diff --git a/latency-aware/half-cheetah/small-policy/sample-factory-qwengr00t-rtx3090-v1/README.md b/latency-aware/half-cheetah/small-policy/sample-factory-qwengr00t-rtx3090-v1/README.md new file mode 100644 index 0000000000000000000000000000000000000000..557b38ca422a906156b84ff1ef0889c03b24f9eb --- /dev/null +++ b/latency-aware/half-cheetah/small-policy/sample-factory-qwengr00t-rtx3090-v1/README.md @@ -0,0 +1,28 @@ +# half-cheetah / sample-factory-appo + +Training condition: `latency-aware`. Run: `half_cheetah_profile_appo_h1_5fps_20260917`. + +Copied from [Standard-Pipeline](https://huggingface.co/latency-sensitive-bench/Standard-Pipeline/tree/2208875f92b21e3f8ec2363a30a6705dfe56cc04/half_cheetah/profile_latency/small_model/GR00T) at `2208875f92b21e3f8ec2363a30a6705dfe56cc04`. + +- Checkpoint: `checkpoint.pth` +- Selection: training_best +- Checkpoint SHA256: `b1fc0590189724808f40c0d27f902cbc53ce84614c3f606bdceacc8919799c77` +- 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. + +## Download + +```python +from huggingface_hub import snapshot_download +prefix = "latency-aware/half-cheetah/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/half-cheetah/small-policy/sample-factory-qwengr00t-rtx3090-v1/checkpoint.pth b/latency-aware/half-cheetah/small-policy/sample-factory-qwengr00t-rtx3090-v1/checkpoint.pth new file mode 100644 index 0000000000000000000000000000000000000000..67499a19e4ee4c307ae26ca69ce09a056da0b19f --- /dev/null +++ b/latency-aware/half-cheetah/small-policy/sample-factory-qwengr00t-rtx3090-v1/checkpoint.pth @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b1fc0590189724808f40c0d27f902cbc53ce84614c3f606bdceacc8919799c77 +size 29877 diff --git a/latency-aware/half-cheetah/small-policy/sample-factory-qwengr00t-rtx3090-v1/config.json b/latency-aware/half-cheetah/small-policy/sample-factory-qwengr00t-rtx3090-v1/config.json new file mode 100644 index 0000000000000000000000000000000000000000..223a34743b3539e7d74276ee541f69df2d0a7311 --- /dev/null +++ b/latency-aware/half-cheetah/small-policy/sample-factory-qwengr00t-rtx3090-v1/config.json @@ -0,0 +1,270 @@ +{ + "help": false, + "algo": "APPO", + "env": "latency_gymnasium", + "experiment": "half_cheetah_profile_appo_h1_5fps_20260917", + "train_dir": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/half_cheetah/profile_latency/teacher/checkpoints", + "restart_behavior": "overwrite", + "device": "gpu", + "seed": 2222, + "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": 1, + "pixel_format": "CHW", + "use_record_episode_statistics": false, + "with_wandb": true, + "wandb_user": "dongqianyu99-zhejiang-university", + "wandb_project": "latency-sensitive-bench", + "wandb_group": "gr00t-six-env-h1-5fps", + "wandb_job_type": "profile_teacher", + "wandb_tags": null, + "with_pbt": false, + "pbt_mix_policies_in_one_env": true, + "pbt_period_env_steps": 5000000, + "pbt_start_mutation": 20000000, + "pbt_replace_fraction": 0.3, + "pbt_mutation_rate": 0.15, + "pbt_replace_reward_gap": 0.1, + "pbt_replace_reward_gap_absolute": 1e-06, + "pbt_optimize_gamma": false, + "pbt_target_objective": "true_objective", + "pbt_perturb_min": 1.1, + "pbt_perturb_max": 1.5, + "gym_task_name": "half_cheetah", + "gym_env_id": "LatencyBench/HalfCheetahContinuous-v0", + "gym_make_kwargs_json": "{\"base_env_id\": \"HalfCheetah-v4\", \"render_mode\": \"rgb_array\", \"base_make_kwargs\": {}}", + "gym_registration_imports_json": "[\"latency_bench.envs.gymnasium_half_cheetah\"]", + "gym_action_space_json": "{\"type\": \"box\", \"labels\": [\"back_thigh_torque\", \"back_shin_torque\", \"back_foot_torque\", \"front_thigh_torque\", \"front_shin_torque\", \"front_foot_torque\"], \"low\": [-1.0, -1.0, -1.0, -1.0, -1.0, -1.0], \"high\": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0], \"dtype\": \"float32\"}", + "gym_action_labels_json": null, + "gym_action_values_json": null, + "gym_noop_action_id": null, + "gym_noop_action_json": "[0.0, 0.0, 0.0, 0.0, 0.0, 0.0]", + "gym_base_prompt": "Make the Half Cheetah run forward as fast as possible. Predict six continuous torques in [-1, 1] ordered as back thigh, back shin, back foot, front thigh, front shin, and front foot.", + "gym_state_labels_json": "[\"front_tip_height\", \"front_tip_angle\", \"back_thigh_angle\", \"back_shin_angle\", \"back_foot_angle\", \"front_thigh_angle\", \"front_shin_angle\", \"front_foot_angle\", \"front_tip_x_velocity\", \"front_tip_z_velocity\", \"front_tip_angular_velocity\", \"back_thigh_angular_velocity\", \"back_shin_angular_velocity\", \"back_foot_angular_velocity\", \"front_thigh_angular_velocity\", \"front_shin_angular_velocity\", \"front_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/half_cheetah/profile_latency/profile_3090_20260917T171246114715Z/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/half_cheetah/profile_latency/teacher/checkpoints/half_cheetah_profile_appo_h1_5fps_20260917/episode_metrics.jsonl", + "ppo": null, + "command_line": "--mode train --algo APPO --env latency_gymnasium --experiment half_cheetah_profile_appo_h1_5fps_20260917 --train_dir /mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/half_cheetah/profile_latency/teacher/checkpoints --restart_behavior overwrite --device gpu --seed 2222 --episode_metrics_path /mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/half_cheetah/profile_latency/teacher/checkpoints/half_cheetah_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 --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/half_cheetah/profile_latency/profile_3090_20260917T171246114715Z/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 half_cheetah --gym-env-id LatencyBench/HalfCheetahContinuous-v0 --gym-make-kwargs-json {\"base_env_id\": \"HalfCheetah-v4\", \"render_mode\": \"rgb_array\", \"base_make_kwargs\": {}} --gym-registration-imports-json [\"latency_bench.envs.gymnasium_half_cheetah\"] --gym-action-space-json {\"type\": \"box\", \"labels\": [\"back_thigh_torque\", \"back_shin_torque\", \"back_foot_torque\", \"front_thigh_torque\", \"front_shin_torque\", \"front_foot_torque\"], \"low\": [-1.0, -1.0, -1.0, -1.0, -1.0, -1.0], \"high\": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0], \"dtype\": \"float32\"} --gym-noop-action-json [0.0, 0.0, 0.0, 0.0, 0.0, 0.0] --gym-base-prompt Make the Half Cheetah run forward as fast as possible. Predict six continuous torques in [-1, 1] ordered as back thigh, back shin, back foot, front thigh, front shin, and front foot. --gym-state-labels-json [\"front_tip_height\", \"front_tip_angle\", \"back_thigh_angle\", \"back_shin_angle\", \"back_foot_angle\", \"front_thigh_angle\", \"front_shin_angle\", \"front_foot_angle\", \"front_tip_x_velocity\", \"front_tip_z_velocity\", \"front_tip_angular_velocity\", \"back_thigh_angular_velocity\", \"back_shin_angular_velocity\", \"back_foot_angular_velocity\", \"front_thigh_angular_velocity\", \"front_shin_angular_velocity\", \"front_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": "half_cheetah_profile_appo_h1_5fps_20260917", + "train_dir": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/half_cheetah/profile_latency/teacher/checkpoints", + "restart_behavior": "overwrite", + "device": "gpu", + "seed": 2222, + "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, + "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": "half_cheetah", + "gym_env_id": "LatencyBench/HalfCheetahContinuous-v0", + "gym_make_kwargs_json": "{\"base_env_id\": \"HalfCheetah-v4\", \"render_mode\": \"rgb_array\", \"base_make_kwargs\": {}}", + "gym_registration_imports_json": "[\"latency_bench.envs.gymnasium_half_cheetah\"]", + "gym_action_space_json": "{\"type\": \"box\", \"labels\": [\"back_thigh_torque\", \"back_shin_torque\", \"back_foot_torque\", \"front_thigh_torque\", \"front_shin_torque\", \"front_foot_torque\"], \"low\": [-1.0, -1.0, -1.0, -1.0, -1.0, -1.0], \"high\": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0], \"dtype\": \"float32\"}", + "gym_noop_action_json": "[0.0, 0.0, 0.0, 0.0, 0.0, 0.0]", + "gym_base_prompt": "Make the Half Cheetah run forward as fast as possible. Predict six continuous torques in [-1, 1] ordered as back thigh, back shin, back foot, front thigh, front shin, and front foot.", + "gym_state_labels_json": "[\"front_tip_height\", \"front_tip_angle\", \"back_thigh_angle\", \"back_shin_angle\", \"back_foot_angle\", \"front_thigh_angle\", \"front_shin_angle\", \"front_foot_angle\", \"front_tip_x_velocity\", \"front_tip_z_velocity\", \"front_tip_angular_velocity\", \"back_thigh_angular_velocity\", \"back_shin_angular_velocity\", \"back_foot_angular_velocity\", \"front_thigh_angular_velocity\", \"front_shin_angular_velocity\", \"front_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/half_cheetah/profile_latency/profile_3090_20260917T171246114715Z/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/half_cheetah/profile_latency/teacher/checkpoints/half_cheetah_profile_appo_h1_5fps_20260917/episode_metrics.jsonl", + "output_dir": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/half_cheetah/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/half_cheetah/profile_latency/teacher/training", + "wandb_unique_id": "half_cheetah_profile_appo_h1_5fps_20260917" +} \ No newline at end of file diff --git a/latency-aware/half-cheetah/small-policy/sample-factory-qwengr00t-rtx3090-v1/evaluation/selection_final/config.yaml b/latency-aware/half-cheetah/small-policy/sample-factory-qwengr00t-rtx3090-v1/evaluation/selection_final/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..9674c57b6acf457cf6287c06375bc406d3ca84ae --- /dev/null +++ b/latency-aware/half-cheetah/small-policy/sample-factory-qwengr00t-rtx3090-v1/evaluation/selection_final/config.yaml @@ -0,0 +1,167 @@ +experiment: + name: half_cheetah_profile_appo_h1_5fps_20260917 + seed: 42 +backend: + type: sample_factory + algo: APPO + device: cuda + train_dir: /mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/half_cheetah/profile_latency/teacher/checkpoints + restart_behavior: overwrite + run_mode: eval + extra_args: + - --wandb_user + - dongqianyu99-zhejiang-university +executor: + mode: simulated + simulated_worker_capacity: 1 + simulated_inference_pool: true + inference_devices: + - cuda:0 +env: + name: gymnasium + task_name: half_cheetah + env_id: LatencyBench/HalfCheetahContinuous-v0 + registration_imports: + - latency_bench.envs.gymnasium_half_cheetah + make_kwargs: + base_env_id: HalfCheetah-v4 + render_mode: rgb_array + base_make_kwargs: {} + env_fps: 5 + obs_fps: 5 + frame_stack: 1 + state_space: + labels: + - front_tip_height + - front_tip_angle + - back_thigh_angle + - back_shin_angle + - back_foot_angle + - front_thigh_angle + - front_shin_angle + - front_foot_angle + - front_tip_x_velocity + - front_tip_z_velocity + - front_tip_angular_velocity + - back_thigh_angular_velocity + - back_shin_angular_velocity + - back_foot_angular_velocity + - front_thigh_angular_velocity + - front_shin_angular_velocity + - front_foot_angular_velocity + action_space: + type: box + labels: + - back_thigh_torque + - back_shin_torque + - back_foot_torque + - front_thigh_torque + - front_shin_torque + - front_foot_torque + low: + - -1.0 + - -1.0 + - -1.0 + - -1.0 + - -1.0 + - -1.0 + high: + - 1.0 + - 1.0 + - 1.0 + - 1.0 + - 1.0 + - 1.0 + dtype: float32 + noop_action: + - 0.0 + - 0.0 + - 0.0 + - 0.0 + - 0.0 + - 0.0 + base_prompt: Make the Half Cheetah run forward as fast as possible. Predict six + continuous torques in [-1, 1] ordered as back thigh, back shin, back foot, front + thigh, front shin, and front foot. +latency: + method: iid + profile_path: /mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/half_cheetah/profile_latency/profile_3090_20260917T171246114715Z/profile.json + profile_worker_slot: 0 + seed: 0 + add_latency_info: false +scheduler: + hold_policy: hold + ordering_policy: issue_order_fifo +policy: + type: gymnasium_sf + checkpoint_path: /mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/half_cheetah/profile_latency/teacher/checkpoints/half_cheetah_profile_appo_h1_5fps_20260917/checkpoint_p0/checkpoint_000019536_10002432.pth + device: cuda:0 + deterministic: true +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 + 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/half_cheetah/profile_latency/selection_final + video: + enabled: false + num_bins: 1 + save_step_records: true + save_action_records: true + save_latency_records: true diff --git a/latency-aware/half-cheetah/small-policy/sample-factory-qwengr00t-rtx3090-v1/evaluation/selection_final/episode_metrics.jsonl b/latency-aware/half-cheetah/small-policy/sample-factory-qwengr00t-rtx3090-v1/evaluation/selection_final/episode_metrics.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..5686d913bb016f0476b3c7a79246996dfc11231f --- /dev/null +++ b/latency-aware/half-cheetah/small-policy/sample-factory-qwengr00t-rtx3090-v1/evaluation/selection_final/episode_metrics.jsonl @@ -0,0 +1,20 @@ +{"episode_id": 0, "episode_return": 1857.2374405048977, "episode_return_env": 1857.2374405048977, "game_score": null, "mean_latency_ms": 111.24583427755603, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/half_cheetah/profile_latency/teacher/checkpoints/half_cheetah_profile_appo_h1_5fps_20260917/checkpoint_p0/checkpoint_000019536_10002432.pth", "config_name": "half_cheetah_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/HalfCheetahContinuous-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/half_cheetah/profile_latency/selection_final", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "half_cheetah_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/half_cheetah/profile_latency/profile_3090_20260917T171246114715Z/profile.json", "source_run_id": "20260917T171246114715Z", "submitted_observation_frames": 1000, "workload_id": "half_cheetah"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 116.5727127719872, "p99_latency_ms": 118.703630845592, "return_raw": null, "survival_steps": 1000} +{"episode_id": 1, "episode_return": 1867.494728237067, "episode_return_env": 1867.494728237067, "game_score": null, "mean_latency_ms": 110.89120968248179, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/half_cheetah/profile_latency/teacher/checkpoints/half_cheetah_profile_appo_h1_5fps_20260917/checkpoint_p0/checkpoint_000019536_10002432.pth", "config_name": "half_cheetah_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/HalfCheetahContinuous-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/half_cheetah/profile_latency/selection_final", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "half_cheetah_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/half_cheetah/profile_latency/profile_3090_20260917T171246114715Z/profile.json", "source_run_id": "20260917T171246114715Z", "submitted_observation_frames": 1000, "workload_id": "half_cheetah"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 116.4121053434171, "p99_latency_ms": 118.23758657459547, "return_raw": null, "survival_steps": 1000} +{"episode_id": 2, "episode_return": 1834.5174806519765, "episode_return_env": 1834.5174806519765, "game_score": null, "mean_latency_ms": 110.72416504640468, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/half_cheetah/profile_latency/teacher/checkpoints/half_cheetah_profile_appo_h1_5fps_20260917/checkpoint_p0/checkpoint_000019536_10002432.pth", "config_name": "half_cheetah_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/HalfCheetahContinuous-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/half_cheetah/profile_latency/selection_final", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "half_cheetah_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/half_cheetah/profile_latency/profile_3090_20260917T171246114715Z/profile.json", "source_run_id": "20260917T171246114715Z", "submitted_observation_frames": 