_return_gt3000_100ep_openvla_native_continuous_projector_sft_5k/dataset_statistics.json @@ -0,0 +1,123 @@ +{ + "new_embodiment": { + "action": { + "mean": [ + 0.03865884989500046, + 0.5777319669723511, + -0.13737158477306366 + ], + "std": [ + 0.37739884853363037, + 0.689887523651123, + 0.6762200593948364 + ], + "max": [ + 1.0, + 1.0, + 1.0 + ], + "min": [ + -1.0, + -1.0, + -1.0 + ], + "q01": [ + -0.9043600207567215, + -1.0, + -1.0 + ], + "q99": [ + 1.0, + 1.0, + 1.0 + ], + "mask": [ + true, + true, + true + ] + }, + "state": { + "mean": [ + 0.40871402621269226, + -0.008348838426172733, + 0.45469287037849426, + 0.6798303723335266, + 0.269776850938797, + -0.1342218965291977, + 0.05999795347452164, + 0.3777448832988739, + -0.007698146626353264, + -0.002836690517142415, + 0.015370956622064114 + ], + "std": [ + 0.3633615970611572, + 0.37869560718536377, + 0.25348955392837524, + 0.30314165353775024, + 0.6127533316612244, + 0.21872857213020325, + 0.4510791301727295, + 0.21530857682228088, + 0.3457658588886261, + 0.4274502992630005, + 0.6040666103363037 + ], + "max": [ + 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 + ], + "q01": [ + -0.5281156021356582, + -0.9567392879724502, + -0.2079367846250534, + -0.12157819628715515, + -0.9323781633377075, + -0.7369779413938522, + -0.6852116698026657, + -0.10524752676486969, + -0.5511617350578308, + -1.0, + -1.0 + ], + "q99": [ + 0.9306531548500061, + 0.7862786006927489, + 0.9816559219360351, + 0.9127335548400879, + 0.9102470433712005, + 0.2971139311790463, + 0.8896300184726715, + 0.9198877382278442, + 1.0, + 0.981240049600601, + 1.0 + ] + }, + "num_transitions": 88212, + "num_trajectories": 90 + } +} \ No newline at end of file diff --git a/zero-latency/hopper/vla/starvla-qwenoft-h1/hopper_rgb_state_l0_return_gt3000_100ep_openvla_native_continuous_projector_sft_5k/dataset_statistics_eval.json b/zero-latency/hopper/vla/starvla-qwenoft-h1/hopper_rgb_state_l0_return_gt3000_100ep_openvla_native_continuous_projector_sft_5k/dataset_statistics_eval.json new file mode 100644 index 0000000000000000000000000000000000000000..3c867e844917ae2dbab65bb030f171c834348353 --- /dev/null +++ b/zero-latency/hopper/vla/starvla-qwenoft-h1/hopper_rgb_state_l0_return_gt3000_100ep_openvla_native_continuous_projector_sft_5k/dataset_statistics_eval.json @@ -0,0 +1,123 @@ +{ + "new_embodiment": { + "action": { + "mean": [ + 0.038164444267749786, + 0.5853713154792786, + -0.14019179344177246 + ], + "std": [ + 0.37648600339889526, + 0.6837278008460999, + 0.6710580587387085 + ], + "max": [ + 1.0, + 1.0, + 1.0 + ], + "min": [ + -1.0, + -1.0, + -1.0 + ], + "q01": [ + -0.8856967091560364, + -1.0, + -1.0 + ], + "q99": [ + 1.0, + 1.0, + 1.0 + ], + "mask": [ + true, + true, + true + ] + }, + "state": { + "mean": [ + 0.4167287349700928, + -0.011946885846555233, + 0.45969364047050476, + 0.6854016780853271, + 0.27180203795433044, + -0.13894067704677582, + 0.060337137430906296, + 0.37753522396087646, + -0.006504729390144348, + -0.0025828757788985968, + 0.01615341380238533 + ], + "std": [ + 0.3655839264392853, + 0.3772760033607483, + 0.2494690865278244, + 0.2961997389793396, + 0.6115167140960693, + 0.21205303072929382, + 0.45317184925079346, + 0.2155265510082245, + 0.3430417776107788, + 0.42244163155555725, + 0.6003957390785217 + ], + "max": [ + 0.997422456741333, + 0.9132802486419678, + 0.996956467628479, + 0.9437090158462524, + 0.9562854766845703, + 0.6047350168228149, + 0.9966599941253662, + 0.992188572883606, + 1.0, + 1.0, + 1.0 + ], + "min": [ + -0.9813680648803711, + -0.9943915605545044, + -0.240229070186615, + -0.7879800796508789, + -0.953981876373291, + -1.0, + -0.7385008335113525, + -0.6288865208625793, + -0.6290589570999146, + -1.0, + -1.0 + ], + "q01": [ + -0.527071304321289, + -0.9531343531608581, + -0.20607015371322632, + -0.10409546375274659, + -0.9325236558914185, + -0.7288819265365601, + -0.68619873046875, + -0.08794799327850342, + -0.5419587254524231, + -1.0, + -1.0 + ], + "q99": [ + 0.9375574779510498, + 0.7395017147064211, + 0.982518982887268, + 0.912756371498108, + 0.9081219291687013, + 0.23814808368682872, + 0.8914690685272217, + 0.922366099357605, + 1.0, + 0.9213636970520022, + 1.0 + ] + }, + "num_transitions": 9917, + "num_trajectories": 10 + } +} \ No newline at end of file diff --git a/zero-latency/hopper/vla/starvla-qwenoft-h1/hopper_rgb_state_l0_return_gt3000_100ep_openvla_native_continuous_projector_sft_5k/provenance.json