1000, "workload_id": "half_cheetah"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 116.05834305979504, "p99_latency_ms": 118.36271124083396, "return_raw": null, "survival_steps": 1000} +{"episode_id": 3, "episode_return": 1865.8690548707461, "episode_return_env": 1865.8690548707461, "game_score": null, "mean_latency_ms": 110.9907417554073, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/half_cheetah/profile_latency/teacher/checkpoints/half_cheetah_profile_appo_h1_5fps_20260917/checkpoint_p0/checkpoint_000019536_10002432.pth", "config_name": "half_cheetah_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/HalfCheetahContinuous-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/half_cheetah/profile_latency/selection_final", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "half_cheetah_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/half_cheetah/profile_latency/profile_3090_20260917T171246114715Z/profile.json", "source_run_id": "20260917T171246114715Z", "submitted_observation_frames": 1000, "workload_id": "half_cheetah"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 116.05442772201623, "p99_latency_ms": 118.4610390043948, "return_raw": null, "survival_steps": 1000} +{"episode_id": 4, "episode_return": 1801.7301131165686, "episode_return_env": 1801.7301131165686, "game_score": null, "mean_latency_ms": 111.04510103247057, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/half_cheetah/profile_latency/teacher/checkpoints/half_cheetah_profile_appo_h1_5fps_20260917/checkpoint_p0/checkpoint_000019536_10002432.pth", "config_name": "half_cheetah_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/HalfCheetahContinuous-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/half_cheetah/profile_latency/selection_final", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "half_cheetah_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/half_cheetah/profile_latency/profile_3090_20260917T171246114715Z/profile.json", "source_run_id": "20260917T171246114715Z", "submitted_observation_frames": 1000, "workload_id": "half_cheetah"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 116.0403861199309, "p99_latency_ms": 118.49525461948608, "return_raw": null, "survival_steps": 1000} +{"episode_id": 5, "episode_return": 1876.8244417767962, "episode_return_env": 1876.8244417767962, "game_score": null, "mean_latency_ms": 110.84752142739005, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/half_cheetah/profile_latency/teacher/checkpoints/half_cheetah_profile_appo_h1_5fps_20260917/checkpoint_p0/checkpoint_000019536_10002432.pth", "config_name": "half_cheetah_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/HalfCheetahContinuous-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/half_cheetah/profile_latency/selection_final", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "half_cheetah_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/half_cheetah/profile_latency/profile_3090_20260917T171246114715Z/profile.json", "source_run_id": "20260917T171246114715Z", "submitted_observation_frames": 1000, "workload_id": "half_cheetah"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 115.8723213773917, "p99_latency_ms": 118.40766092046617, "return_raw": null, "survival_steps": 1000} +{"episode_id": 6, "episode_return": 1865.6403856027441, "episode_return_env": 1865.6403856027441, "game_score": null, "mean_latency_ms": 110.93093487552736, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/half_cheetah/profile_latency/teacher/checkpoints/half_cheetah_profile_appo_h1_5fps_20260917/checkpoint_p0/checkpoint_000019536_10002432.pth", "config_name": "half_cheetah_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/HalfCheetahContinuous-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/half_cheetah/profile_latency/selection_final", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "half_cheetah_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/half_cheetah/profile_latency/profile_3090_20260917T171246114715Z/profile.json", "source_run_id": "20260917T171246114715Z", "submitted_observation_frames": 1000, "workload_id": "half_cheetah"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 116.17502496369826, "p99_latency_ms": 118.37483083815783, "return_raw": null, "survival_steps": 1000} +{"episode_id": 7, "episode_return": 1762.3797913309936, "episode_return_env": 1762.3797913309936, "game_score": null, "mean_latency_ms": 110.91768366022798, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/half_cheetah/profile_latency/teacher/checkpoints/half_cheetah_profile_appo_h1_5fps_20260917/checkpoint_p0/checkpoint_000019536_10002432.pth", "config_name": "half_cheetah_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/HalfCheetahContinuous-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/half_cheetah/profile_latency/selection_final", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "half_cheetah_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/half_cheetah/profile_latency/profile_3090_20260917T171246114715Z/profile.json", "source_run_id": "20260917T171246114715Z", "submitted_observation_frames": 1000, "workload_id": "half_cheetah"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 116.44383566008594, "p99_latency_ms": 118.9625626349772, "return_raw": null, "survival_steps": 1000} +{"episode_id": 8, "episode_return": 1835.7552201531942, "episode_return_env": 1835.7552201531942, "game_score": null, "mean_latency_ms": 110.97265965999314, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/half_cheetah/profile_latency/teacher/checkpoints/half_cheetah_profile_appo_h1_5fps_20260917/checkpoint_p0/checkpoint_000019536_10002432.pth", "config_name": "half_cheetah_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/HalfCheetahContinuous-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/half_cheetah/profile_latency/selection_final", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "half_cheetah_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/half_cheetah/profile_latency/profile_3090_20260917T171246114715Z/profile.json", "source_run_id": "20260917T171246114715Z", "submitted_observation_frames": 1000, "workload_id": "half_cheetah"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 116.30948159669535, "p99_latency_ms": 118.74422798450378, "return_raw": null, "survival_steps": 1000} +{"episode_id": 9, "episode_return": 1842.555192265558, "episode_return_env": 1842.555192265558, "game_score": null, "mean_latency_ms": 110.7400888862513, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/half_cheetah/profile_latency/teacher/checkpoints/half_cheetah_profile_appo_h1_5fps_20260917/checkpoint_p0/checkpoint_000019536_10002432.pth", "config_name": "half_cheetah_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/HalfCheetahContinuous-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/half_cheetah/profile_latency/selection_final", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "half_cheetah_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/half_cheetah/profile_latency/profile_3090_20260917T171246114715Z/profile.json", "source_run_id": "20260917T171246114715Z", "submitted_observation_frames": 1000, "workload_id": "half_cheetah"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 116.14594901709111, "p99_latency_ms": 118.35435209056368, "return_raw": null, "survival_steps": 1000} +{"episode_id": 10, "episode_return": 1857.8608419027873, "episode_return_env": 1857.8608419027873, "game_score": null, "mean_latency_ms": 110.9610735286344, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/half_cheetah/profile_latency/teacher/checkpoints/half_cheetah_profile_appo_h1_5fps_20260917/checkpoint_p0/checkpoint_000019536_10002432.pth", "config_name": "half_cheetah_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/HalfCheetahContinuous-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/half_cheetah/profile_latency/selection_final", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "half_cheetah_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/half_cheetah/profile_latency/profile_3090_20260917T171246114715Z/profile.json", "source_run_id": "20260917T171246114715Z", "submitted_observation_frames": 1000, "workload_id": "half_cheetah"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 116.08791093680271, "p99_latency_ms": 118.57852585751068, "return_raw": null, "survival_steps": 1000} +{"episode_id": 11, "episode_return": 1828.4981316832227, "episode_return_env": 1828.4981316832227, "game_score": null, "mean_latency_ms": 111.02514784905733, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/half_cheetah/profile_latency/teacher/checkpoints/half_cheetah_profile_appo_h1_5fps_20260917/checkpoint_p0/checkpoint_000019536_10002432.pth", "config_name": "half_cheetah_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/HalfCheetahContinuous-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/half_cheetah/profile_latency/selection_final", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "half_cheetah_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/half_cheetah/profile_latency/profile_3090_20260917T171246114715Z/profile.json", "source_run_id": "20260917T171246114715Z", "submitted_observation_frames": 1000, "workload_id": "half_cheetah"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 116.23220724586743, "p99_latency_ms": 118.30526355507156, "return_raw": null, "survival_steps": 1000} +{"episode_id": 12, "episode_return": 1858.3761893131157, "episode_return_env": 1858.3761893131157, "game_score": null, "mean_latency_ms": 110.78938788053553, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/half_cheetah/profile_latency/teacher/checkpoints/half_cheetah_profile_appo_h1_5fps_20260917/checkpoint_p0/checkpoint_000019536_10002432.pth", "config_name": "half_cheetah_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/HalfCheetahContinuous-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/half_cheetah/profile_latency/selection_final", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "half_cheetah_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/half_cheetah/profile_latency/profile_3090_20260917T171246114715Z/profile.json", "source_run_id": "20260917T171246114715Z", "submitted_observation_frames": 1000, "workload_id": "half_cheetah"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 116.01787974736017, "p99_latency_ms": 118.58826783447219, "return_raw": null, "survival_steps": 1000} +{"episode_id": 13, "episode_return": 1870.122593021503, "episode_return_env": 1870.122593021503, "game_score": null, "mean_latency_ms": 110.95709203860189, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/half_cheetah/profile_latency/teacher/checkpoints/half_cheetah_profile_appo_h1_5fps_20260917/checkpoint_p0/checkpoint_000019536_10002432.pth", "config_name": "half_cheetah_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/HalfCheetahContinuous-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/half_cheetah/profile_latency/selection_final", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "half_cheetah_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/half_cheetah/profile_latency/profile_3090_20260917T171246114715Z/profile.json", "source_run_id": "20260917T171246114715Z", "submitted_observation_frames": 1000, "workload_id": "half_cheetah"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 116.24608836340131, "p99_latency_ms": 118.18154033221496, "return_raw": null, "survival_steps": 1000} +{"episode_id": 14, "episode_return": 1842.1905916416376, "episode_return_env": 1842.1905916416376, "game_score": null, "mean_latency_ms": 110.89318622702031, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/half_cheetah/profile_latency/teacher/checkpoints/half_cheetah_profile_appo_h1_5fps_20260917/checkpoint_p0/checkpoint_000019536_10002432.pth", "config_name": "half_cheetah_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/HalfCheetahContinuous-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/half_cheetah/profile_latency/selection_final", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "half_cheetah_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/half_cheetah/profile_latency/profile_3090_20260917T171246114715Z/profile.json", "source_run_id": "20260917T171246114715Z", "submitted_observation_frames": 1000, "workload_id": "half_cheetah"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 116.2686548252319, "p99_latency_ms": 118.57311098329794, "return_raw": null, "survival_steps": 1000} +{"episode_id": 15, "episode_return": 1853.096570888896, "episode_return_env": 1853.096570888896, "game_score": null, "mean_latency_ms": 110.91502063789433, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/half_cheetah/profile_latency/teacher/checkpoints/half_cheetah_profile_appo_h1_5fps_20260917/checkpoint_p0/checkpoint_000019536_10002432.pth", "config_name": "half_cheetah_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/HalfCheetahContinuous-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/half_cheetah/profile_latency/selection_final", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "half_cheetah_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/half_cheetah/profile_latency/profile_3090_20260917T171246114715Z/profile.json", "source_run_id": "20260917T171246114715Z", "submitted_observation_frames": 1000, "workload_id": "half_cheetah"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 116.58213277385921, "p99_latency_ms": 118.82122257258376, "return_raw": null, "survival_steps": 1000} +{"episode_id": 16, "episode_return": 1874.5922690761886, "episode_return_env": 1874.5922690761886, "game_score": null, "mean_latency_ms": 110.81657893489775, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/half_cheetah/profile_latency/teacher/checkpoints/half_cheetah_profile_appo_h1_5fps_20260917/checkpoint_p0/checkpoint_000019536_10002432.pth", "config_name": "half_cheetah_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/HalfCheetahContinuous-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/half_cheetah/profile_latency/selection_final", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "half_cheetah_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/half_cheetah/profile_latency/profile_3090_20260917T171246114715Z/profile.json", "source_run_id": "20260917T171246114715Z", "submitted_observation_frames": 1000, "workload_id": "half_cheetah"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 116.15639521447764, "p99_latency_ms": 119.11157094082903, "return_raw": null, "survival_steps": 1000} +{"episode_id": 17, "episode_return": 1804.164970094345, "episode_return_env": 1804.164970094345, "game_score": null, "mean_latency_ms": 110.76658662771891, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/half_cheetah/profile_latency/teacher/checkpoints/half_cheetah_profile_appo_h1_5fps_20260917/checkpoint_p0/checkpoint_000019536_10002432.pth", "config_name": "half_cheetah_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/HalfCheetahContinuous-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/half_cheetah/profile_latency/selection_final", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "half_cheetah_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/half_cheetah/profile_latency/profile_3090_20260917T171246114715Z/profile.json", "source_run_id": "20260917T171246114715Z", "submitted_observation_frames": 1000, "workload_id": "half_cheetah"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 116.01034373542693, "p99_latency_ms": 118.40297172363019, "return_raw": null, "survival_steps": 1000} +{"episode_id": 18, "episode_return": 1864.918328697703, "episode_return_env": 1864.918328697703, "game_score": null, "mean_latency_ms": 110.80282639843057, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/half_cheetah/profile_latency/teacher/checkpoints/half_cheetah_profile_appo_h1_5fps_20260917/checkpoint_p0/checkpoint_000019536_10002432.pth", "config_name": "half_cheetah_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/HalfCheetahContinuous-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/half_cheetah/profile_latency/selection_final", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "half_cheetah_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/half_cheetah/profile_latency/profile_3090_20260917T171246114715Z/profile.json", "source_run_id": "20260917T171246114715Z", "submitted_observation_frames": 1000, "workload_id": "half_cheetah"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 116.23651719565382, "p99_latency_ms": 118.4700052289859, "return_raw": null, "survival_steps": 1000} +{"episode_id": 19, "episode_return": 1799.2772620012308, "episode_return_env": 1799.2772620012308, "game_score": null, "mean_latency_ms": 111.01576867826434, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/half_cheetah/profile_latency/teacher/checkpoints/half_cheetah_profile_appo_h1_5fps_20260917/checkpoint_p0/checkpoint_000019536_10002432.pth", "config_name": "half_cheetah_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/HalfCheetahContinuous-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/half_cheetah/profile_latency/selection_final", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "half_cheetah_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/half_cheetah/profile_latency/profile_3090_20260917T171246114715Z/profile.json", "source_run_id": "20260917T171246114715Z", "submitted_observation_frames": 1000, "workload_id": "half_cheetah"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 116.08249123050322, "p99_latency_ms": 118.64038701855631, "return_raw": null, "survival_steps": 1000} diff --git a/latency-aware/half-cheetah/small-policy/sample-factory-qwengr00t-rtx3090-v1/evaluation/selection_training_best/config.yaml b/latency-aware/half-cheetah/small-policy/sample-factory-qwengr00t-rtx3090-v1/evaluation/selection_training_best/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..31deada267f021dcd0fc626bce3e91e26c16d646 --- /dev/null +++ b/latency-aware/half-cheetah/small-policy/sample-factory-qwengr00t-rtx3090-v1/evaluation/selection_training_best/config.yaml @@ -0,0 +1,167 @@ +experiment: + name: half_cheetah_profile_appo_h1_5fps_20260917 + seed: 42 +backend: + type: sample_factory + algo: APPO + device: cuda + train_dir: /mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/half_cheetah/profile_latency/teacher/checkpoints + restart_behavior: overwrite + run_mode: eval + extra_args: + - --wandb_user + - dongqianyu99-zhejiang-university +executor: + mode: simulated + simulated_worker_capacity: 1 + simulated_inference_pool: true + inference_devices: + - cuda:0 +env: + name: gymnasium + task_name: half_cheetah + env_id: LatencyBench/HalfCheetahContinuous-v0 + registration_imports: + - latency_bench.envs.gymnasium_half_cheetah + make_kwargs: + base_env_id: HalfCheetah-v4 + render_mode: rgb_array + base_make_kwargs: {} + env_fps: 5 + obs_fps: 5 + frame_stack: 1 + state_space: + labels: + - front_tip_height + - front_tip_angle + - back_thigh_angle + - back_shin_angle + - back_foot_angle + - front_thigh_angle + - front_shin_angle + - front_foot_angle + - front_tip_x_velocity + - front_tip_z_velocity + - front_tip_angular_velocity + - back_thigh_angular_velocity + - back_shin_angular_velocity + - back_foot_angular_velocity + - front_thigh_angular_velocity + - front_shin_angular_velocity + - front_foot_angular_velocity + action_space: + type: box + labels: + - back_thigh_torque + - back_shin_torque + - back_foot_torque + - front_thigh_torque + - front_shin_torque + - front_foot_torque + low: + - -1.0 + - -1.0 + - -1.0 + - -1.0 + - -1.0 + - -1.0 + high: + - 1.0 + - 1.0 + - 1.0 + - 1.0 + - 1.0 + - 1.0 + dtype: float32 + noop_action: + - 0.0 + - 0.0 + - 0.0 + - 0.0 + - 0.0 + - 0.0 + base_prompt: Make the Half Cheetah run forward as fast as possible. Predict six + continuous torques in [-1, 1] ordered as back thigh, back shin, back foot, front + thigh, front shin, and front foot. +latency: + method: iid + profile_path: /mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/half_cheetah/profile_latency/profile_3090_20260917T171246114715Z/profile.json + profile_worker_slot: 0 + seed: 0 + add_latency_info: false +scheduler: + hold_policy: hold + ordering_policy: issue_order_fifo +policy: + type: gymnasium_sf + checkpoint_path: /mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/half_cheetah/profile_latency/teacher/checkpoints/half_cheetah_profile_appo_h1_5fps_20260917/checkpoint_p0/best_000018736_9592832_reward_1793.350.pth + device: cuda:0 + deterministic: true +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 + 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/half_cheetah/profile_latency/selection_training_best + video: + enabled: false + num_bins: 1 + save_step_records: true + save_action_records: true + save_latency_records: true diff --git a/latency-aware/half-cheetah/small-policy/sample-factory-qwengr00t-rtx3090-v1/evaluation/selection_training_best/episode_metrics.jsonl b/latency-aware/half-cheetah/small-policy/sample-factory-qwengr00t-rtx3090-v1/evaluation/selection_training_best/episode_metrics.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..473d75eaa841b943a3141db826076a7fa659e8ee --- /dev/null +++ b/latency-aware/half-cheetah/small-policy/sample-factory-qwengr00t-rtx3090-v1/evaluation/selection_training_best/episode_metrics.jsonl @@ -0,0 +1,20 @@ +{"episode_id": 0, "episode_return": 1887.561249494765, "episode_return_env": 1887.561249494765, "game_score": null, "mean_latency_ms": 111.24583427755603, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/half_cheetah/profile_latency/teacher/checkpoints/half_cheetah_profile_appo_h1_5fps_20260917/checkpoint_p0/best_000018736_9592832_reward_1793.350.pth", "config_name": "half_cheetah_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/HalfCheetahContinuous-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/half_cheetah/profile_latency/selection_training_best", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "half_cheetah_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/half_cheetah/profile_latency/profile_3090_20260917T171246114715Z/profile.json", "source_run_id": "20260917T171246114715Z", "submitted_observation_frames": 1000, "workload_id": "half_cheetah"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 116.5727127719872, "p99_latency_ms": 118.703630845592, "return_raw": null, "survival_steps": 1000} +{"episode_id": 1, "episode_return": 1895.0448055201541, "episode_return_env": 1895.0448055201541, "game_score": null, "mean_latency_ms": 110.89120968248179, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/half_cheetah/profile_latency/teacher/checkpoints/half_cheetah_profile_appo_h1_5fps_20260917/checkpoint_p0/best_000018736_9592832_reward_1793.350.pth", "config_name": "half_cheetah_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/HalfCheetahContinuous-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/half_cheetah/profile_latency/selection_training_best", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "half_cheetah_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/half_cheetah/profile_latency/profile_3090_20260917T171246114715Z/profile.json", "source_run_id": "20260917T171246114715Z", "submitted_observation_frames": 1000, "workload_id": "half_cheetah"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 116.4121053434171, "p99_latency_ms": 118.23758657459547, "return_raw": null, "survival_steps": 1000} +{"episode_id": 2, "episode_return": 1858.3344573252318, "episode_return_env": 1858.3344573252318, "game_score": null, "mean_latency_ms": 110.72416504640468, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/half_cheetah/profile_latency/teacher/checkpoints/half_cheetah_profile_appo_h1_5fps_20260917/checkpoint_p0/best_000018736_9592832_reward_1793.350.pth", "config_name": "half_cheetah_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/HalfCheetahContinuous-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/half_cheetah/profile_latency/selection_training_best", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "half_cheetah_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/half_cheetah/profile_latency/profile_3090_20260917T171246114715Z/profile.json", "source_run_id": "20260917T171246114715Z", "submitted_observation_frames": 1000, "workload_id": "half_cheetah"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 116.05834305979504, "p99_latency_ms": 118.36271124083396, "return_raw": null, "survival_steps": 1000} +{"episode_id": 3, "episode_return": 1830.0221069470836, "episode_return_env": 1830.0221069470836, "game_score": null, "mean_latency_ms": 110.9907417554073, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/half_cheetah/profile_latency/teacher/checkpoints/half_cheetah_profile_appo_h1_5fps_20260917/checkpoint_p0/best_000018736_9592832_reward_1793.350.pth", "config_name": "half_cheetah_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/HalfCheetahContinuous-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/half_cheetah/profile_latency/selection_training_best", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "half_cheetah_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/half_cheetah/profile_latency/profile_3090_20260917T171246114715Z/profile.json", "source_run_id": "20260917T171246114715Z", "submitted_observation_frames": 1000, "workload_id": "half_cheetah"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 116.05442772201623, "p99_latency_ms": 118.4610390043948, "return_raw": null, "survival_steps": 1000} +{"episode_id": 4, "episode_return": 1812.0352138530668, "episode_return_env": 1812.0352138530668, "game_score": null, "mean_latency_ms": 111.04510103247057, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/half_cheetah/profile_latency/teacher/checkpoints/half_cheetah_profile_appo_h1_5fps_20260917/checkpoint_p0/best_000018736_9592832_reward_1793.350.pth", "config_name": "half_cheetah_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/HalfCheetahContinuous-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/half_cheetah/profile_latency/selection_training_best", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "half_cheetah_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/half_cheetah/profile_latency/profile_3090_20260917T171246114715Z/profile.json", "source_run_id": "20260917T171246114715Z", "submitted_observation_frames": 1000, "workload_id": "half_cheetah"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 116.0403861199309, "p99_latency_ms": 118.49525461948608, "return_raw": null, "survival_steps": 1000} +{"episode_id": 5, "episode_return": 1793.54399269377, "episode_return_env": 1793.54399269377, "game_score": null, "mean_latency_ms": 110.84752142739005, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/half_cheetah/profile_latency/teacher/checkpoints/half_cheetah_profile_appo_h1_5fps_20260917/checkpoint_p0/best_000018736_9592832_reward_1793.350.pth", "config_name": "half_cheetah_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/HalfCheetahContinuous-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/half_cheetah/profile_latency/selection_training_best", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "half_cheetah_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/half_cheetah/profile_latency/profile_3090_20260917T171246114715Z/profile.json", "source_run_id": "20260917T171246114715Z", "submitted_observation_frames": 1000, "workload_id": "half_cheetah"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 115.8723213773917, "p99_latency_ms": 118.40766092046617, "return_raw": null, "survival_steps": 1000} +{"episode_id": 6, "episode_return": 1893.5209629203653, "episode_return_env": 1893.5209629203653, "game_score": null, "mean_latency_ms": 110.93093487552736, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/half_cheetah/profile_latency/teacher/checkpoints/half_cheetah_profile_appo_h1_5fps_20260917/checkpoint_p0/best_000018736_9592832_reward_1793.350.pth", "config_name": "half_cheetah_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/HalfCheetahContinuous-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/half_cheetah/profile_latency/selection_training_best", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "half_cheetah_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/half_cheetah/profile_latency/profile_3090_20260917T171246114715Z/profile.json", "source_run_id": "20260917T171246114715Z", "submitted_observation_frames": 1000, "workload_id": "half_cheetah"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 116.17502496369826, "p99_latency_ms": 118.37483083815783, "return_raw": null, "survival_steps": 1000} +{"episode_id": 7, "episode_return": 1835.4972454347915, "episode_return_env": 1835.4972454347915, "game_score": null, "mean_latency_ms": 110.91768366022798, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/half_cheetah/profile_latency/teacher/checkpoints/half_cheetah_profile_appo_h1_5fps_20260917/checkpoint_p0/best_000018736_9592832_reward_1793.350.pth", "config_name": "half_cheetah_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/HalfCheetahContinuous-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/half_cheetah/profile_latency/selection_training_best", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "half_cheetah_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/half_cheetah/profile_latency/profile_3090_20260917T171246114715Z/profile.json", "source_run_id": "20260917T171246114715Z", "submitted_observation_frames": 1000, "workload_id": "half_cheetah"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 116.44383566008594, "p99_latency_ms": 118.9625626349772, "return_raw": null, "survival_steps": 1000} +{"episode_id": 8, "episode_return": 1888.6783450211715, "episode_return_env": 1888.6783450211715, "game_score": null, "mean_latency_ms": 110.97265965999314, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/half_cheetah/profile_latency/teacher/checkpoints/half_cheetah_profile_appo_h1_5fps_20260917/checkpoint_p0/best_000018736_9592832_reward_1793.350.pth", "config_name": "half_cheetah_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/HalfCheetahContinuous-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/half_cheetah/profile_latency/selection_training_best", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "half_cheetah_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/half_cheetah/profile_latency/profile_3090_20260917T171246114715Z/profile.json", "source_run_id": "20260917T171246114715Z", "submitted_observation_frames": 1000, "workload_id": "half_cheetah"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 116.30948159669535, "p99_latency_ms": 118.74422798450378, "return_raw": null, "survival_steps": 1000} +{"episode_id": 9, "episode_return": 1831.1448449836785, "episode_return_env": 1831.1448449836785, "game_score": null, "mean_latency_ms": 110.7400888862513, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/half_cheetah/profile_latency/teacher/checkpoints/half_cheetah_profile_appo_h1_5fps_20260917/checkpoint_p0/best_000018736_9592832_reward_1793.350.pth", "config_name": "half_cheetah_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/HalfCheetahContinuous-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/half_cheetah/profile_latency/selection_training_best", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "half_cheetah_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/half_cheetah/profile_latency/profile_3090_20260917T171246114715Z/profile.json", "source_run_id": "20260917T171246114715Z", "submitted_observation_frames": 1000, "workload_id": "half_cheetah"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 116.14594901709111, "p99_latency_ms": 118.35435209056368, "return_raw": null, "survival_steps": 