b/zero-latency/hopper/vla/starvla-qwenoft-h1/hopper_rgb_state_l0_return_gt3000_100ep_openvla_native_continuous_projector_sft_5k/provenance.json new file mode 100644 index 0000000000000000000000000000000000000000..8a95fb0b4721f02823cece9f13b8b69c99fc7b9f --- /dev/null +++ b/zero-latency/hopper/vla/starvla-qwenoft-h1/hopper_rgb_state_l0_return_gt3000_100ep_openvla_native_continuous_projector_sft_5k/provenance.json @@ -0,0 +1,37 @@ +{ + "task": "hopper", + "model": "qwenoft", + "training_condition": "zero-latency", + "training_run_id": "hopper_rgb_state_l0_return_gt3000_100ep_openvla_native_continuous_projector_sft_5k", + "source": { + "repo_id": "latency-sensitive-bench/Standard-Pipeline", + "revision": "2208875f92b21e3f8ec2363a30a6705dfe56cc04", + "path": "hopper/zero_latency/OpenVLA", + "url": "https://huggingface.co/latency-sensitive-bench/Standard-Pipeline/tree/2208875f92b21e3f8ec2363a30a6705dfe56cc04/hopper/zero_latency/OpenVLA" + }, + "hf_repo_id": "latency-sensitive-bench/benchmark-models", + "hf_path": "zero-latency/hopper/vla/starvla-qwenoft-h1/hopper_rgb_state_l0_return_gt3000_100ep_openvla_native_continuous_projector_sft_5k", + "checkpoint": { + "source_file": "hopper/zero_latency/OpenVLA/checkpoints/steps_5000_pytorch_model.pt", + "source_sha256": "87e4cb2bf569982a181bb7a7642a64499012542ab2cbf1545cd1c654b0a3d7cd", + "source_bytes": 9785081249, + "file": "checkpoints/model.pt", + "selection_rule": "only published VLA checkpoint", + "method": "copy", + "sha256": "87e4cb2bf569982a181bb7a7642a64499012542ab2cbf1545cd1c654b0a3d7cd", + "bytes": 9785081249 + }, + "config_source": "hopper/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/hopper/vla/starvla-qwenoft-h1/hopper_rgb_state_l0_return_gt3000_100ep_openvla_native_continuous_projector_sft_5k/source/config.yaml b/zero-latency/hopper/vla/starvla-qwenoft-h1/hopper_rgb_state_l0_return_gt3000_100ep_openvla_native_continuous_projector_sft_5k/source/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..4b6f9c165553b6a9b9618d22e33067d2adcc854d --- /dev/null +++ b/zero-latency/hopper/vla/starvla-qwenoft-h1/hopper_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: hopper_rgb_state_l0_return_gt3000_100ep + dataset_py: lerobot_datasets + eval_data_mix: hopper_rgb_state_l0_return_gt3000_100ep__val + latency_curriculum: + enabled: false + per_device_batch_size: 16 +framework: + action_model: + action_dim: 3 + action_env_dim: 3 + action_hidden_dim: 2560 + action_horizon: 1 + action_model_type: MLP + loss_type: l1 + state_dim: 11 + 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/hopper_rgb_state_l0_return_gt3000_100ep_openvla_native_continuous_projector_sft_5k +rl_games: + env_eval: + enabled: false + task: gymnasium +run_id: hopper_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/hopper/vla/starvla-qwenoft-h1/hopper_rgb_state_l0_return_gt3000_100ep_openvla_native_continuous_projector_sft_5k/task_contract.json b/zero-latency/hopper/vla/starvla-qwenoft-h1/hopper_rgb_state_l0_return_gt3000_100ep_openvla_native_continuous_projector_sft_5k/task_contract.json new file mode 100644 index 0000000000000000000000000000000000000000..a678e1d462c6db0bc1ab23d85f4e827529e15783 --- /dev/null +++ b/zero-latency/hopper/vla/starvla-qwenoft-h1/hopper_rgb_state_l0_return_gt3000_100ep_openvla_native_continuous_projector_sft_5k/task_contract.json @@ -0,0 +1,62 @@ +{ + "action_space": { + "dtype": "float32", + "high": [ + 1.0, + 1.0, + 1.0 + ], + "labels": [ + "thigh_torque", + "leg_torque", + "foot_torque" + ], + "low": [ + -1.0, + -1.0, + -1.0 + ], + "type": "box" + }, + "base_prompt": "Move the Hopper robot forward while keeping its torso upright. Predict three continuous torques in [-1, 1] ordered as thigh, leg, and foot.", + "env_fps": 125.0, + "env_id": "LatencyBench/HopperRgbState-v0", + "frame_stack": 1, + "make_kwargs": { + "base_env_id": "Hopper-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 + ], + "obs_fps": 125.0, + "registration_imports": [ + "latency_bench.envs.gymnasium_hopper_rgb_state" + ], + "state_space": { + "labels": [ + "torso_height", + "torso_angle", + "thigh_angle", + "leg_angle", + "foot_angle", + "torso_x_velocity", + "torso_z_velocity", + "torso_angular_velocity", + "thigh_angular_velocity", + "leg_angular_velocity", + "foot_angular_velocity" + ] + }, + "task_name": "hopper_rgb_state" +} diff --git a/zero-latency/hopper/vla/starvla-qwenpi_v3-h1/hopper_pi05_l0_h1_5fps_g128_20260921/README.md b/zero-latency/hopper/vla/starvla-qwenpi_v3-h1/hopper_pi05_l0_h1_5fps_g128_20260921/README.md