1000} +{"episode_id": 10, "episode_return": 1855.0970322771736, "episode_return_env": 1855.0970322771736, "game_score": null, "mean_latency_ms": 110.9610735286344, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/half_cheetah/profile_latency/teacher/checkpoints/half_cheetah_profile_appo_h1_5fps_20260917/checkpoint_p0/best_000018736_9592832_reward_1793.350.pth", "config_name": "half_cheetah_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/HalfCheetahContinuous-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/half_cheetah/profile_latency/selection_training_best", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "half_cheetah_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/half_cheetah/profile_latency/profile_3090_20260917T171246114715Z/profile.json", "source_run_id": "20260917T171246114715Z", "submitted_observation_frames": 1000, "workload_id": "half_cheetah"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 116.08791093680271, "p99_latency_ms": 118.57852585751068, "return_raw": null, "survival_steps": 1000} +{"episode_id": 11, "episode_return": 1855.3305649185393, "episode_return_env": 1855.3305649185393, "game_score": null, "mean_latency_ms": 111.02514784905733, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/half_cheetah/profile_latency/teacher/checkpoints/half_cheetah_profile_appo_h1_5fps_20260917/checkpoint_p0/best_000018736_9592832_reward_1793.350.pth", "config_name": "half_cheetah_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/HalfCheetahContinuous-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/half_cheetah/profile_latency/selection_training_best", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "half_cheetah_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/half_cheetah/profile_latency/profile_3090_20260917T171246114715Z/profile.json", "source_run_id": "20260917T171246114715Z", "submitted_observation_frames": 1000, "workload_id": "half_cheetah"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 116.23220724586743, "p99_latency_ms": 118.30526355507156, "return_raw": null, "survival_steps": 1000} +{"episode_id": 12, "episode_return": 1835.5406847810145, "episode_return_env": 1835.5406847810145, "game_score": null, "mean_latency_ms": 110.78938788053553, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/half_cheetah/profile_latency/teacher/checkpoints/half_cheetah_profile_appo_h1_5fps_20260917/checkpoint_p0/best_000018736_9592832_reward_1793.350.pth", "config_name": "half_cheetah_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/HalfCheetahContinuous-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/half_cheetah/profile_latency/selection_training_best", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "half_cheetah_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/half_cheetah/profile_latency/profile_3090_20260917T171246114715Z/profile.json", "source_run_id": "20260917T171246114715Z", "submitted_observation_frames": 1000, "workload_id": "half_cheetah"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 116.01787974736017, "p99_latency_ms": 118.58826783447219, "return_raw": null, "survival_steps": 1000} +{"episode_id": 13, "episode_return": 1845.6326210025886, "episode_return_env": 1845.6326210025886, "game_score": null, "mean_latency_ms": 110.95709203860189, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/half_cheetah/profile_latency/teacher/checkpoints/half_cheetah_profile_appo_h1_5fps_20260917/checkpoint_p0/best_000018736_9592832_reward_1793.350.pth", "config_name": "half_cheetah_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/HalfCheetahContinuous-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/half_cheetah/profile_latency/selection_training_best", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "half_cheetah_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/half_cheetah/profile_latency/profile_3090_20260917T171246114715Z/profile.json", "source_run_id": "20260917T171246114715Z", "submitted_observation_frames": 1000, "workload_id": "half_cheetah"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 116.24608836340131, "p99_latency_ms": 118.18154033221496, "return_raw": null, "survival_steps": 1000} +{"episode_id": 14, "episode_return": 1803.1661607436554, "episode_return_env": 1803.1661607436554, "game_score": null, "mean_latency_ms": 110.89318622702031, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/half_cheetah/profile_latency/teacher/checkpoints/half_cheetah_profile_appo_h1_5fps_20260917/checkpoint_p0/best_000018736_9592832_reward_1793.350.pth", "config_name": "half_cheetah_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/HalfCheetahContinuous-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/half_cheetah/profile_latency/selection_training_best", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "half_cheetah_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/half_cheetah/profile_latency/profile_3090_20260917T171246114715Z/profile.json", "source_run_id": "20260917T171246114715Z", "submitted_observation_frames": 1000, "workload_id": "half_cheetah"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 116.2686548252319, "p99_latency_ms": 118.57311098329794, "return_raw": null, "survival_steps": 1000} +{"episode_id": 15, "episode_return": 1830.418271781488, "episode_return_env": 1830.418271781488, "game_score": null, "mean_latency_ms": 110.91502063789433, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/half_cheetah/profile_latency/teacher/checkpoints/half_cheetah_profile_appo_h1_5fps_20260917/checkpoint_p0/best_000018736_9592832_reward_1793.350.pth", "config_name": "half_cheetah_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/HalfCheetahContinuous-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/half_cheetah/profile_latency/selection_training_best", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "half_cheetah_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/half_cheetah/profile_latency/profile_3090_20260917T171246114715Z/profile.json", "source_run_id": "20260917T171246114715Z", "submitted_observation_frames": 1000, "workload_id": "half_cheetah"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 116.58213277385921, "p99_latency_ms": 118.82122257258376, "return_raw": null, "survival_steps": 1000} +{"episode_id": 16, "episode_return": 1845.491576391448, "episode_return_env": 1845.491576391448, "game_score": null, "mean_latency_ms": 110.81657893489775, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/half_cheetah/profile_latency/teacher/checkpoints/half_cheetah_profile_appo_h1_5fps_20260917/checkpoint_p0/best_000018736_9592832_reward_1793.350.pth", "config_name": "half_cheetah_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/HalfCheetahContinuous-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/half_cheetah/profile_latency/selection_training_best", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "half_cheetah_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/half_cheetah/profile_latency/profile_3090_20260917T171246114715Z/profile.json", "source_run_id": "20260917T171246114715Z", "submitted_observation_frames": 1000, "workload_id": "half_cheetah"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 116.15639521447764, "p99_latency_ms": 119.11157094082903, "return_raw": null, "survival_steps": 1000} +{"episode_id": 17, "episode_return": 1854.8196809449396, "episode_return_env": 1854.8196809449396, "game_score": null, "mean_latency_ms": 110.76658662771891, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/half_cheetah/profile_latency/teacher/checkpoints/half_cheetah_profile_appo_h1_5fps_20260917/checkpoint_p0/best_000018736_9592832_reward_1793.350.pth", "config_name": "half_cheetah_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/HalfCheetahContinuous-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/half_cheetah/profile_latency/selection_training_best", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "half_cheetah_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/half_cheetah/profile_latency/profile_3090_20260917T171246114715Z/profile.json", "source_run_id": "20260917T171246114715Z", "submitted_observation_frames": 1000, "workload_id": "half_cheetah"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 116.01034373542693, "p99_latency_ms": 118.40297172363019, "return_raw": null, "survival_steps": 1000} +{"episode_id": 18, "episode_return": 1845.2868341282708, "episode_return_env": 1845.2868341282708, "game_score": null, "mean_latency_ms": 110.80282639843057, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/half_cheetah/profile_latency/teacher/checkpoints/half_cheetah_profile_appo_h1_5fps_20260917/checkpoint_p0/best_000018736_9592832_reward_1793.350.pth", "config_name": "half_cheetah_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/HalfCheetahContinuous-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/half_cheetah/profile_latency/selection_training_best", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "half_cheetah_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/half_cheetah/profile_latency/profile_3090_20260917T171246114715Z/profile.json", "source_run_id": "20260917T171246114715Z", "submitted_observation_frames": 1000, "workload_id": "half_cheetah"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 116.23651719565382, "p99_latency_ms": 118.4700052289859, "return_raw": null, "survival_steps": 1000} +{"episode_id": 19, "episode_return": 1848.1539286138711, "episode_return_env": 1848.1539286138711, "game_score": null, "mean_latency_ms": 111.01576867826434, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/half_cheetah/profile_latency/teacher/checkpoints/half_cheetah_profile_appo_h1_5fps_20260917/checkpoint_p0/best_000018736_9592832_reward_1793.350.pth", "config_name": "half_cheetah_profile_appo_h1_5fps_20260917", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/HalfCheetahContinuous-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/half_cheetah/profile_latency/selection_training_best", "policy_id": "gymnasium_sf", "profile_ref": null, "run_name": "half_cheetah_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/half_cheetah/profile_latency/profile_3090_20260917T171246114715Z/profile.json", "source_run_id": "20260917T171246114715Z", "submitted_observation_frames": 1000, "workload_id": "half_cheetah"}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 116.08249123050322, "p99_latency_ms": 118.64038701855631, "return_raw": null, "survival_steps": 1000} diff --git a/latency-aware/half-cheetah/small-policy/sample-factory-qwengr00t-rtx3090-v1/gate-summary.json b/latency-aware/half-cheetah/small-policy/sample-factory-qwengr00t-rtx3090-v1/gate-summary.json new file mode 100644 index 0000000000000000000000000000000000000000..88e484d44f6de05237efd5f7e32b3f6cd1e543b8 --- /dev/null +++ b/latency-aware/half-cheetah/small-policy/sample-factory-qwengr00t-rtx3090-v1/gate-summary.json @@ -0,0 +1,17 @@ +{ + "state": "NO_QUALIFYING_EPISODES", + "selected": "training_best", + "selected_mean": 1847.2160289888534, + "selected_population_std": 27.628393653569436, + "accepted_evaluation_episodes": 0, + "evaluation_episodes": 20, + "strict_gate": ">3000", + "not_run": [ + "E10", + "probe", + "data_collection", + "profile_VLA", + "F" + ], + "next": "Stop profilepipeline atscientificgate; noadditionalbudgetorlowergate" +} diff --git a/latency-aware/half-cheetah/small-policy/sample-factory-qwengr00t-rtx3090-v1/profile/DONE b/latency-aware/half-cheetah/small-policy/sample-factory-qwengr00t-rtx3090-v1/profile/DONE new file mode 100644 index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391 diff --git a/latency-aware/half-cheetah/small-policy/sample-factory-qwengr00t-rtx3090-v1/profile/hardware.json b/latency-aware/half-cheetah/small-policy/sample-factory-qwengr00t-rtx3090-v1/profile/hardware.json new file mode 100644 index 0000000000000000000000000000000000000000..e8204e81807044975c0e0a0953d16d6835192f31 --- /dev/null +++ b/latency-aware/half-cheetah/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/half-cheetah/small-policy/sample-factory-qwengr00t-rtx3090-v1/profile/latency_burst_model.json b/latency-aware/half-cheetah/small-policy/sample-factory-qwengr00t-rtx3090-v1/profile/latency_burst_model.json new file mode 100644 index 0000000000000000000000000000000000000000..654b3b2d2078aedfc4118025b1a90261c4363c66 --- /dev/null +++ b/latency-aware/half-cheetah/small-policy/sample-factory-qwengr00t-rtx3090-v1/profile/latency_burst_model.json @@ -0,0 +1,1308 @@ +{ + "burst_dwell_distribution": { + "distribution_type": "inverse_cdf", + "latency_ms": [ + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0 + ], + "quantile_levels": [ + 0.0, + 0.0001, + 0.0005, + 0.001, + 0.002, + 0.003, + 0.004, + 0.005, + 0.006, + 0.007, + 0.008, + 0.009, + 0.01, + 0.02, + 0.03, + 0.04, + 0.05, + 0.06, + 0.07, + 0.08, + 0.09, + 0.1, + 0.11, + 0.12, + 0.13, + 0.14, + 0.15, + 0.16, + 0.17, + 0.18, + 0.19, + 0.2, + 0.21, + 0.22, + 0.23, + 0.24, + 0.25, + 0.26, + 0.27, + 0.28, + 0.29, + 0.3, + 0.31, + 0.32, + 0.33, + 0.34, + 0.35, + 0.36, + 0.37, + 0.38, + 0.39, + 0.4, + 0.41, + 0.42, + 0.43, + 0.44, + 0.45, + 0.46, + 0.47, + 0.48, + 0.49, + 0.5, + 0.51, + 0.52, + 0.53, + 0.54, + 0.55, + 0.56, + 0.57, + 0.58, + 0.59, + 0.6, + 0.61, + 0.62, + 0.63, + 0.64, + 0.65, + 0.66, + 0.67, + 0.68, + 0.69, + 0.7, + 0.71, + 0.72, + 0.73, + 0.74, + 0.75, + 0.76, + 0.77, + 0.78, + 0.79, + 0.8, + 0.81, + 0.82, + 0.83, + 0.84, + 0.85, + 0.86, + 0.87, + 0.88, + 0.89, + 0.9, + 0.91, + 0.92, + 0.93, + 0.94, + 0.95, + 0.96, + 0.97, + 0.98, + 0.99, + 0.991, + 0.992, + 0.993, + 0.994, + 0.995, + 0.996, + 0.997, + 0.998, + 0.999, + 0.9995, + 0.9999, + 1.0 + ] + }, + "burst_dwell_lengths": [ + 1 + ], + "burst_merge_gap_records": 30, + "burst_rank_processes": [ + { + "draw_count": 1, + "dwell_length_spearman_rho": 0.0, + "level_ms": { + "distribution_type": "inverse_cdf", + "latency_ms": [ + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947, + 139.9069287776947 + ], + "quantile_levels": [ + 0.0, + 0.0001, + 0.0005, + 0.001, + 0.002, + 0.003, + 0.004, + 0.005, + 0.006, + 0.007, + 0.008, + 0.009, + 0.01, + 0.02, + 0.03, + 0.04, + 0.05, + 0.06, + 0.07, + 0.08, + 0.09, + 0.1, + 0.11, + 0.12, + 0.13, + 0.14, + 0.15, + 0.16, + 0.17, + 0.18, + 0.19, + 0.2, + 0.21, + 0.22, + 0.23, + 0.24, + 0.25, + 0.26, + 0.27, + 0.28, + 0.29, + 0.3, + 0.31, + 0.32, + 0.33, + 0.34, + 0.35, + 0.36, + 0.37, + 0.38, + 0.39, + 0.4, + 0.41, + 0.42, + 0.43, + 0.44, + 0.45, + 0.46, + 0.47, + 0.48, + 0.49, + 0.5, + 0.51, + 0.52, + 0.53, + 0.54, + 0.55, + 0.56, + 0.57, + 0.58, + 0.59, + 0.6, + 0.61, + 0.62, + 0.63, + 0.64, + 0.65, + 0.66, + 0.67, + 0.68, + 0.69, + 0.7, + 0.71, + 0.72, + 0.73, + 0.74, + 0.75, + 0.76, + 0.77, + 0.78, + 0.79, + 0.8, + 0.81, + 0.82, + 0.83, + 0.84, + 0.85, + 0.86, + 0.87, + 0.88, + 0.89, + 0.9, + 0.91, + 0.92, + 0.93, + 0.94, + 0.95, + 0.96, + 0.97, + 0.98, + 0.99, + 0.991, + 0.992, + 0.993, + 0.994, + 0.995, + 0.996, + 0.997, + 0.998, + 0.999, + 0.9995, + 0.9999, + 1.0 + ] + }, + "level_residual_ms": { + "distribution_type": "inverse_cdf", + "latency_ms": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ], + "quantile_levels": [ + 0.0, + 0.0001, + 0.0005, + 0.001, + 0.002, + 0.003, + 0.004, + 0.005, + 0.006, + 0.007, + 0.008, + 0.009, + 0.01, + 0.02, + 0.03, + 0.04, + 0.05, + 0.06, + 0.07, + 0.08, + 0.09, + 0.1, + 0.11, + 0.12, + 0.13, + 0.14, + 0.15, + 0.16, + 0.17, + 0.18, + 0.19, + 0.2, + 0.21, + 0.22, + 0.23, + 0.24, + 0.25, + 0.26, + 0.27, + 0.28, + 0.29, + 0.3, + 0.31, + 0.32, + 0.33, + 0.34, + 0.35, + 0.36, + 0.37, + 0.38, + 0.39, + 0.4, + 0.41, + 0.42, + 0.43, + 0.44, + 0.45, + 0.46, + 0.47, + 0.48, + 0.49, + 0.5, + 0.51, + 0.52, + 0.53, + 0.54, + 0.55, + 0.56, + 0.57, + 0.58, + 0.59, + 0.6, + 0.61, + 0.62, + 0.63, + 0.64, + 0.65, + 0.66, + 0.67, + 0.68, + 0.69, + 0.7, + 0.71, + 0.72, + 0.73, + 0.74, + 0.75, + 0.76, + 0.77, + 0.78, + 0.79, + 0.8, + 0.81, + 0.82, + 0.83, + 0.84, + 0.85, + 0.86, + 0.87, + 0.88, + 0.89, + 0.9, + 0.91, + 0.92, + 0.93, + 0.94, + 0.95, + 0.96, + 0.97, + 0.98, + 0.99, + 0.991, + 0.992, + 0.993, + 0.994, + 0.995, + 0.996, + 0.997, + 0.998, + 0.999, + 0.9995, + 0.9999, + 1.0 + ] + }, + "severity": { + "distribution_type": "inverse_cdf", + "latency_ms": [ + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0 + ], + "quantile_levels": [ + 0.0, + 0.0001, + 0.0005, + 0.001, + 0.002, + 0.003, + 0.004, + 0.005, + 0.006, + 0.007, + 0.008, + 0.009, + 0.01, + 0.02, + 0.03, + 0.04, + 0.05, + 0.06, + 0.07, + 0.08, + 0.09, + 0.1, + 0.11, + 0.12, + 0.13, + 0.14, + 0.15, + 0.16, + 0.17, + 0.18, + 0.19, + 0.2, + 0.21, + 0.22, + 0.23, + 0.24, + 0.25, + 0.26, + 0.27, + 0.28, + 0.29, + 0.3, + 0.31, + 0.32, + 0.33, + 0.34, + 0.35, + 0.36, + 0.37, + 0.38, + 0.39, + 0.4, + 0.41, + 0.42, + 0.43, + 0.44, + 0.45, + 0.46, + 0.47, + 0.48, + 0.49, + 0.5, + 0.51, + 0.52, + 0.53, + 0.54, + 0.55, + 0.56, + 0.57, + 0.58, + 0.59, + 0.6, + 0.61, + 0.62, + 0.63, + 0.64, + 0.65, + 0.66, + 0.67, + 0.68, + 0.69, + 0.7, + 0.71, + 0.72, + 0.73, + 0.74, + 0.75, + 0.76, + 0.77, + 0.78, + 0.79, + 0.8, + 0.81, + 0.82, + 0.83, + 0.84, + 0.85, + 0.86, + 0.87, + 0.88, + 0.89, + 0.9, + 0.91, + 0.92, + 0.93, + 0.94, + 0.95, + 0.96, + 0.97, + 0.98, + 0.99, + 0.991, + 0.992, + 0.993, + 0.994, + 0.995, + 0.996, + 0.997, + 0.998, + 0.999, + 0.9995, + 0.9999, + 1.0 + ] + }, + "spike_count": 1 + } + ], + "burst_slot_rank_templates": [ + { + "dwell_length": 1, + "slot_to_rank": { + "0": 0 + } + } + ], + "model_type": "hidden_regime", + "pre_worker_time_ms": { + "distribution_type": "inverse_cdf", + "latency_ms": [ + 0.30553507804870605, + 0.3077211380004883, + 0.3164808750152588, + 0.31953859329223633, + 0.32552826404571533, + 0.32983291149139404, + 0.33376097679138184, + 0.33717095851898193, + 0.33983349800109863, + 0.34331464767456055, + 0.34679698944091797, + 0.348968505859375, + 0.3508632183074951, + 0.368625283241272, + 0.38565897941589355, + 0.4070460796356201, + 0.4611990451812744, + 0.6316664218902588, + 0.8840409517288211, + 1.1662169694900513, + 1.1765408515930176, + 1.1826759576797485, + 1.1880431175231934, + 1.192684531211853, + 1.1972424983978271, + 1.2018431425094604, + 1.2057654857635498, + 1.2101205587387085, + 1.2142618894577026, + 1.2186999320983887, + 1.222942590713501, + 1.2273939847946167, + 1.2317519187927246, + 1.2365156412124634, + 1.2413979768753052, + 1.2463014125823975, + 1.2513395547866821, + 1.2571754455566406, + 1.263430118560791, + 1.2700271606445312, + 1.275714635848999, + 1.2796019315719604, + 1.2827589511871338, + 1.2852050065994263, + 1.287244439125061, + 1.2892138957977295, + 1.2908059358596802, + 1.2922614812850952, + 1.29354989528656, + 1.294797420501709, + 1.2959740161895752, + 1.2971434593200684, + 1.2982224225997925, + 1.2992639541625977, + 1.3003689050674438, + 1.3014429807662964, + 1.3025270700454712, + 1.3035889863967896, + 1.3046104907989502, + 1.305594563484192, + 1.3065420389175415, + 1.3075339794158936, + 1.3084988594055176, + 1.3095744848251343, + 1.3105474710464478, + 1.3115888833999634, + 1.3126180171966553, + 1.3135755062103271, + 1.3145705461502075, + 1.3156801462173462, + 1.31684148311615, + 1.3178833723068237, + 1.3190351724624634, + 1.3201690912246704, + 1.321287751197815, + 1.3223395347595215, + 1.3234862089157104, + 1.324728012084961, + 1.3260740041732788, + 1.3272501230239868, + 1.328466534614563, + 1.3298245668411255, + 1.3311378955841064, + 1.3325191736221313, + 1.3339595794677734, + 1.3355170488357544, + 1.3370128870010376, + 1.3387715816497803, + 1.3404450416564941, + 1.342273473739624, + 1.3439849615097046, + 1.3459140062332153, + 1.347856879234314, + 1.3500744104385376, + 1.352415919303894, + 1.3550053834915161, + 1.3578065633773804, + 1.360755443572998, + 1.3641879558563232, + 1.3683834075927734, + 1.3727985620498657, + 1.3783611059188843, + 1.3850845098495483, + 1.3929460048675537, + 1.4022895097732544, + 1.4133398532867432, + 1.4243890047073364, + 1.4366841316223145, + 1.4502850770950317, + 1.4661744832992554, + 1.4898686408996582, + 1.4927234649658203, + 1.496643304824829, + 1.499802827835083, + 1.5039091110229492, + 1.5099905729293823, + 1.5158015489578247, + 1.5254449844360352, + 1.5403350591659546, + 1.587038516998291, + 1.626986026763916, + 1.685157060623169, + 1.759110927581787 + ], + "quantile_levels": [ + 0.0, + 0.0001, + 0.0005, + 0.001, + 0.002, + 0.003, + 0.004, + 0.005, + 0.006, + 0.007, + 0.008, + 0.009, + 0.01, + 0.02, + 0.03, + 0.04, + 0.05, + 0.06, + 0.07, + 0.08, + 0.09, + 0.1, + 0.11, + 0.12, + 0.13, + 0.14, + 0.15, + 0.16, + 0.17, + 0.18, + 0.19, + 0.2, + 0.21, + 0.22, + 0.23, + 0.24, + 0.25, + 0.26, + 0.27, + 0.28, + 0.29, + 0.3, + 0.31, + 0.32, + 0.33, + 0.34, + 0.35, + 0.36, + 0.37, + 0.38, + 0.39, + 0.4, + 0.41, + 0.42, + 0.43, + 0.44, + 0.45, + 0.46, + 0.47, + 0.48, + 0.49, + 0.5, + 0.51, + 0.52, + 0.53, + 0.54, + 0.55, + 0.56, + 0.57, + 0.58, + 0.59, + 0.6, + 0.61, + 0.62, + 0.63, + 0.64, + 0.65, + 0.66, + 0.67, + 0.68, + 0.69, + 0.7, + 0.71, + 0.72, + 0.73, + 0.74, + 0.75, + 0.76, + 0.77, + 0.78, + 0.79, + 0.8, + 0.81, + 0.82, + 0.83, + 0.84, + 0.85, + 0.86, + 0.87, + 0.88, + 0.89, + 0.9, + 0.91, + 0.92, + 0.93, + 0.94, + 0.95, + 0.96, + 0.97, + 0.98, + 0.99, + 0.991, + 0.992, + 0.993, + 0.994, + 0.995, + 0.996, + 0.997, + 0.998, + 0.999, + 0.9995, + 0.9999, + 1.0 + ] + }, + "regime_step_counts": { + "burst": 1, + "calm": 44999 + }, + "regime_transition_counts": { + "burst": { + "burst": 0, + "calm": 1 + }, + "calm": { + "burst": 1, + "calm": 44993 + } + }, + "reset_scope": "session", + "schema_version": 12, + "spike_median_multiplier": 1.25, + "spike_threshold_ms_by_worker_slot": { + "0": 136.5302787721157 + }, + "worker_count": 1 +} diff --git a/latency-aware/half-cheetah/small-policy/sample-factory-qwengr00t-rtx3090-v1/profile/latency_distribution.json b/latency-aware/half-cheetah/small-policy/sample-factory-qwengr00t-rtx3090-v1/profile/latency_distribution.json new file mode 100644 index 0000000000000000000000000000000000000000..04f26e776d708822f7a53b45e5d3f12baeba9a61 --- /dev/null +++ b/latency-aware/half-cheetah/small-policy/sample-factory-qwengr00t-rtx3090-v1/profile/latency_distribution.json @@ -0,0 +1,1027 @@ +{ + "schema_version": 3, + "worker_slots": { + "0": { + "all": { + "count": 45000, + "observation_to_action_latency_ms": { + "distribution_type": "inverse_cdf", + "latency_ms": [ + 98.70396614074707, + 98.93569874763489, + 99.48969292640686, + 99.74139285087585, + 99.99378800392151, + 100.23676657676697, + 100.45681989192963, + 100.61310255527496, + 100.76971435546875, + 100.91665184497833, + 101.04355335235596, + 101.16939151287079, + 101.26515638828278, + 102.14995896816254, + 102.71126055717468, + 103.18618357181549, + 103.5731360912323, + 103.8581645488739, + 104.08811604976654, + 104.26839756965637, + 104.44286751747131, + 104.59254956245422, + 104.73005056381226, + 104.86736297607422, + 104.99852454662323, + 105.11451184749603, + 105.22956800460815, + 105.34380328655243, + 105.46738255023956, + 105.5861269235611, + 105.72314298152924, + 105.86401557922363, + 106.00613284111023, + 106.1586788892746, + 106.3304455280304, + 106.49429941177368, + 106.6677074432373, + 106.88013243675232, + 107.11437559127808, + 107.39064860343933, + 107.76535844802856, + 108.26877391338348, + 109.1403124332428, + 109.8295773267746, + 110.14285898208618, + 110.38829505443573, + 110.55411946773529, + 110.7079405784607, + 110.84284853935242, + 110.94373548030853, + 111.0390625, + 111.1330189704895, + 111.21616291999817, + 111.30723702907562, + 111.39569699764252, + 111.4632009267807, + 111.53299856185913, + 111.6087236404419, + 111.68404424190521, + 111.75892806053162, + 111.83013546466827, + 111.89721739292145, + 111.96227693557739, + 112.0322414636612, + 112.09205996990204, + 112.1550350189209, + 112.21599662303925, + 112.27896404266357, + 112.34797739982605, + 112.41241264343262, + 112.47833442687988, + 112.5394195318222, + 112.59503257274628, + 112.66607999801636, + 112.74150538444519, + 112.81656002998352, + 112.88787245750427, + 112.96066892147064, + 113.03999197483063, + 113.12193381786346, + 113.20433247089386, + 113.29065752029419, + 113.38739097118378, + 113.48024344444275, + 113.57242012023926, + 113.67222535610199, + 113.78520202636719, + 113.90836346149445, + 114.03478968143463, + 114.17524063587189, + 114.33557045459747, + 114.5150785446167, + 114.69101011753082, + 114.87553763389587, + 115.04817807674408, + 115.21558105945587, + 115.4046676158905, + 115.56282246112823, + 115.73450660705566, + 115.90149509906769, + 116.06068742275238, + 116.21335291862488, + 116.38565921783447, + 116.53832197189331, + 116.72133803367615, + 116.9329149723053, + 117.14166796207428, + 117.37930107116699, + 117.6389229297638, + 118.00152051448822, + 118.53854954242706, + 118.61444401741028, + 118.718177318573, + 118.81481945514679, + 118.91746199131012, + 119.05122864246368, + 119.20978999137878, + 119.4237105846405, + 119.78174543380737, + 120.79663491249084, + 125.07273578643799, + 131.77028608322144, + 140.54522109031677 + ], + "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": 0.9992538887917277, + "worker_service_time_ms": { + "distribution_type": "inverse_cdf", + "latency_ms": [ + 98.31866312026978, + 98.56582283973694, + 99.09732794761658, + 99.34384250640869, + 99.60134279727936, + 99.86176609992981, + 100.06919705867767, + 100.20859134197235, + 100.35521805286407, + 100.51889312267303, + 100.64560198783875, + 100.76041841506958, + 100.85606694221497, + 101.66479992866516, + 102.09854483604431, + 102.38846600055695, + 102.62049758434296, + 102.81858789920807, + 102.98463356494904, + 103.14342212677002, + 103.28345096111298, + 103.41380250453949, + 103.54721701145172, + 103.67480552196503, + 103.77950954437256, + 103.88921844959259, + 104.00332796573639, + 104.11410200595856, + 104.23153555393219, + 104.35140717029572, + 104.48387706279755, + 104.61716043949127, + 104.76020801067352, + 104.91312956809998, + 105.07158708572388, + 105.24114692211151, + 105.41247749328613, + 105.62178659439087, + 105.85350358486176, + 106.14066958427429, + 106.5293470621109, + 107.06386351585388, + 107.92944240570068, + 108.54334259033203, + 108.85813748836517, + 109.0984525680542, + 109.2556928396225, + 109.41076350212097, + 109.53469741344452, + 109.63675737380981, + 109.73488652706146, + 109.83300447463989, + 109.91255807876587, + 110.00031197071075, + 110.08124887943268, + 110.15699744224548, + 110.22898960113525, + 110.29925954341888, + 110.37904095649719, + 110.44585633277893, + 110.51946687698364, + 110.5860960483551, + 110.6517504453659, + 110.72050642967224, + 110.77966952323914, + 110.84107339382172, + 110.89812338352203, + 110.96228384971619, + 111.02794694900513, + 111.09109902381897, + 111.15636229515076, + 111.21828901767731, + 111.28044390678406, + 111.34816014766693, + 111.42458152770996, + 111.49758005142212, + 111.57378840446472, + 111.6487375497818, + 111.72672939300537, + 111.80910456180573, + 111.88192296028137, + 111.97343039512634, + 112.0639683008194, + 112.15919697284698, + 112.25180351734161, + 112.35117554664612, + 112.46263563632965, + 112.585289478302, + 112.71991741657257, + 112.85827541351318, + 113.02608108520508, + 113.19591450691223, + 113.37514591217041, + 113.5558249950409, + 113.73285055160522, + 113.89902758598328, + 114.09989607334137, + 114.25527513027191, + 114.43847215175629, + 114.60090148448944, + 114.7521880865097, + 114.90961241722107, + 115.07374906539917, + 115.23099088668823, + 115.41548895835876, + 115.63097894191742, + 115.84239614009857, + 116.07322025299072, + 116.32566285133362, + 116.69259297847748, + 117.25221645832062, + 117.32926750183105, + 117.41624450683594, + 117.49305152893066, + 117.62031185626984, + 117.74097692966461, + 117.92407608032227, + 118.13507807254791, + 118.49272584915161, + 119.61601066589355, + 123.79435396194458, + 131.1158401966095, + 139.9069287776947 + ], + "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": 44999, + "observation_to_action_latency_ms": { + "distribution_type": "inverse_cdf", + "latency_ms": [ + 98.70396614074707, + 98.93569474406242, + 99.48969086396694, + 99.7413560910225, + 99.99378479576112, + 100.23674687576293, + 100.45678990745544, + 100.61309240221978, + 100.76970933151244, + 100.91663708996772, + 101.04355195236207, + 101.16937319874764, + 101.26514303922653, + 102.1499203300476, + 102.71111395835877, + 103.18615472316742, + 103.5730320930481, + 103.8581521320343, + 104.08799927473068, + 104.26837124824524, + 104.44285610198975, + 104.59245924949646, + 104.72997303962707, + 104.86732120513916, + 104.9985189986229, + 105.11449979782104, + 105.2294695854187, + 105.3437799024582, + 105.46737934827804, + 105.58610138893127, + 105.7231175684929, + 105.86395378112793, + 106.00595810413361, + 106.15865334510804, + 106.3304360961914, + 106.49419790267945, + 106.66770577430725, + 106.87996209144592, + 107.11434404850006, + 107.3905671596527, + 107.76500264644622, + 108.26873979568481, + 109.13974598884583, + 109.82951240062714, + 110.14282467842102, + 110.38825346946716, + 110.55410786867142, + 110.7078928565979, + 110.842807431221, + 110.94358916282654, + 111.0390558052063, + 111.13293657302856, + 111.2161580324173, + 111.30711696624756, + 111.39564727544784, + 111.46318865299224, + 111.532990193367, + 111.60866814613343, + 111.68401275396347, + 111.75892714500428, + 111.83005964517594, + 111.89674973487854, + 111.96217710733414, + 112.03223152160645, + 112.09183992862701, + 112.15503047943115, + 112.21598188877105, + 112.27885273933411, + 112.34791289329529, + 112.4121667098999, + 112.47822159051896, + 112.53936698436738, + 112.59497592687607, + 112.66591523170472, + 112.74125365257264, + 112.81647316932678, + 112.88769792318344, + 112.96064425468445, + 113.03990176916122, + 113.1216405916214, + 113.20431551218033, + 113.29031753540039, + 113.38733742952347, + 113.47987905025482, + 113.5723169875145, + 113.67212121963502, + 113.78512352705002, + 113.90804686069488, + 114.03463499546051, + 114.17505489349365, + 114.33522748231888, + 114.51444413661957, + 114.69055737257004, + 114.87542158126831, + 115.04800778150559, + 115.21496387481689, + 115.40456523895264, + 115.56248826980591, + 115.73442641496658, + 115.90141519069671, + 116.06050712585449, + 116.2129159450531, + 116.38519371986389, + 116.53765635490417, + 116.72071683883667, + 116.93244757652283, + 117.14148443937302, + 117.37877433776856, + 117.63874564647675, + 118.00030923843383, + 118.53773995399476, + 118.61231872749327, + 118.71476497268677, + 118.81297756981849, + 118.9169585442543, + 119.04759360313415, + 119.20824273872375, + 119.41981980085372, + 119.77135396146775, + 120.72904401683795, + 123.87360811496012, + 131.25776838946214, + 135.37514996528625 + ], + "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": 0.9992538390487696, + "worker_service_time_ms": { + "distribution_type": "inverse_cdf", + "latency_ms": [ + 98.31866312026978, + 98.56581710674763, + 99.09732095921039, + 99.34383623552323, + 99.60132966184617, + 99.86176584815979, + 100.06919284248352, + 100.20857570767403, + 100.35520109701157, + 100.51886550021172, + 100.64558459663391, + 100.76037700605393, + 100.85604187965393, + 101.66479192733765, + 102.09853091716766, + 102.38840856075286, + 102.62048434019088, + 102.81856811523437, + 102.98463207960128, + 103.143373336792, + 103.28344969511032, + 103.41374378204345, + 103.54720513105393, + 103.67479965686798, + 103.77949875831604, + 103.88920065879822, + 104.00327035188675, + 104.11408411502838, + 104.23152489423752, + 104.35132833480834, + 104.48386677980423, + 104.61714541912079, + 104.76005975961685, + 104.91307124137879, + 105.07148816108703, + 105.24111230373383, + 105.4121242761612, + 