new file mode 100644 index 0000000000000000000000000000000000000000..a0e738db863e6f034aaa266570e00ae2d1ef2be3 --- /dev/null +++ b/zero-latency/hopper/vla/starvla-qwenpi_v3-h1/hopper_pi05_l0_h1_5fps_g128_20260921/README.md @@ -0,0 +1,31 @@ +# hopper / qwenpi_v3 + +Training condition: `zero-latency`. Run: `hopper_pi05_l0_h1_5fps_g128_20260921`. + +Copied from [Standard-Pipeline](https://huggingface.co/latency-sensitive-bench/Standard-Pipeline/tree/2208875f92b21e3f8ec2363a30a6705dfe56cc04/hopper/zero_latency/Pi05) at `2208875f92b21e3f8ec2363a30a6705dfe56cc04`. + +- Checkpoint: `checkpoints/model.pt` +- Selection: only published VLA checkpoint +- Checkpoint SHA256: `c35ddc57fef8fd096adca3df9ecf94662824c26b311c3884508f07646231f05b` +- 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/hopper/vla/starvla-qwenpi_v3-h1/hopper_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/hopper/vla/starvla-qwenpi_v3-h1/hopper_pi05_l0_h1_5fps_g128_20260921/checkpoints/model.pt b/zero-latency/hopper/vla/starvla-qwenpi_v3-h1/hopper_pi05_l0_h1_5fps_g128_20260921/checkpoints/model.pt new file mode 100644 index 0000000000000000000000000000000000000000..514e74dd5da12106f4e368c5520163c4326b14aa --- /dev/null +++ b/zero-latency/hopper/vla/starvla-qwenpi_v3-h1/hopper_pi05_l0_h1_5fps_g128_20260921/checkpoints/model.pt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c35ddc57fef8fd096adca3df9ecf94662824c26b311c3884508f07646231f05b +size 10922629277 diff --git a/zero-latency/hopper/vla/starvla-qwenpi_v3-h1/hopper_pi05_l0_h1_5fps_g128_20260921/config.full.yaml b/zero-latency/hopper/vla/starvla-qwenpi_v3-h1/hopper_pi05_l0_h1_5fps_g128_20260921/config.full.yaml new file mode 100644 index 0000000000000000000000000000000000000000..d6ec72abce705e3d32bbd258d18bcc4bcce838aa --- /dev/null +++ b/zero-latency/hopper/vla/starvla-qwenpi_v3-h1/hopper_pi05_l0_h1_5fps_g128_20260921/config.full.yaml @@ -0,0 +1,246 @@ +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: 3 + state_dim: 11 + 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: 3 + 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 + labels: + - thigh_torque + - leg_torque + - foot_torque + low: + - -1.0 + - -1.0 + - -1.0 + type: box + base_prompt: Move the Hopper robot forward while keeping its torso upright. + Predict three continuous torques in [-1, 1] ordered as thigh, leg, and foot. + env_fps: 5 + env_id: LatencyBench/HopperRgbState-v0 + frame_stack: 1 + make_kwargs: + base_env_id: Hopper-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 + obs_fps: 5 + registration_imports: + - latency_bench.envs.gymnasium_hopper_rgb_state + state_space: + labels: + - torso_height + - torso_angle + - thigh_angle + - leg_angle + - foot_angle + - torso_x_velocity + - torso_z_velocity + - torso_angular_velocity + - thigh_angular_velocity + - leg_angular_velocity + - foot_angular_velocity + task_name: hopper_rgb_state + active_action_dim: 3 +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: true + pretrained_checkpoint: ${PI05_RUN_DIR}/recovery/source_3000/steps_3000_state + resume_step: 3000 + 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: hopper_pi05_l0_h1_5fps_g128_20260921 +run_root_dir: ${PI05_RUN_DIR}/recovery/training +seed: 42 +is_debug: false +version_id: '0.21' +wandb_project: latency-sensitive-bench +wandb_entity: dongqianyu99-zhejiang-university +wandb_name: hopper_pi05_l0_h1_5fps_g128_20260921 +wandb_group: pi05-seven-env +wandb_tags: +- hopper +- 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}/recovery/train-resume3000.yaml +output_dir: ${PI05_RUN_DIR}/recovery/training/hopper_pi05_l0_h1_5fps_g128_20260921 diff --git a/zero-latency/hopper/vla/starvla-qwenpi_v3-h1/hopper_pi05_l0_h1_5fps_g128_20260921/config.yaml b/zero-latency/hopper/vla/starvla-qwenpi_v3-h1/hopper_pi05_l0_h1_5fps_g128_20260921/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..abd3b258077ddcbb1fabad859d50967ecc4287bf --- /dev/null +++ b/zero-latency/hopper/vla/starvla-qwenpi_v3-h1/hopper_pi05_l0_h1_5fps_g128_20260921/config.yaml @@ -0,0 +1,123 @@ +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: 3 + state_dim: 11 + 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: 3 + 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 + labels: + - thigh_torque + - leg_torque + - foot_torque + low: + - -1.0 + - -1.0 + - -1.0 + type: box + base_prompt: Move the Hopper robot forward while keeping its torso upright. + Predict three continuous torques in [-1, 1] ordered