105.6217562198639, + 105.85334130048751, + 106.14065583229065, + 106.52853278398514, + 107.06374320983886, + 107.92898904323577, + 108.54316497802735, + 108.8574897313118, + 109.09838366508484, + 109.25567122697831, + 109.41071063041687, + 109.5345829987526, + 109.63613604545593, + 109.73469749212265, + 109.83290796279907, + 109.91245387554169, + 110.00021874427796, + 110.08120243787765, + 110.15696618080139, + 110.22889561653137, + 110.29889290809632, + 110.37904006004334, + 110.44581078529357, + 110.51936477184296, + 110.58607697486877, + 110.65169381380082, + 110.7204720878601, + 110.77962331056595, + 110.84104709625244, + 110.89811874628067, + 110.96207599163056, + 111.02792620897293, + 111.09100631713868, + 111.15635010957718, + 111.2181872367859, + 111.28041862487792, + 111.34810885429383, + 111.42450404405594, + 111.49753837585449, + 111.57377830743789, + 111.64865688323975, + 111.72656315088273, + 111.80902991771698, + 111.88165298700332, + 111.9731110572815, + 112.06394392251968, + 112.15879735469818, + 112.25163966655731, + 112.35110059738159, + 112.46214687824249, + 112.58521132946015, + 112.71982661724091, + 112.85818910598755, + 113.0256680059433, + 113.19586725234986, + 113.3748426270485, + 113.55558583259582, + 113.73247645378113, + 113.89891651153565, + 114.09933540821075, + 114.25469061851501, + 114.4374224114418, + 114.60077642917634, + 114.75160539627075, + 114.90944337844849, + 115.07355502843858, + 115.2304388999939, + 115.41513421297073, + 115.63058570861817, + 115.84229958057404, + 116.0729815864563, + 116.32491849422455, + 116.69249981880188, + 117.25177101373673, + 117.32898230147362, + 117.41230021953584, + 117.49239917516708, + 117.61818783283233, + 117.73604384064673, + 117.91679067230226, + 118.13051911377906, + 118.48173347473144, + 119.57613303613635, + 122.88888999116595, + 130.4164038644296, + 134.48226380348206 + ], + "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/half-cheetah/small-policy/sample-factory-qwengr00t-rtx3090-v1/profile/latency_profile.png b/latency-aware/half-cheetah/small-policy/sample-factory-qwengr00t-rtx3090-v1/profile/latency_profile.png new file mode 100644 index 0000000000000000000000000000000000000000..fd191b77d7ba009a8681e040f49133af16c9bf36 Binary files /dev/null and b/latency-aware/half-cheetah/small-policy/sample-factory-qwengr00t-rtx3090-v1/profile/latency_profile.png differ diff --git a/latency-aware/half-cheetah/small-policy/sample-factory-qwengr00t-rtx3090-v1/profile/profile.json b/latency-aware/half-cheetah/small-policy/sample-factory-qwengr00t-rtx3090-v1/profile/profile.json new file mode 100644 index 0000000000000000000000000000000000000000..b8fa0a0d9cebea9febc023918a57fb7bd406dbb6 --- /dev/null +++ b/latency-aware/half-cheetah/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": 45000, + "n_capacity_drops": 0, + "n_observation_attempts": 45000, + "per_slot_summary": { + "0": { + "admitted_count": 45000, + "mean_observation_to_action_latency_ms": 110.87737190819317, + "mean_worker_service_time_ms": 109.62982732170953, + "p95_observation_to_action_latency_ms": 117.14155541658401, + "p95_worker_service_time_ms": 115.84231406450272, + "p99_worker_service_time_ms": 117.25199367284775 + } + }, + "provenance": { + "base_config": "/workspace/lzj/latency-sensitive-bench/runs/half_cheetah/gr00t_h1_5fps_20260917/P.yaml", + "checkpoint_kind": "latest", + "model_artifact": { + "action_horizon": 1, + "checkpoint_sha256": "8f007f801c45ed8476546ebf7cc96e60283d2ffd05a6d30feaa529d7ea7f658f", + "clock_fps": 5, + "hf_prefix": "half_cheetah/zero_latency/GR00T", + "hf_repo": "latency-sensitive-bench/Standard-Pipeline", + "hf_revision": "543d6d4ab81112a0307af9f0a5c9cf4d6cd355de", + "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": "20260917T171246114715Z", + "summary": { + "frame_ms": 200.0, + "max_ms": 140.54522109031677, + "mean_effective_frames": 0.5543868595409659, + "mean_ms": 110.87737190819317, + "min_ms": 98.70396614074707, + "n_samples": 45000, + "p50_frames": 0.5594860869646072, + "p50_ms": 111.89721739292145, + "p90_frames": 0.5810652406215668, + "p90_ms": 116.21304812431336, + "p95_frames": 0.5857077770829201, + "p95_ms": 117.14155541658401, + "p99_frames": 0.5926902026772499, + "p99_ms": 118.53804053544998, + "prob_latency_gt_1_frame": 0.0, + "prob_latency_gt_2_frames": 0.0, + "prob_latency_gt_3_frames": 0.0, + "std_ms": 4.364779955431135 + }, + "visualization_path": "latency_profile.png", + "workload_id": "half_cheetah" +} diff --git a/latency-aware/half-cheetah/small-policy/sample-factory-qwengr00t-rtx3090-v1/provenance.json b/latency-aware/half-cheetah/small-policy/sample-factory-qwengr00t-rtx3090-v1/provenance.json new file mode 100644 index 0000000000000000000000000000000000000000..e1bee27eab55f53f2371eb50a99b173d303eb5fa --- /dev/null +++ b/latency-aware/half-cheetah/small-policy/sample-factory-qwengr00t-rtx3090-v1/provenance.json @@ -0,0 +1,44 @@ +{ + "task": "half-cheetah", + "model": "sample-factory-appo", + "training_condition": "latency-aware", + "training_run_id": "half_cheetah_profile_appo_h1_5fps_20260917", + "source": { + "repo_id": "latency-sensitive-bench/Standard-Pipeline", + "revision": "2208875f92b21e3f8ec2363a30a6705dfe56cc04", + "path": "half_cheetah/profile_latency/small_model/GR00T", + "url": "https://huggingface.co/latency-sensitive-bench/Standard-Pipeline/tree/2208875f92b21e3f8ec2363a30a6705dfe56cc04/half_cheetah/profile_latency/small_model/GR00T" + }, + "hf_repo_id": "latency-sensitive-bench/benchmark-models", + "hf_path": "latency-aware/half-cheetah/small-policy/sample-factory-qwengr00t-rtx3090-v1", + "checkpoint": { + "source_file": "half_cheetah/profile_latency/small_model/GR00T/checkpoints/best_000018736_9592832_reward_1793.350.pth", + "source_sha256": "01e50eb4902c9c6cc99879bbd8bc34333921ffc02a7371f092f51b7e8795bd59", + "source_bytes": 83973, + "file": "checkpoint.pth", + "selection_rule": "training_best", + "method": "inference_export", + "sha256": "b1fc0590189724808f40c0d27f902cbc53ce84614c3f606bdceacc8919799c77", + "bytes": 29877, + "train_step": 18736, + "env_steps": 9592832, + "tensor_count": 15, + "tensor_equality_verified": true, + "removed_fields": [ + "best_performance", + "curr_lr", + "optimizer" + ] + }, + "config_source": "half_cheetah/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/half-cheetah/small-policy/sample-factory-qwengr00t-rtx3090-v1/recipe.yaml b/latency-aware/half-cheetah/small-policy/sample-factory-qwengr00t-rtx3090-v1/recipe.yaml new file mode 100644 index 0000000000000000000000000000000000000000..07b993e2ab5a093d45000e27bf1adf0a67c1bc78 --- /dev/null +++ b/latency-aware/half-cheetah/small-policy/sample-factory-qwengr00t-rtx3090-v1/recipe.yaml @@ -0,0 +1,163 @@ +experiment: + name: half_cheetah_profile_appo_h1_5fps_20260917 + seed: 2222 +backend: + type: sample_factory + algo: APPO + device: cuda + train_dir: /mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/half_cheetah/profile_latency/teacher/checkpoints + restart_behavior: overwrite + run_mode: train + extra_args: + - --wandb_user + - dongqianyu99-zhejiang-university +executor: + mode: simulated +env: + name: gymnasium + task_name: half_cheetah + env_id: LatencyBench/HalfCheetahContinuous-v0 + registration_imports: + - latency_bench.envs.gymnasium_half_cheetah + make_kwargs: + base_env_id: HalfCheetah-v4 + render_mode: rgb_array + base_make_kwargs: {} + env_fps: 5 + obs_fps: 5 + frame_stack: 1 + state_space: + labels: + - front_tip_height + - front_tip_angle + - back_thigh_angle + - back_shin_angle + - back_foot_angle + - front_thigh_angle + - front_shin_angle + - front_foot_angle + - front_tip_x_velocity + - front_tip_z_velocity + - front_tip_angular_velocity + - back_thigh_angular_velocity + - back_shin_angular_velocity + - back_foot_angular_velocity + - front_thigh_angular_velocity + - front_shin_angular_velocity + - front_foot_angular_velocity + action_space: + type: box + labels: + - back_thigh_torque + - back_shin_torque + - back_foot_torque + - front_thigh_torque + - front_shin_torque + - front_foot_torque + low: + - -1.0 + - -1.0 + - -1.0 + - -1.0 + - -1.0 + - -1.0 + high: + - 1.0 + - 1.0 + - 1.0 + - 1.0 + - 1.0 + - 1.0 + dtype: float32 + noop_action: + - 0.0 + - 0.0 + - 0.0 + - 0.0 + - 0.0 + - 0.0 + base_prompt: Make the Half Cheetah run forward as fast as possible. Predict six + continuous torques in [-1, 1] ordered as back thigh, back shin, back foot, front + thigh, front shin, and front foot. +latency: + method: iid + profile_path: /mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/half_cheetah/profile_latency/profile_3090_20260917T171246114715Z/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 + 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/half_cheetah/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/half-cheetah/small-policy/sample-factory-qwengr00t-rtx3090-v1/selection.json b/latency-aware/half-cheetah/small-policy/sample-factory-qwengr00t-rtx3090-v1/selection.json new file mode 100644 index 0000000000000000000000000000000000000000..a1452acf581a6debcb744de959335a9a6be9bfc6 --- /dev/null +++ b/latency-aware/half-cheetah/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/half_cheetah/profile_latency/teacher/checkpoints/half_cheetah_profile_appo_h1_5fps_20260917/checkpoint_p0/checkpoint_000019536_10002432.pth", + "sha256": "d99185b8e6af48bdaf4f3a152b7eb391d96c7d8fdbcda6a7d9b9f36f25b8b7b8", + "episodes": 20, + "returns": [ + 1857.2374405048977, + 1867.494728237067, + 1834.5174806519765, + 1865.8690548707461, + 1801.7301131165686, + 1876.8244417767962, + 1865.6403856027441, + 1762.3797913309936, + 1835.7552201531942, + 1842.555192265558, + 1857.8608419027873, + 1828.4981316832227, + 1858.3761893131157, + 1870.122593021503, + 1842.1905916416376, + 1853.096570888896, + 1874.5922690761886, + 1804.164970094345, + 1864.918328697703, + 1799.2772620012308 + ], + "mean": 1843.1550798415585, + "std": 29.76027331862149, + "strict_gt3000": 0 + }, + "training_best": { + "checkpoint": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/half_cheetah/profile_latency/teacher/checkpoints/half_cheetah_profile_appo_h1_5fps_20260917/checkpoint_p0/best_000018736_9592832_reward_1793.350.pth", + "sha256": "01e50eb4902c9c6cc99879bbd8bc34333921ffc02a7371f092f51b7e8795bd59", + "episodes": 20, + "returns": [ + 1887.561249494765, + 1895.0448055201541, + 1858.3344573252318, + 1830.0221069470836, + 1812.0352138530668, + 1793.54399269377, + 1893.5209629203653, + 1835.4972454347915, + 1888.6783450211715, + 1831.1448449836785, + 1855.0970322771736, + 1855.3305649185393, + 1835.5406847810145, + 1845.6326210025886, + 1803.1661607436554, + 1830.418271781488, + 1845.491576391448, + 1854.8196809449396, + 1845.2868341282708, + 1848.1539286138711 + ], + "mean": 1847.2160289888534, + "std": 27.628393653569436, + "strict_gt3000": 0 + } + }, + "completed_at": "2026-09-17T20:08:30.143735+00:00" +} diff --git a/latency-aware/half-cheetah/small-policy/sample-factory-qwengr00t-rtx3090-v1/source/README.md b/latency-aware/half-cheetah/small-policy/sample-factory-qwengr00t-rtx3090-v1/source/README.md new file mode 100644 index 0000000000000000000000000000000000000000..67c5810638d3139cafa401d96675c37d31b8247e --- /dev/null +++ b/latency-aware/half-cheetah/small-policy/sample-factory-qwengr00t-rtx3090-v1/source/README.md @@ -0,0 +1,12 @@ +# HalfCheetah H1 / 5 FPS: verified stage status + +Zero-latency GR00T training completed 5000 updates. L0 simulated evaluation: **5012.472804 ± 742.293683** (population SD), 20 episodes, seeds 42–61, all 1000 raw steps. + +Measured P and B are complete. B20 return: **-262.465910 ± 74.942120**, all 1000 raw steps, zero dropped/invalid actions. B combines 16 completed original episodes whose core metrics were reconstructed with frozen `compute_episode_metrics`, plus four newly evaluated remaining seeds 58–61 after the container restart. P and the completed 16 episodes were not rerun. Some metadata for those 16 episodes, including actual environment FPS and end-drain counters, could not be recovered and remains `None`; this is not lossless full-aggregate metadata recovery. + +The actual-profile APPO teacher completed its original 10M budget (10,002,432 actual environment steps). Final20: 1843.155080 ± 29.760273; training-best20: 1847.216029 ± 27.628394. Both had **0/20 returns strictly greater than 3000**. Training-best was selected, but the scientific gate failed. E10, probe, profile data collection, profile VLA training and F were not run. The budget and gate were not relaxed. Full pipeline acceptance is not established. + +- [Pinned final B and recovery evidence](https://huggingface.co/datasets/latency-sensitive-bench/Standard-Pipeline/tree/223f2106c10b264b820963391ec546d8dce039a5/half_cheetah/profile_latency/GR00T/h1_5fps_20260917/baseline) +- [Pinned teacher gate audit](https://huggingface.co/datasets/latency-sensitive-bench/Standard-Pipeline/tree/6391d8254f2b9d38d6f4a8c5d5168cfd5f10e30a/half_cheetah/profile_latency/GR00T/h1_5fps_20260917/small_model/evaluation) + +Existing inference weights and historical validation receipts are unchanged. diff --git a/latency-aware/half-cheetah/small-policy/sample-factory-qwengr00t-rtx3090-v1/teacher-verification.json b/latency-aware/half-cheetah/small-policy/sample-factory-qwengr00t-rtx3090-v1/teacher-verification.json new file mode 100644 index 0000000000000000000000000000000000000000..2c6b078937f759f8d63647dd88290f5f37da7b05 --- /dev/null +++ b/latency-aware/half-cheetah/small-policy/sample-factory-qwengr00t-rtx3090-v1/teacher-verification.json @@ -0,0 +1,132 @@ +{ + "state": "TEACHER_GATE_FAILED_VERIFIED", + "verified_at": "2026-09-17T20:16:11.793455+00:00", + "seed": 2222, + "target_env_steps": 10000000, + "actual_final_env_steps": 10002432, + "state_dim": 17, + "action_dim": 6, + "env_fps": 5, + "obs_fps": 5, + "latency": "actualprofileIIDseed0", + "profile_sha256": "bec664bd742b851acad96d46b8d0ecde5447f26097c4bea79ae839bba440037f", + "selected": "training_best", + "evaluations": { + "final": { + "mean": 1843.1550798415585, + "population_std": 29.76027331862149, + "episodes": 20, + "seeds": [ + 104771, + 104772, + 104773, + 104774, + 104775, + 104776, + 104777, + 104778, + 104779, + 104780, + 104781, + 104782, + 104783, + 104784, + 104785, + 104786, + 104787, + 104788, + 104789, + 104790 + ], + "lengths": [ + 1000, + 1000, + 1000, + 1000, + 1000, + 1000, + 1000, + 1000, + 1000, + 1000, + 1000, + 1000, + 1000, + 1000, + 1000, + 1000, + 1000, + 1000, + 1000, + 1000 + ], + "strict_gt3000": 0, + "invalid_actions": 0, + "raw_step_rewards_and_indices_verified": true, + "checkpoint_sha256": "d99185b8e6af48bdaf4f3a152b7eb391d96c7d8fdbcda6a7d9b9f36f25b8b7b8" + }, + "training_best": { + "mean": 1847.2160289888534, + "population_std": 27.628393653569436, + "episodes": 20, + "seeds": [ + 104771, + 104772, + 104773, + 104774, + 104775, + 104776, + 104777, + 104778, + 104779, + 104780, + 104781, + 104782, + 104783, + 104784, + 104785, + 104786, + 104787, + 104788, + 104789, + 104790 + ], + "lengths": [ + 1000, + 1000, + 1000, + 1000, + 1000, + 1000, + 1000, + 1000, + 1000, + 1000, + 1000, + 1000, + 1000, + 1000, + 1000, + 1000, + 1000, + 1000, + 1000, + 1000 + ], + "strict_gt3000": 0, + "invalid_actions": 0, + "raw_step_rewards_and_indices_verified": true, + "checkpoint_sha256": "01e50eb4902c9c6cc99879bbd8bc34333921ffc02a7371f092f51b7e8795bd59" + } + }, + "strict_gate": ">3000", + "not_run": [ + "E10", + "probe", + "data_collection", + "profile_VLA", + "F" + ], + "B": "Still running as reported by parent; not inspected by this audit", + "budget_or_gate_changed": false +} diff --git a/latency-aware/half-cheetah/small-policy/sample-factory-qwenoft-rtx3090-v1/README.md b/latency-aware/half-cheetah/small-policy/sample-factory-qwenoft-rtx3090-v1/README.md new file mode 100644 index 