as thigh, leg, and foot. + env_fps: 5 + env_id: LatencyBench/HopperRgbState-v0 + frame_stack: 1 + make_kwargs: + base_env_id: Hopper-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 + obs_fps: 5 + registration_imports: + - latency_bench.envs.gymnasium_hopper_rgb_state + state_space: + labels: + - torso_height + - torso_angle + - thigh_angle + - leg_angle + - foot_angle + - torso_x_velocity + - torso_z_velocity + - torso_angular_velocity + - thigh_angular_velocity + - leg_angular_velocity + - foot_angular_velocity + task_name: hopper_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/hopper/vla/starvla-qwenpi_v3-h1/hopper_pi05_l0_h1_5fps_g128_20260921/dataset_statistics.json b/zero-latency/hopper/vla/starvla-qwenpi_v3-h1/hopper_pi05_l0_h1_5fps_g128_20260921/dataset_statistics.json new file mode 100644 index 0000000000000000000000000000000000000000..bb92bf6761b388f4009b3ee0beb94cee8054651c --- /dev/null +++ b/zero-latency/hopper/vla/starvla-qwenpi_v3-h1/hopper_pi05_l0_h1_5fps_g128_20260921/dataset_statistics.json @@ -0,0 +1,123 @@ +{ + "new_embodiment": { + "action": { + "mean": [ + 0.03865884989500046, + 0.5777319669723511, + -0.13737158477306366 + ], + "std": [ + 0.37739884853363037, + 0.689887523651123, + 0.6762200593948364 + ], + "max": [ + 1.0, + 1.0, + 1.0 + ], + "min": [ + -1.0, + -1.0, + -1.0 + ], + "q01": [ + -0.9043600207567215, + -1.0, + -1.0 + ], + "q99": [ + 1.0, + 1.0, + 1.0 + ], + "mask": [ + true, + true, + true + ] + }, + "state": { + "mean": [ + 0.40871402621269226, + -0.008348838426172733, + 0.45469287037849426, + 0.6798303723335266, + 0.269776850938797, + -0.1342218965291977, + 0.05999795347452164, + 0.3777448832988739, + -0.007698146626353264, + -0.002836690517142415, + 0.015370956622064114 + ], + "std": [ + 0.3633615970611572, + 0.37869560718536377, + 0.25348955392837524, + 0.30314165353775024, + 0.6127533316612244, + 0.21872857213020325, + 0.4510791301727295, + 0.21530857682228088, + 0.3457658588886261, + 0.4274502992630005, + 0.6040666103363037 + ], + "max": [ + 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 + ], + "q01": [ + -0.5281156021356582, + -0.9567392879724502, + -0.2079367846250534, + -0.12157819628715515, + -0.9323781633377075, + -0.7369779413938522, + -0.6852116698026657, + -0.10524752676486969, + -0.5511617350578308, + -1.0, + -1.0 + ], + "q99": [ + 0.9306531548500061, + 0.7862786006927489, + 0.9816559219360351, + 0.9127335548400879, + 0.9102470433712005, + 0.2971139311790463, + 0.8896300184726715, + 0.9198877382278442, + 1.0, + 0.981240049600601, + 1.0 + ] + }, + "num_transitions": 88212, + "num_trajectories": 90 + } +} \ No newline at end of file diff --git a/zero-latency/hopper/vla/starvla-qwenpi_v3-h1/hopper_pi05_l0_h1_5fps_g128_20260921/latency_prompt_map.json b/zero-latency/hopper/vla/starvla-qwenpi_v3-h1/hopper_pi05_l0_h1_5fps_g128_20260921/latency_prompt_map.json new file mode 100644 index 0000000000000000000000000000000000000000..f2b7129bb06cf782fa08ced52a646bb6c42dfaec --- /dev/null +++ b/zero-latency/hopper/vla/starvla-qwenpi_v3-h1/hopper_pi05_l0_h1_5fps_g128_20260921/latency_prompt_map.json @@ -0,0 +1,7 @@ +{ + "0": { + "prompt": "Move the Hopper robot forward while keeping its torso upright. Predict three continuous torques in [-1, 1] ordered as thigh, leg, and 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/hopper/vla/starvla-qwenpi_v3-h1/hopper_pi05_l0_h1_5fps_g128_20260921/manifest.json b/zero-latency/hopper/vla/starvla-qwenpi_v3-h1/hopper_pi05_l0_h1_5fps_g128_20260921/manifest.json new file mode 100644 index 0000000000000000000000000000000000000000..3a27a795874aae90d48af4b9a5efeeba56a5ba02 --- /dev/null +++ b/zero-latency/hopper/vla/starvla-qwenpi_v3-h1/hopper_pi05_l0_h1_5fps_g128_20260921/manifest.json @@ -0,0 +1,142 @@ +{ + "dataset_name": "hopper_h1_5fps_l0", + "env_name": "hopper_rgb_state", + "episodes": 90, + "frames": 88212, + "task_prompts": [ + "Move the Hopper robot forward while keeping its torso upright. Predict three continuous torques in [-1, 1] ordered as thigh, leg, and 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": "hopper_rgb_state", + "action_layout": "gymnasium_continuous_v1", + "action_labels": [ + "thigh_torque", + "leg_torque", + "foot_torque" + ], + "carrier_action_labels": [ + "thigh_torque", + "leg_torque", + "foot_torque" + ], + "action_dim": 3, + "active_action_dim": 3, + "action_carrier": "native", + "rows_unit": "decision_step", + "obs_fps": 5, + "obs_stride_raw_frames": 1, + "uses_state": true, + "state_dim": 11, + "state_labels": [ + "torso_height", + "torso_angle", + "thigh_angle", + "leg_angle", + "foot_angle", + "torso_x_velocity", + "torso_z_velocity", + "torso_angular_velocity", + "thigh_angular_velocity", + "leg_angular_velocity", + "foot_angular_velocity" + ], + "state_normalization": { + "type": "min_max", + "min": [ + 0.7007287740707397, + -0.19482052326202393, + -1.6461776494979858, + -1.9154568910598755, + -0.9631047248840332, + -0.1792079657316208, + -3.757478713989258, + -6.126420021057129, + -10.0, + -10.0, + -10.0 + ], + "max": [ + 1.6431776285171509, + 0.17731712758541107, + 0.059119515120983124, + 0.14385102689266205, + 0.9273068308830261, + 6.297187805175781, + 3.283433198928833, + 2.7663865089416504, + 10.0, + 10.0, + 10.0 + ] + }, + "robot_type": "rl_games_gymnasium", + "latency_prompt_map_path": "${PI05_RUN_DIR}/zero_latency/data/lerobot/hopper_h1_5fps_l0/latency_prompt_map.json", + "custom_mixtures_path": "${PI05_RUN_DIR}/zero_latency/data/lerobot/_generated_mixtures/hopper_h1_5fps_l0.json", + "gymnasium_task": { + "action_space": { + "dtype": "float32", + "high": [ + 1.0, + 1.0, + 1.0 + ], + "labels": [ + "thigh_torque", + "leg_torque", + "foot_torque" + ], + "low": [ + -1.0, + -1.0, + -1.0 + ], + "type": "box" + }, + "base_prompt": "Move the Hopper robot forward while keeping its torso upright. Predict three continuous torques in [-1, 1] ordered as thigh, leg, and foot.", + "env_fps": 5, + "env_id": "LatencyBench/HopperRgbState-v0", + "frame_stack": 1, + "make_kwargs": { + "base_env_id": "Hopper-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 + ], + "obs_fps": 5, + "registration_imports": [ + "latency_bench.envs.gymnasium_hopper_rgb_state" + ], + "state_space": { + "labels": [ + "torso_height", + "torso_angle", + "thigh_angle", + "leg_angle", + "foot_angle", + "torso_x_velocity", + "torso_z_velocity", + "torso_angular_velocity", + "thigh_angular_velocity", + "leg_angular_velocity", + "foot_angular_velocity" + ] + }, + "task_name": "hopper_rgb_state" + }, + "validation_dataset_name": "hopper_h1_5fps_l0__val", + "validation_episodes": 10, + "validation_frames": 9917 +} \ No newline at end of file diff --git a/zero-latency/hopper/vla/starvla-qwenpi_v3-h1/hopper_pi05_l0_h1_5fps_g128_20260921/provenance.json b/zero-latency/hopper/vla/starvla-qwenpi_v3-h1/hopper_pi05_l0_h1_5fps_g128_20260921/provenance.json new file mode 100644 index 0000000000000000000000000000000000000000..23b9d1d9fc1797238788969647564d97618a55b7 --- /dev/null +++ b/zero-latency/hopper/vla/starvla-qwenpi_v3-h1/hopper_pi05_l0_h1_5fps_g128_20260921/provenance.json @@ -0,0 +1,34 @@ +{ + "task": "hopper", + "model": "qwenpi_v3", + "training_condition": "zero-latency", + "training_run_id": "hopper_pi05_l0_h1_5fps_g128_20260921", + "source": { + "repo_id": "latency-sensitive-bench/Standard-Pipeline", + "revision": "2208875f92b21e3f8ec2363a30a6705dfe56cc04", + "path": "hopper/zero_latency/Pi05", + "url": "https://huggingface.co/latency-sensitive-bench/Standard-Pipeline/tree/2208875f92b21e3f8ec2363a30a6705dfe56cc04/hopper/zero_latency/Pi05" + }, + "hf_repo_id": "latency-sensitive-bench/benchmark-models", + "hf_path": "zero-latency/hopper/vla/starvla-qwenpi_v3-h1/hopper_pi05_l0_h1_5fps_g128_20260921", + "checkpoint": { + "source_file": "hopper/zero_latency/Pi05/checkpoints/model.pt", + "source_sha256": "c35ddc57fef8fd096adca3df9ecf94662824c26b311c3884508f07646231f05b", + "source_bytes": 10922629277, + "file": "checkpoints/model.pt", + "selection_rule": "only published VLA checkpoint", + "method": "copy", + "sha256": "c35ddc57fef8fd096adca3df9ecf94662824c26b311c3884508f07646231f05b", + "bytes": 10922629277 + }, + "config_source": "hopper/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/hopper/vla/starvla-qwenpi_v3-h1/hopper_pi05_l0_h1_5fps_g128_20260921/recovery.json b/zero-latency/hopper/vla/starvla-qwenpi_v3-h1/hopper_pi05_l0_h1_5fps_g128_20260921/recovery.json new file mode 100644 index 0000000000000000000000000000000000000000..d2a5acb498c1993fb3dd7d45cba6bde94c6c7130 --- /dev/null +++ b/zero-latency/hopper/vla/starvla-qwenpi_v3-h1/hopper_pi05_l0_h1_5fps_g128_20260921/recovery.json @@ -0,0 +1,86 @@ +{ + "state": "RECOVERY_PREPARED_NOT_LAUNCHED", + "source_archive": [ + { + "source": "${PI05_RUN_DIR}/zero_latency/vla/training/hopper_pi05_l0_h1_5fps_g128_20260921/checkpoints/steps_3000_pytorch_model.pt", + "archive": "${PI05_RUN_DIR}/recovery/source_3000/steps_3000_pytorch_model.pt", + "device": 