0000000000000000000000000000000000000000..078cdd93f759e7d544135cd619230bf044212735 --- /dev/null +++ b/latency-aware/half-cheetah/small-policy/sample-factory-qwenoft-rtx3090-v1/README.md @@ -0,0 +1,26 @@ +# half-cheetah / sample-factory-appo + +Training condition: `latency-aware`. Run: `half_cheetah_profile_20260912T023319Z`. + +Copied from [Standard-Pipeline](https://huggingface.co/latency-sensitive-bench/Standard-Pipeline/tree/2208875f92b21e3f8ec2363a30a6705dfe56cc04/half_cheetah/profile_latency/small_model) at `2208875f92b21e3f8ec2363a30a6705dfe56cc04`. + +- Checkpoint: `checkpoint.pth` +- Selection: best_reward_after_full_training_budget +- Checkpoint SHA256: `323a1b8a6b7050ad692bf684645ec03e518fb0f2b20cbe3ec98fb92df0ceb640` +- 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/half-cheetah/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/half-cheetah/small-policy/sample-factory-qwenoft-rtx3090-v1/checkpoint.pth b/latency-aware/half-cheetah/small-policy/sample-factory-qwenoft-rtx3090-v1/checkpoint.pth new file mode 100644 index 0000000000000000000000000000000000000000..ba35d976949b609637038b187cbc8d90234e4b06 --- /dev/null +++ b/latency-aware/half-cheetah/small-policy/sample-factory-qwenoft-rtx3090-v1/checkpoint.pth @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:323a1b8a6b7050ad692bf684645ec03e518fb0f2b20cbe3ec98fb92df0ceb640 +size 29877 diff --git a/latency-aware/half-cheetah/small-policy/sample-factory-qwenoft-rtx3090-v1/config.json b/latency-aware/half-cheetah/small-policy/sample-factory-qwenoft-rtx3090-v1/config.json new file mode 100644 index 0000000000000000000000000000000000000000..0f828f22b67208f701257009561f8e26ed83fca9 --- /dev/null +++ b/latency-aware/half-cheetah/small-policy/sample-factory-qwenoft-rtx3090-v1/config.json @@ -0,0 +1,276 @@ +{ + "help": false, + "algo": "APPO", + "env": "latency_gymnasium", + "experiment": "half_cheetah_profile_20260912T023319Z", + "train_dir": "/mnt/results/latency-sensitive-bench/profile-latency-new/half_cheetah/20260912T023319Z-halfcheetah/small_models", + "restart_behavior": "overwrite", + "device": "gpu", + "seed": 2222, + "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": 1, + "pixel_format": "CHW", + "use_record_episode_statistics": false, + "with_wandb": false, + "wandb_user": null, + "wandb_project": "sample_factory", + "wandb_group": null, + "wandb_job_type": "SF", + "wandb_tags": [], + "with_pbt": false, + "pbt_mix_policies_in_one_env": true, + "pbt_period_env_steps": 5000000, + "pbt_start_mutation": 20000000, + "pbt_replace_fraction": 0.3, + "pbt_mutation_rate": 0.15, + "pbt_replace_reward_gap": 0.1, + "pbt_replace_reward_gap_absolute": 1e-06, + "pbt_optimize_gamma": false, + "pbt_target_objective": "true_objective", + "pbt_perturb_min": 1.1, + "pbt_perturb_max": 1.5, + "gym_task_name": "half_cheetah_rgb_state", + "gym_env_id": "LatencyBench/HalfCheetahContinuous-v0", + "gym_make_kwargs_json": "{\"base_env_id\": \"HalfCheetah-v4\", \"base_make_kwargs\": {}, \"render_mode\": \"rgb_array\"}", + "gym_registration_imports_json": "[\"latency_bench.envs.gymnasium_half_cheetah\"]", + "gym_action_space_json": "{\"dtype\": \"float32\", \"high\": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0], \"labels\": [\"back_thigh_torque\", \"back_shin_torque\", \"back_foot_torque\", \"front_thigh_torque\", \"front_shin_torque\", \"front_foot_torque\"], \"low\": [-1.0, -1.0, -1.0, -1.0, -1.0, -1.0], \"type\": \"box\"}", + "gym_action_labels_json": null, + "gym_action_values_json": null, + "gym_noop_action_id": null, + "gym_noop_action_json": "[0.0, 0.0, 0.0, 0.0, 0.0, 0.0]", + "gym_base_prompt": "Make the Half Cheetah run forward as fast as possible. Predict six continuous torques in [-1, 1] ordered as back thigh, back shin, back foot, front thigh, front shin, and front foot.", + "gym_state_labels_json": "[\"front_tip_height\", \"front_tip_angle\", \"back_thigh_angle\", \"back_shin_angle\", \"back_foot_angle\", \"front_thigh_angle\", \"front_shin_angle\", \"front_foot_angle\", \"front_tip_x_velocity\", \"front_tip_z_velocity\", \"front_tip_angular_velocity\", \"back_thigh_angular_velocity\", \"back_shin_angular_velocity\", \"back_foot_angular_velocity\", \"front_thigh_angular_velocity\", \"front_shin_angular_velocity\", \"front_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/half_cheetah/20260912T023319Z-halfcheetah/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/half_cheetah/20260912T023319Z-halfcheetah/small_models/half_cheetah_profile_20260912T023319Z/episode_metrics.jsonl", + "command_line": "--mode train --algo APPO --env latency_gymnasium --experiment half_cheetah_profile_20260912T023319Z --train_dir /mnt/results/latency-sensitive-bench/profile-latency-new/half_cheetah/20260912T023319Z-halfcheetah/small_models --restart_behavior overwrite --device gpu --seed 2222 --episode_metrics_path /mnt/results/latency-sensitive-bench/profile-latency-new/half_cheetah/20260912T023319Z-halfcheetah/small_models/half_cheetah_profile_20260912T023319Z/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 1 --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 --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 --encoder_mlp_layers 64 64 --latency-type iid --latency-seed 0 --latency-profile-path /mnt/results/latency-sensitive-bench/profile-latency-new/half_cheetah/20260912T023319Z-halfcheetah/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 half_cheetah_rgb_state --gym-env-id LatencyBench/HalfCheetahContinuous-v0 --gym-make-kwargs-json {\"base_env_id\": \"HalfCheetah-v4\", \"base_make_kwargs\": {}, \"render_mode\": \"rgb_array\"} --gym-registration-imports-json [\"latency_bench.envs.gymnasium_half_cheetah\"] --gym-action-space-json {\"dtype\": \"float32\", \"high\": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0], \"labels\": [\"back_thigh_torque\", \"back_shin_torque\", \"back_foot_torque\", \"front_thigh_torque\", \"front_shin_torque\", \"front_foot_torque\"], \"low\": [-1.0, -1.0, -1.0, -1.0, -1.0, -1.0], \"type\": \"box\"} --gym-noop-action-json [0.0, 0.0, 0.0, 0.0, 0.0, 0.0] --gym-base-prompt Make the Half Cheetah run forward as fast as possible. Predict six continuous torques in [-1, 1] ordered as back thigh, back shin, back foot, front thigh, front shin, and front foot. --gym-state-labels-json [\"front_tip_height\", \"front_tip_angle\", \"back_thigh_angle\", \"back_shin_angle\", \"back_foot_angle\", \"front_thigh_angle\", \"front_shin_angle\", \"front_foot_angle\", \"front_tip_x_velocity\", \"front_tip_z_velocity\", \"front_tip_angular_velocity\", \"back_thigh_angular_velocity\", \"back_shin_angular_velocity\", \"back_foot_angular_velocity\", \"front_thigh_angular_velocity\", \"front_shin_angular_velocity\", \"front_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": "half_cheetah_profile_20260912T023319Z", + "train_dir": "/mnt/results/latency-sensitive-bench/profile-latency-new/half_cheetah/20260912T023319Z-halfcheetah/small_models", + "restart_behavior": "overwrite", + "device": "gpu", + "seed": 2222, + "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, + "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": 1, + "with_wandb": false, + "gym_task_name": "half_cheetah_rgb_state", + "gym_env_id": "LatencyBench/HalfCheetahContinuous-v0", + "gym_make_kwargs_json": "{\"base_env_id\": \"HalfCheetah-v4\", \"base_make_kwargs\": {}, \"render_mode\": \"rgb_array\"}", + "gym_registration_imports_json": "[\"latency_bench.envs.gymnasium_half_cheetah\"]", + "gym_action_space_json": "{\"dtype\": \"float32\", \"high\": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0], \"labels\": [\"back_thigh_torque\", \"back_shin_torque\", \"back_foot_torque\", \"front_thigh_torque\", \"front_shin_torque\", \"front_foot_torque\"], \"low\": [-1.0, -1.0, -1.0, -1.0, -1.0, -1.0], \"type\": \"box\"}", + "gym_noop_action_json": "[0.0, 0.0, 0.0, 0.0, 0.0, 0.0]", + "gym_base_prompt": "Make the Half Cheetah run forward as fast as possible. Predict six continuous torques in [-1, 1] ordered as back thigh, back shin, back foot, front thigh, front shin, and front foot.", + "gym_state_labels_json": "[\"front_tip_height\", \"front_tip_angle\", \"back_thigh_angle\", \"back_shin_angle\", \"back_foot_angle\", \"front_thigh_angle\", \"front_shin_angle\", \"front_foot_angle\", \"front_tip_x_velocity\", \"front_tip_z_velocity\", \"front_tip_angular_velocity\", \"back_thigh_angular_velocity\", \"back_shin_angular_velocity\", \"back_foot_angular_velocity\", \"front_thigh_angular_velocity\", \"front_shin_angular_velocity\", \"front_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/half_cheetah/20260912T023319Z-halfcheetah/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/half_cheetah/20260912T023319Z-halfcheetah/small_models/half_cheetah_profile_20260912T023319Z/episode_metrics.jsonl", + "output_dir": "/mnt/results/latency-sensitive-bench/profile-latency-new/half_cheetah/20260912T023319Z-halfcheetah/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/half_cheetah/20260912T023319Z-halfcheetah/small_train" +} \ No newline at end of file diff --git a/latency-aware/half-cheetah/small-policy/sample-factory-qwenoft-rtx3090-v1/profile/DONE b/latency-aware/half-cheetah/small-policy/sample-factory-qwenoft-rtx3090-v1/profile/DONE new file mode 100644 index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391 diff --git a/latency-aware/half-cheetah/small-policy/sample-factory-qwenoft-rtx3090-v1/profile/hardware.json b/latency-aware/half-cheetah/small-policy/sample-factory-qwenoft-rtx3090-v1/profile/hardware.json new file mode 100644 index 0000000000000000000000000000000000000000..e8204e81807044975c0e0a0953d16d6835192f31 --- /dev/null +++ b/latency-aware/half-cheetah/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/half-cheetah/small-policy/sample-factory-qwenoft-rtx3090-v1/profile/latency_burst_model.json b/latency-aware/half-cheetah/small-policy/sample-factory-qwenoft-rtx3090-v1/profile/latency_burst_model.json new file mode 100644 index 0000000000000000000000000000000000000000..aa5e6d42f5c6a718266afa349f01bef857224531 --- /dev/null +++ b/latency-aware/half-cheetah/small-policy/sample-factory-qwenoft-rtx3090-v1/profile/latency_burst_model.json @@ -0,0 +1,1329 @@ +{ + "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, + 1 + ], + "burst_merge_gap_records": 30, + "burst_rank_processes": [ + { + "draw_count": 4, + "dwell_length_spearman_rho": 0.0, + "level_ms": { + "distribution_type": "inverse_cdf", + "latency_ms": [ + 91.29252696037292, + 91.29252696037292, + 91.29252696037292, + 91.29252696037292, + 91.29252696037292, + 91.29252696037292, + 91.29252696037292, + 91.29252696037292, + 91.29252696037292, + 91.29252696037292, + 91.29252696037292, + 91.29252696037292, + 91.29252696037292, + 91.29252696037292, + 91.29252696037292, + 91.29252696037292, + 91.29252696037292, + 91.29252696037292, + 91.29252696037292, + 91.29252696037292, + 91.29252696037292, + 91.29252696037292, + 91.29252696037292, + 91.29252696037292, + 91.48080760002136, + 91.85736887931824, + 92.23393015861511, + 92.61049143791199, + 92.98705271720887, + 93.36361399650573, + 93.74017527580261, + 94.1167365550995, + 94.49329783439636, + 94.86985911369324, + 95.24642039299012, + 95.62298167228698, + 95.99954295158386, + 96.37610423088074, + 96.75266551017761, + 97.12922678947449, + 97.50578806877137, + 97.88234934806823, + 98.25891062736511, + 98.63547190666199, + 99.01203318595887, + 99.38859446525574, + 99.76515574455262, + 100.14171702384948, + 100.51827830314636, + 100.78247780323028, + 100.93431552410125, + 101.08615324497222, + 101.2379909658432, + 101.38982868671417, + 101.54166640758514, + 101.69350412845611, + 101.84534184932708, + 101.99717957019806, + 102.14901729106903, + 102.30085501194, + 102.45269273281097, + 102.60453045368195, + 102.75636817455292, + 102.90820589542389, + 103.06004361629486, + 103.21188133716583, + 103.3637190580368, + 103.51555677890778, + 103.66739449977875, + 103.81923222064972, + 103.9710699415207, + 104.12290766239167, + 104.27474538326264, + 104.42658310413361, + 104.64846988677978, + 104.94040573120117, + 105.23234157562256, + 105.52427742004394, + 105.81621326446533, + 106.10814910888672, + 106.40008495330811, + 106.6920207977295, + 106.98395664215087, + 107.27589248657226, + 107.56782833099365, + 107.85976417541504, + 108.15170001983643, + 108.44363586425781, + 108.7355717086792, + 109.02750755310059, + 109.31944339752198, + 109.61137924194335, + 109.90331508636474, + 110.19525093078613, + 110.48718677520752, + 110.77912261962891, + 111.0710584640503, + 111.36299430847168, + 111.65493015289307, + 111.80089807510376, + 111.80089807510376, + 111.80089807510376, + 111.80089807510376, + 111.80089807510376, + 111.80089807510376, + 111.80089807510376, + 111.80089807510376, + 111.80089807510376, + 111.80089807510376, + 111.80089807510376, + 111.80089807510376, + 111.80089807510376, + 111.80089807510376, + 111.80089807510376, + 111.80089807510376, + 111.80089807510376, + 111.80089807510376, + 111.80089807510376, + 111.80089807510376, + 111.80089807510376, + 111.80089807510376, + 111.80089807510376, + 111.80089807510376 + ], + "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": 4 + } + ], + "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 + } + }, + { + "dwell_length": 1, + "slot_to_rank": { + "0": 0 + } + } + ], + "model_type": "hidden_regime", + "pre_worker_time_ms": { + "distribution_type": "inverse_cdf", + "latency_ms": [ + 0.28966736793518066, + 0.29661628777980803, + 0.30831514048576353, + 0.31727498722076414, + 0.338593578338623, + 0.3461133544445038, + 0.3511757125854492, + 0.3563822603225708, + 0.3618279318809509, + 0.3671014323234558, + 0.3725587911605835, + 0.3779346108436584, + 0.38324137926101687, + 0.4556529235839844, + 0.509889326095581, + 0.5281284952163696, + 0.5403126955032349, + 0.5479220390319824, + 0.5539253664016723, + 0.5594458866119385, + 0.5640100693702698, + 0.5677289962768555, + 0.5709687566757202, + 0.5738528633117675, + 0.5767292833328247, + 0.5795547199249268, + 0.5822810411453248, + 0.5850470447540284, + 0.5876750636100769, + 0.5903967952728272, + 0.592800943851471, + 0.5953300476074219, + 0.5980806803703308, + 0.6004024124145508, + 0.6025746560096741, + 0.6047077465057373, + 0.6067898273468018, + 0.6086850166320801, + 0.6106685948371887, + 0.612722282409668, + 0.6146113967895508, + 0.6165520191192627, + 0.6182403135299682, + 0.6199717569351196, + 0.6218879985809326, + 0.6237829494476318, + 0.6255278944969177, + 0.6274885368347168, + 0.6292352557182312, + 0.6305854225158691, + 0.6320900011062622, + 0.6335686922073365, + 0.6350811290740966, + 0.6365709590911866, + 0.6379385328292847, + 0.6395668697357177, + 0.640962290763855, + 0.642304630279541, + 0.6439359402656555, + 0.645436725616455, + 0.6472484803199768, + 0.6488471031188965, + 0.6504889988899231, + 0.6521338987350463, + 0.6537751984596253, + 0.655202226638794, + 0.6567079544067383, + 0.6582039451599122, + 0.6598467111587524, + 0.6612679958343506, + 0.6625553750991822, + 0.6639010190963746, + 0.6652205157279968, + 0.6667079925537109, + 0.6680989265441895, + 0.6693964767456054, + 0.6709147572517395, + 0.6722634792327881, + 0.6739575719833374, + 0.6756732177734375, + 0.6773415374755859, + 0.6790953636169433, + 0.6808587408065796, + 0.6830189752578736, + 0.6850442719459534, + 0.687199592590332, + 0.68963623046875, + 0.6920534992218017, + 0.6945298147201538, + 0.6971553516387939, + 0.6997041010856628, + 0.7023355007171631, + 0.7050988674163818, + 0.7076102638244629, + 0.7099640583992004, + 0.7127210712432861, + 0.7153138279914856, + 0.7181663990020751, + 0.7208955335617065, + 0.7237677955627442, + 0.7270624494552612, + 0.7306365489959716, + 0.7349855732917786, + 0.739589581489563, + 0.7454291129112245, + 0.7518837356567382, + 0.7594980955123901, + 0.7684542274475097, + 0.7794348359107971, + 0.799935255050659, + 0.8361522078514098, + 