2304, + "inode": 38084051, + "bytes": 10922629277 + }, + { + "source": "${PI05_RUN_DIR}/zero_latency/vla/training/hopper_pi05_l0_h1_5fps_g128_20260921/checkpoints/steps_3000_state/latest", + "archive": "${PI05_RUN_DIR}/recovery/source_3000/steps_3000_state/latest", + "device": 2304, + "inode": 38084048, + "bytes": 13 + }, + { + "source": "${PI05_RUN_DIR}/zero_latency/vla/training/hopper_pi05_l0_h1_5fps_g128_20260921/checkpoints/steps_3000_state/random_states_1.pkl", + "archive": "${PI05_RUN_DIR}/recovery/source_3000/steps_3000_state/random_states_1.pkl", + "device": 2304, + "inode": 38084050, + "bytes": 15017 + }, + { + "source": "${PI05_RUN_DIR}/zero_latency/vla/training/hopper_pi05_l0_h1_5fps_g128_20260921/checkpoints/steps_3000_state/random_states_0.pkl", + "archive": "${PI05_RUN_DIR}/recovery/source_3000/steps_3000_state/random_states_0.pkl", + "device": 2304, + "inode": 38084049, + "bytes": 15017 + }, + { + "source": "${PI05_RUN_DIR}/zero_latency/vla/training/hopper_pi05_l0_h1_5fps_g128_20260921/checkpoints/steps_3000_state/zero_to_fp32.py", + "archive": "${PI05_RUN_DIR}/recovery/source_3000/steps_3000_state/zero_to_fp32.py", + "device": 2304, + "inode": 38084047, + "bytes": 33272 + }, + { + "source": "${PI05_RUN_DIR}/zero_latency/vla/training/hopper_pi05_l0_h1_5fps_g128_20260921/checkpoints/steps_3000_state/pytorch_model/bf16_zero_pp_rank_0_mp_rank_00_optim_states.pt", + "archive": "${PI05_RUN_DIR}/recovery/source_3000/steps_3000_state/pytorch_model/bf16_zero_pp_rank_0_mp_rank_00_optim_states.pt", + "device": 2304, + "inode": 38084065, + "bytes": 30432724243 + }, + { + "source": "${PI05_RUN_DIR}/zero_latency/vla/training/hopper_pi05_l0_h1_5fps_g128_20260921/checkpoints/steps_3000_state/pytorch_model/mp_rank_00_model_states.pt", + "archive": "${PI05_RUN_DIR}/recovery/source_3000/steps_3000_state/pytorch_model/mp_rank_00_model_states.pt", + "device": 2304, + "inode": 38084043, + "bytes": 10144702589 + }, + { + "source": "${PI05_RUN_DIR}/zero_latency/vla/training/hopper_pi05_l0_h1_5fps_g128_20260921/checkpoints/steps_3000_state/pytorch_model/bf16_zero_pp_rank_1_mp_rank_00_optim_states.pt", + "archive": "${PI05_RUN_DIR}/recovery/source_3000/steps_3000_state/pytorch_model/bf16_zero_pp_rank_1_mp_rank_00_optim_states.pt", + "device": 2304, + "inode": 38084066, + "bytes": 30432702675 + } + ], + "shared_trainer_sha256": "ace9d82c8b1203cbd0b06385bbdba5ca11de5571fd58e49d7f4869d34541215d", + "isolated_trainer_sha256": "99b2e4fb1c67ae703b5e231ac1562ca298140b7b186795340378cd621e170ff8", + "source_step": 3000, + "target_total_steps": 5000, + "recomputed_failed_updates": 342, + "remaining_logical_updates": 2000, + "changed_config_keys": [ + "trainer.is_resume", + "trainer.pretrained_checkpoint", + "trainer.resume_step", + "run_root_dir" + ], + "wandb_run_id": "hopper-pi05-l0-h1-5fps-g128-20260921", + "wandb_resume": "must", + "claim": "Restored model/optimizer/LR/RNG and proved data cursor, not bitwise future CUDA trajectory", + "initialization_attempt1": { + "state": "FAILED_BEFORE_OPTIMIZER_UPDATE", + "updates": 0, + "reason": "Non-line-anchored assertion insertion altered prepare_training; restored exact original function", + "log": "zero_latency/formal_recovery.log", + "status": "zero_latency/formal_recovery.status.json", + "bad_snapshot_sha256": "9b00e38bafff5405f620c54536d3832bc48fa0cd8218c67017ebe66c800cb19b" + }, + "worker_rng_restored": false, + "data_equivalence_scope": "Deterministic normal transforms and epoch/index cursor verified; main RNG restored after iterator. No worker RNG or future CUDA bitwise claim." +} \ No newline at end of file diff --git a/zero-latency/hopper/vla/starvla-qwenpi_v3-h1/hopper_pi05_l0_h1_5fps_g128_20260921/source/README.md b/zero-latency/hopper/vla/starvla-qwenpi_v3-h1/hopper_pi05_l0_h1_5fps_g128_20260921/source/README.md new file mode 100644 index 0000000000000000000000000000000000000000..7b73691d4188671b3f031723e43ee347c45604c1 --- /dev/null +++ b/zero-latency/hopper/vla/starvla-qwenpi_v3-h1/hopper_pi05_l0_h1_5fps_g128_20260921/source/README.md @@ -0,0 +1,5 @@ +# Hopper Pi0.5 H1 — zero latency + +QwenPI_v3 with pinned Qwen3-VL-4B-Instruct, fresh action head, 5000 updates, seed42, global128=micro64 x accumulation1 x2H200, gradient checkpointing disabled, ZeRO2. Environment/observation5/5FPS. RGB224, 11D state with train-only minmax applied once and continuous projector, 3 native continuous actions, H1, four inference timesteps. Training and saved-model forward verified; 3090 profile/B/F pending. This is the project's Pi0.5-style StarVLA implementation. + +Recovery: initial attempt stopped after3342 updates due toNCCL ALLREDUCE timeout. Complete step3000 model/optimizer/RNG restored from hardlinked archive; scheduler and deterministic data cursor repaired in isolated trainer. Updates3001–3342 recomputed(342 repeated), target5000 unchanged; no extra scientific budget. Details andsnapshot SHA inrecovery.json/provenance.json; no optimizer published. FutureCUDA trajectory is notclaimedbitwise. diff --git a/zero-latency/hopper/vla/starvla-qwenpi_v3-h1/hopper_pi05_l0_h1_5fps_g128_20260921/source/provenance.json b/zero-latency/hopper/vla/starvla-qwenpi_v3-h1/hopper_pi05_l0_h1_5fps_g128_20260921/source/provenance.json new file mode 100644 index 0000000000000000000000000000000000000000..e2c7a44f703ed6e6aa7293d3c01c506a4a177fb0 --- /dev/null +++ b/zero-latency/hopper/vla/starvla-qwenpi_v3-h1/hopper_pi05_l0_h1_5fps_g128_20260921/source/provenance.json @@ -0,0 +1,204 @@ +{ + "task": "hopper", + "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": "hopper_pi05_l0_h1_5fps_g128_20260921", + "condition": "zero_latency", + "source": { + "task": "hopper", + "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": "hopper/zero_latency/shared/source_100ep_v1/demonstrations/raw/", + "sha256": { + "metadata.json": "6f6ba3dc2556aa2c16f81defde6231458cbe051d082ba7e054150ba1b65e39cd", + "train.parquet": "fcd0e418fbc33c531235c5765c1b5d3288dacd5b0a7bdcc6b288f1d4e1009b75", + "val.parquet": "1e53ba71d9efa4dd8cb8839a2e40206a9d682480c753e2bcdab8c3f417836e20" + }, + "timing_derivation": { + "source_env_fps": 125, + "source_obs_fps": 125, + "target_env_fps": 5, + "target_obs_fps": 5, + "source_revision": "b0cdd0505f6c6f2dadfca67638e4fec1b252756c", + "source_prefix": "hopper/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": 88212, + "episodes": 90, + "return_min": 3036.393681228161, + "return_mean": 3552.4750215795307, + "action_min": -1.0, + "action_max": 1.0 + }, + "val": { + "rows": 9917, + "episodes": 10, + "return_min": 3405.0092537999153, + "return_mean": 3579.1710964143276, + "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": 11, + "action_dim": 3, + "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": [ + 6, + 7 + ], + "gradient_checkpointing": false, + "zero_stage": 2, + "allocator": "expandable_segments:True", + "state_encoding": "continuous_projector" + }, + "recovery": { + "state": "RECOVERY_PREPARED_NOT_LAUNCHED", + "source_archive": [ + { + "source": "${PI05_RUN_DIR}/zero_latency/vla/training/hopper_pi05_l0_h1_5fps_g128_20260921/checkpoints/steps_3000_pytorch_model.pt", + "archive": "${PI05_RUN_DIR}/recovery/source_3000/steps_3000_pytorch_model.pt", + "device": 2304, + "inode": 38084051, + "bytes": 10922629277 + }, + { + "source": "${PI05_RUN_DIR}/zero_latency/vla/training/hopper_pi05_l0_h1_5fps_g128_20260921/checkpoints/steps_3000_state/latest", + "archive": "${PI05_RUN_DIR}/recovery/source_3000/steps_3000_state/latest", + "device": 2304, + "inode": 38084048, + "bytes": 13 + }, + { + "source": "${PI05_RUN_DIR}/zero_latency/vla/training/hopper_pi05_l0_h1_5fps_g128_20260921/checkpoints/steps_3000_state/random_states_1.pkl", + "archive": "${PI05_RUN_DIR}/recovery/source_3000/steps_3000_state/random_states_1.pkl", + "device": 2304, + "inode": 38084050, + "bytes": 15017 + }, + { + "source": "${PI05_RUN_DIR}/zero_latency/vla/training/hopper_pi05_l0_h1_5fps_g128_20260921/checkpoints/steps_3000_state/random_states_0.pkl", + "archive": "${PI05_RUN_DIR}/recovery/source_3000/steps_3000_state/random_states_0.pkl", + "device": 2304, + "inode": 38084049, + "bytes": 15017 + }, + { + "source": "${PI05_RUN_DIR}/zero_latency/vla/training/hopper_pi05_l0_h1_5fps_g128_20260921/checkpoints/steps_3000_state/zero_to_fp32.py", + "archive": "${PI05_RUN_DIR}/recovery/source_3000/steps_3000_state/zero_to_fp32.py", + "device": 2304, + "inode": 38084047, + "bytes": 33272 + }, + { + "source": "${PI05_RUN_DIR}/zero_latency/vla/training/hopper_pi05_l0_h1_5fps_g128_20260921/checkpoints/steps_3000_state/pytorch_model/bf16_zero_pp_rank_0_mp_rank_00_optim_states.pt", + "archive": "${PI05_RUN_DIR}/recovery/source_3000/steps_3000_state/pytorch_model/bf16_zero_pp_rank_0_mp_rank_00_optim_states.pt", + "device": 2304, + "inode": 38084065, + "bytes": 30432724243 + }, + { + "source": "${PI05_RUN_DIR}/zero_latency/vla/training/hopper_pi05_l0_h1_5fps_g128_20260921/checkpoints/steps_3000_state/pytorch_model/mp_rank_00_model_states.pt", + "archive": "${PI05_RUN_DIR}/recovery/source_3000/steps_3000_state/pytorch_model/mp_rank_00_model_states.pt", + "device": 2304, + "inode": 38084043, + "bytes": 10144702589 + }, + { + "source": "${PI05_RUN_DIR}/zero_latency/vla/training/hopper_pi05_l0_h1_5fps_g128_20260921/checkpoints/steps_3000_state/pytorch_model/bf16_zero_pp_rank_1_mp_rank_00_optim_states.pt", + "archive": "${PI05_RUN_DIR}/recovery/source_3000/steps_3000_state/pytorch_model/bf16_zero_pp_rank_1_mp_rank_00_optim_states.pt", + "device": 2304, + "inode": 38084066, + "bytes": 30432702675 + } + ], + "shared_trainer_sha256": "ace9d82c8b1203cbd0b06385bbdba5ca11de5571fd58e49d7f4869d34541215d", + "isolated_trainer_sha256": "99b2e4fb1c67ae703b5e231ac1562ca298140b7b186795340378cd621e170ff8", + "source_step": 3000, + "target_total_steps": 5000, + "recomputed_failed_updates": 342, + "remaining_logical_updates": 2000, + "changed_config_keys": [ + "trainer.is_resume", + "trainer.pretrained_checkpoint", + "trainer.resume_step", + "run_root_dir" + ], + "wandb_run_id": "hopper-pi05-l0-h1-5fps-g128-20260921", + "wandb_resume": "must", + "claim": "Restored model/optimizer/LR/RNG and proved data cursor, not bitwise future CUDA trajectory", + "initialization_attempt1": { + "state": "FAILED_BEFORE_OPTIMIZER_UPDATE", + "updates": 0, + "reason": "Non-line-anchored assertion insertion altered prepare_training; restored exact original function", + "log": "zero_latency/formal_recovery.log", + "status": "zero_latency/formal_recovery.status.json", + "bad_snapshot_sha256": "9b00e38bafff5405f620c54536d3832bc48fa0cd8218c67017ebe66c800cb19b" + }, + "worker_rng_restored": false, + "data_equivalence_scope": "Deterministic normal transforms and epoch/index cursor verified; main RNG restored after iterator. No worker RNG or future CUDA bitwise claim." + } + }, + "training_config_sha256": "f55ffc7e8359cedf1aea045be9b4128835d34ad8f7097de9ae2b65c51e65f81c", + "dataset_manifest_sha256": "8a42290a98fd9bc023644142b1562fac49d9045b5e6a27937c4ef4f02964bccc" +} diff --git a/zero-latency/hopper/vla/starvla-qwenpi_v3-h1/hopper_pi05_l0_h1_5fps_g128_20260921/task_contract.json b/zero-latency/hopper/vla/starvla-qwenpi_v3-h1/hopper_pi05_l0_h1_5fps_g128_20260921/task_contract.json new file mode 100644 index 0000000000000000000000000000000000000000..9f46ce4391f55da768343fd4aca8cf9c2d035daf --- /dev/null +++ b/zero-latency/hopper/vla/starvla-qwenpi_v3-h1/hopper_pi05_l0_h1_5fps_g128_20260921/task_contract.json @@ -0,0 +1,62 @@ +{ + "action_space": { + "dtype": "float32", + "high": [ + 1.0, + 1.0, + 1.0 + ], + "labels": [ + "thigh_torque", + "leg_torque", + "foot_torque" + ], + "low": [ + -1.0, + -1.0, + -1.0 + ], + "type": "box" + }, + "base_prompt": "Move the Hopper robot forward while keeping its torso upright. Predict three continuous torques in [-1, 1] ordered as thigh, leg, and foot.", + "env_fps": 5, + "env_id": "LatencyBench/HopperRgbState-v0", + "frame_stack": 1, + "make_kwargs": { + "base_env_id": "Hopper-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 + ], + "obs_fps": 5, + "registration_imports": [ + "latency_bench.envs.gymnasium_hopper_rgb_state" + ], + "state_space": { + "labels": [ + "torso_height", + "torso_angle", + "thigh_angle", + "leg_angle", + "foot_angle", + "torso_x_velocity", + "torso_z_velocity", + "torso_angular_velocity", + "thigh_angular_velocity", + "leg_angular_velocity", + "foot_angular_velocity" + ] + }, + "task_name": "hopper_rgb_state" +} diff --git a/zero-latency/hopper/vla/starvla-qwenpi_v3-h1/hopper_pi05_l0_h1_5fps_g128_20260921/validation.json b/zero-latency/hopper/vla/starvla-qwenpi_v3-h1/hopper_pi05_l0_h1_5fps_g128_20260921/validation.json new file mode 100644 index 0000000000000000000000000000000000000000..c2624c34a25745cd631de144c111b68ef7d67494 --- /dev/null +++ b/zero-latency/hopper/vla/starvla-qwenpi_v3-h1/hopper_pi05_l0_h1_5fps_g128_20260921/validation.json @@ -0,0 +1,18 @@ +{ + "state": "SAVED_BUNDLE_RELOAD_AND_REAL_SAMPLE_FORWARD_VERIFIED", + "verified_at": "2026-09-21T00:03:04.596509+00:00", + "forward_shape": [ + 1, + 1, + 3 + ], + "forward_finite": true, + "heldout_sample": 0, + "loader_rows": 9917, + "action_horizon": 1, + "state_dim": 11, + "action_dim": 3, + "state_encoding": "continuous_projector", + "checkpoint_sha256": "c35ddc57fef8fd096adca3df9ecf94662824c26b311c3884508f07646231f05b", + "reload": "Existing strict key check with documented tied Qwen lm_head equivalence only" +}