0.841301315307617, + 0.8511774196624763, + 0.858581780910492, + 0.8691914596557617, + 0.8781117153167721, + 0.8919882164001455, + 0.9130556282997128, + 0.9367391667366031, + 0.9679305844306949, + 1.0121961189508684, + 1.2404262654781233, + 1.388310194015503 + ], + "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": 4, + "calm": 29993 + }, + "regime_transition_counts": { + "burst": { + "burst": 0, + "calm": 4 + }, + "calm": { + "burst": 4, + "calm": 29984 + } + }, + "reset_scope": "session", + "schema_version": 12, + "spike_median_multiplier": 1.25, + "spike_threshold_ms_by_worker_slot": { + "0": 88.3233892917633 + }, + "worker_count": 1 +} diff --git a/latency-aware/half-cheetah/small-policy/sample-factory-qwenoft-rtx3090-v1/profile/latency_distribution.json b/latency-aware/half-cheetah/small-policy/sample-factory-qwenoft-rtx3090-v1/profile/latency_distribution.json new file mode 100644 index 0000000000000000000000000000000000000000..b5dc4b58602cca48b6383606d7ca634ea28cec4e --- /dev/null +++ b/latency-aware/half-cheetah/small-policy/sample-factory-qwenoft-rtx3090-v1/profile/latency_distribution.json @@ -0,0 +1,1027 @@ +{ + "schema_version": 3, + "worker_slots": { + "0": { + "all": { + "count": 29997, + "observation_to_action_latency_ms": { + "distribution_type": "inverse_cdf", + "latency_ms": [ + 65.39556503295898, + 65.90319257864952, + 66.34795825266838, + 66.52524732637406, + 66.76708299684525, + 67.02418623113633, + 67.17869547843934, + 67.26114100694656, + 67.40597652959823, + 67.50079720020294, + 67.60124465465546, + 67.68513790631295, + 67.77476126432418, + 69.4245816898346, + 70.02091380119323, + 70.28190787315368, + 70.46180428266526, + 70.5905613231659, + 70.6786233663559, + 70.76199055194854, + 70.82302750349045, + 70.87952952384948, + 70.93279217720031, + 70.97876193523408, + 71.02215361356735, + 71.05534663200379, + 71.08903501033782, + 71.12099472522736, + 71.14889664411545, + 71.17835633277893, + 71.20590738773346, + 71.23764076232911, + 71.26494863510132, + 71.29445910453796, + 71.31955314636231, + 71.3448329114914, + 71.370949447155, + 71.39764987945557, + 71.42063557147979, + 71.44203779697418, + 71.46336093187332, + 71.48176407814026, + 71.50155672311783, + 71.52068632602692, + 71.53738250017166, + 71.55426707267762, + 71.5741131901741, + 71.59390948295594, + 71.61351494789123, + 71.63126242637634, + 71.65004153251648, + 71.6698139667511, + 71.68906754732132, + 71.7093546485901, + 71.72822314500809, + 71.74877698421479, + 71.77002807855607, + 71.79165102005005, + 71.81265570163727, + 71.83448543548585, + 71.85390951871872, + 71.872318983078, + 71.88918613672257, + 71.90831651687623, + 71.92598719120025, + 71.94410138130188, + 71.96174719333649, + 71.97731454849243, + 71.99444168806076, + 72.01067091941833, + 72.02784201383591, + 72.04347264766693, + 72.06015002250672, + 72.07703278541565, + 72.09355116605758, + 72.11092311382293, + 72.12674032449722, + 72.14582781791687, + 72.1675558257103, + 72.18711300849914, + 72.20661722660064, + 72.22612509727477, + 72.24460446119309, + 72.26352282524108, + 72.28893879890443, + 72.3113342666626, + 72.33358442783356, + 72.35914145946502, + 72.38095099925995, + 72.40431155204773, + 72.42851207256317, + 72.45446176528931, + 72.48137830018997, + 72.5092474937439, + 72.53562726974488, + 72.56389167308808, + 72.59069626331329, + 72.62336647033692, + 72.65377742528915, + 72.68993539810181, + 72.72776058673858, + 72.770436668396, + 72.82028721570968, + 72.8761532688141, + 72.93663378715515, + 73.01012805938721, + 73.0942009806633, + 73.18615770816803, + 73.34689612150193, + 73.59420322418212, + 74.26737931489944, + 74.39098821926116, + 74.46888185977936, + 74.59232448458671, + 74.7213032426834, + 74.85509558796882, + 75.02051337623595, + 75.32228709888452, + 75.84211740064622, + 78.21195779466629, + 80.96420217287572, + 96.67231945006563, + 112.48595118522644 + ], + "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": 0.9963489496805269, + "worker_service_time_ms": { + "distribution_type": "inverse_cdf", + "latency_ms": [ + 65.03665804862976, + 65.55401456711293, + 65.98787849330903, + 66.16043975687028, + 66.39510622596741, + 66.65682306718826, + 66.81381133651733, + 66.91495643734932, + 67.01536854743958, + 67.10630764722823, + 67.22047565555573, + 67.31252860784531, + 67.39065796613693, + 68.88474659919738, + 69.45113959550858, + 69.69205457687377, + 69.84725253582, + 69.97218856811523, + 70.05851617097855, + 70.12988825798034, + 70.19403784513473, + 70.24813051223755, + 70.29781345844269, + 70.34338568687438, + 70.38068360805511, + 70.41919645309449, + 70.44898892641068, + 70.4820789384842, + 70.50974248409271, + 70.53648761749268, + 70.5667546749115, + 70.59351196289063, + 70.61864386320114, + 70.64630803108216, + 70.6727023935318, + 70.69872972011567, + 70.72493976354599, + 70.7476874256134, + 70.77046714782715, + 70.79256957054139, + 70.81106184720993, + 70.83241534233093, + 70.8515542602539, + 70.86944600582123, + 70.89099317789078, + 70.90784223556518, + 70.926378428936, + 70.94538230895996, + 70.96381083726882, + 70.98396788597107, + 71.00406824111938, + 71.02347581386566, + 71.04286310195923, + 71.06384484291077, + 71.08473001003266, + 71.10399283409119, + 71.12429804801941, + 71.14437225341797, + 71.1641768026352, + 71.1816377735138, + 71.20087228775024, + 71.21854376792908, + 71.23626034259796, + 71.25274047851562, + 71.27102743148804, + 71.28886438369751, + 71.30608870983124, + 71.32378898143769, + 71.33962363481521, + 71.35465386390686, + 71.371551156044, + 71.3889540195465, + 71.40592097759247, + 71.42115729331971, + 71.43740501642228, + 71.45302256584168, + 71.47052863836288, + 71.4882806968689, + 71.50748840570449, + 71.52952627658844, + 71.55059865713119, + 71.57056875228882, + 71.58898196697236, + 71.60883347988128, + 71.6305475282669, + 71.65208287239075, + 71.67324340343475, + 71.69656970024108, + 71.71778207540513, + 71.7423820400238, + 71.76660402297973, + 71.79135773181915, + 71.8179384636879, + 71.84612865447998, + 71.87385843992233, + 71.90183074951172, + 71.92967116832733, + 71.96011421203613, + 71.99151601076126, + 72.02691771984101, + 72.06483766317368, + 72.10874972343444, + 72.15644047737122, + 72.21186874389649, + 72.2747768831253, + 72.34141691207886, + 72.41462646722793, + 72.5150628232956, + 72.65761122465133, + 72.89338689804077, + 73.48060830593109, + 73.56621590280531, + 73.65627765464784, + 73.81442364430428, + 73.93480096769333, + 74.05676769256591, + 74.22994810676575, + 74.53286233878123, + 75.0761570343972, + 77.40913844251634, + 80.2655718396903, + 96.00236716117574, + 111.80089807510376 + ], + "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": 29993, + "observation_to_action_latency_ms": { + "distribution_type": "inverse_cdf", + "latency_ms": [ + 65.39556503295898, + 65.90317262949944, + 66.34794480299949, + 66.52523650598526, + 66.76706639719009, + 67.02408304548264, + 67.17864943504334, + 67.26108434915542, + 67.40594648313522, + 67.5005348443985, + 67.6011360502243, + 67.68505244469642, + 67.77472558736801, + 69.4243199634552, + 70.02058695793151, + 70.28187300682067, + 70.46149944067001, + 70.59055726051331, + 70.67848117351532, + 70.76189371585846, + 70.82301428794861, + 70.87942266464233, + 70.93270378589631, + 70.97858119487762, + 71.02201527357101, + 71.05530756950378, + 71.0888930439949, + 71.12097437858581, + 71.14887702703476, + 71.17826912879944, + 71.2058483171463, + 71.23736724853515, + 71.26493341445922, + 71.2944363117218, + 71.31952383756638, + 71.34473480701446, + 71.3709145784378, + 71.39759300231934, + 71.42056230068206, + 71.44201333999634, + 71.46333529710769, + 71.4817370414734, + 71.50154428243637, + 71.52057462692261, + 71.5372909116745, + 71.55419871330261, + 71.5740326166153, + 71.59386352539063, + 71.61349762916565, + 71.6311550617218, + 71.64978809833526, + 71.6697144985199, + 71.68892177343369, + 71.70919153213501, + 71.72815980434417, + 71.74859975814819, + 71.7698770403862, + 71.79157061576844, + 71.81258868217468, + 71.83437560558319, + 71.8536979842186, + 71.87217211723328, + 71.88915539979935, + 71.90820856571197, + 71.92582542896271, + 71.94395114898681, + 71.96166801452637, + 71.97716438293457, + 71.99436103343963, + 72.01059600830078, + 72.02775737524033, + 72.04330060482025, + 72.0600642299652, + 72.07685089111328, + 72.09341137647628, + 72.11067230224609, + 72.12661106586457, + 72.14578086853027, + 72.16744079828263, + 72.18705845355987, + 72.20654620409012, + 72.22568397521972, + 72.2445086312294, + 72.26336144447326, + 72.28845200538635, + 72.3110949420929, + 72.33329623937607, + 72.35886358737946, + 72.38054457426071, + 72.40417922973633, + 72.42825816869735, + 72.45425968170166, + 72.48099433660508, + 72.50895282745361, + 72.53547874927521, + 72.56302463531495, + 72.59042900800705, + 72.62289143562317, + 72.65363789081573, + 72.68952176094055, + 72.72733676195145, + 72.76983900070191, + 72.81949627876281, + 72.87578072071075, + 72.93594701766968, + 73.00869238853454, + 73.09237577915192, + 73.1850743484497, + 73.34378555059433, + 73.58954241752625, + 74.25404654741287, + 74.377984126091, + 74.45795891284942, + 74.56692840909957, + 74.68882406949997, + 74.83917311906815, + 74.99441411018371, + 75.2778499314785, + 75.79855277395247, + 77.6527411787511, + 79.8600849787, + 83.74479532060701, + 87.96991801261902 + ], + "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": 0.9963474887247316, + "worker_service_time_ms": { + "distribution_type": "inverse_cdf", + "latency_ms": [ + 65.03665804862976, + 65.55398055746555, + 65.9878485212326, + 66.16043788576125, + 66.39510260963439, + 66.65668116617202, + 66.81379850769044, + 66.91483943581581, + 67.0153457736969, + 67.10624616384506, + 67.22031239414216, + 67.31242878675461, + 67.39064136266708, + 68.884574842453, + 69.4511189675331, + 69.69204290390014, + 69.84723336696625, + 69.97216024398804, + 70.05845548868179, + 70.1298413324356, + 70.19400545597077, + 70.24808297157287, + 70.29777021169663, + 70.34335553169251, + 70.3806136393547, + 70.41918155670166, + 70.448968064785, + 70.48199189662934, + 70.50970908641816, + 70.53647972106934, + 70.56663146018983, + 70.59342155456542, + 70.61852590322495, + 70.64628098487854, + 70.67265600442886, + 70.69867686271668, + 70.72488385438919, + 70.7476558971405, + 70.7703990483284, + 70.79254083633423, + 70.81092375040055, + 70.83221793174744, + 70.85147241830826, + 70.8694316148758, + 70.89091994285583, + 70.9078179550171, + 70.92635341882706, + 70.94532262325286, + 70.9637482881546, + 70.98387693405151, + 71.00404205799103, + 71.02344303131103, + 71.04280935287476, + 71.06381531715392, + 71.08464890480042, + 71.10389989852905, + 71.12422089576721, + 71.14431077003479, + 71.16399006843567, + 71.18156445026398, + 71.2007286643982, + 71.21847820281982, + 71.23595291376114, + 71.25258999824524, + 71.27089961290359, + 71.28884199142456, + 71.3059007525444, + 71.32369264602661, + 71.33935357570648, + 71.35447284698486, + 71.37133013010025, + 71.38878374099731, + 71.40563994407654, + 71.42104347229004, + 71.43730006217956, + 71.4528238105774, + 71.47043471336364, + 71.48799346923828, + 71.50737279176712, + 71.52930597782135, + 71.55040246009827, + 71.57041568756104, + 71.58885224580764, + 71.60867823123932, + 71.63035848617554, + 71.65184557914733, + 71.67304146289825, + 71.69649241924286, + 71.71739944934845, + 71.74205615997315, + 71.76628437042237, + 71.79092388153076, + 71.81759873390197, + 71.84595065116882, + 71.87376455783844, + 71.9015589761734, + 71.92951682806014, + 71.95996829032897, + 71.99123099327088, + 72.02651390075684, + 72.06419330596924, + 72.10861415863037, + 72.15558389186859, + 72.21108134746551, + 72.27408450841904, + 72.34028102874755, + 72.4134205698967, + 72.51264576911926, + 72.65399513483048, + 72.89117924690247, + 73.4721660399437, + 73.55530694293975, + 73.64461615943908, + 73.80222330904007, + 73.91535479927065, + 74.04386379599572, + 74.20484306907652, + 74.46073522233964, + 75.0037772812843, + 76.84873060560238, + 79.33595724523077, + 82.89885435822055, + 87.28808784484863 + ], + "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/half-cheetah/small-policy/sample-factory-qwenoft-rtx3090-v1/profile/latency_profile.png b/latency-aware/half-cheetah/small-policy/sample-factory-qwenoft-rtx3090-v1/profile/latency_profile.png new file mode 100644 index 0000000000000000000000000000000000000000..935d1345a0421a6f634df4db3f9010ec284e1bd4 Binary files /dev/null and b/latency-aware/half-cheetah/small-policy/sample-factory-qwenoft-rtx3090-v1/profile/latency_profile.png differ diff --git a/latency-aware/half-cheetah/small-policy/sample-factory-qwenoft-rtx3090-v1/profile/profile.json b/latency-aware/half-cheetah/small-policy/sample-factory-qwenoft-rtx3090-v1/profile/profile.json new file mode 100644 index 0000000000000000000000000000000000000000..7dc2634bd101f638e0549818d7b64c92bb028e57 --- /dev/null +++ b/latency-aware/half-cheetah/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": 29997, + "n_capacity_drops": 3, + "n_observation_attempts": 30000, + "per_slot_summary": { + "0": { + "admitted_count": 29997, + "mean_observation_to_action_latency_ms": 71.8201196935092, + "mean_worker_service_time_ms": 71.17275555820454, + "p95_observation_to_action_latency_ms": 73.09400281906129, + "p95_worker_service_time_ms": 72.4146143436432, + "p99_worker_service_time_ms": 73.4799683380127 + } + }, + "provenance": { + "base_config": "/workspace/tasks/20260911T023128Z-p-only4/half_cheetah/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/half_cheetah_rgb_state_l0_return_gt3000_100ep_openvla_native_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": "20260911T035806295710Z", + "summary": { + "frame_ms": 100.0, + "max_ms": 112.48595118522644, + "mean_effective_frames": 0.718201196935092, + "mean_ms": 71.8201196935092, + "min_ms": 65.39556503295898, + "n_samples": 29997, + "p50_frames": 0.71872318983078, + "p50_ms": 71.872318983078, + "p90_frames": 0.7277030391693116, + "p90_ms": 72.77030391693116, + "p95_frames": 0.7309400281906128, + "p95_ms": 73.09400281906127, + "p99_frames": 0.7426679273605348, + "p99_ms": 74.26679273605347, + "prob_latency_gt_1_frame": 0.00010001000100010001, + "prob_latency_gt_2_frames": 0.0, + "prob_latency_gt_3_frames": 0.0, + "std_ms": 1.0660114911848162 + }, + "visualization_path": "latency_profile.png", + "workload_id": "half_cheetah" +} diff --git a/latency-aware/half-cheetah/small-policy/sample-factory-qwenoft-rtx3090-v1/provenance.json b/latency-aware/half-cheetah/small-policy/sample-factory-qwenoft-rtx3090-v1/provenance.json new file mode 100644 index 0000000000000000000000000000000000000000..0341bdbd13bc5b964c21cde7139261c8f87f89c4 --- /dev/null +++ b/latency-aware/half-cheetah/small-policy/sample-factory-qwenoft-rtx3090-v1/provenance.json @@ -0,0 +1,44 @@ +{ + "task": "half-cheetah", + "model": "sample-factory-appo", + "training_condition": "latency-aware", + "training_run_id": "half_cheetah_profile_20260912T023319Z", + "source": { + "repo_id": "latency-sensitive-bench/Standard-Pipeline", + "revision": "2208875f92b21e3f8ec2363a30a6705dfe56cc04", + "path": "half_cheetah/profile_latency/small_model", + "url": "https://huggingface.co/latency-sensitive-bench/Standard-Pipeline/tree/2208875f92b21e3f8ec2363a30a6705dfe56cc04/half_cheetah/profile_latency/small_model" + }, + "hf_repo_id": "latency-sensitive-bench/benchmark-models", + "hf_path": "latency-aware/half-cheetah/small-policy/sample-factory-qwenoft-rtx3090-v1", + "checkpoint": { + "source_file": "half_cheetah/profile_latency/small_model/checkpoint_p0/best_000019544_10006528_reward_4867.430.pth", + "source_sha256": "0692b81428ca7c99d327185e68f53f19d14e775aba15a7586351bc06c9c370e5", + "source_bytes": 83973, + "file": "checkpoint.pth", + "selection_rule": "best_reward_after_full_training_budget", + "method": "inference_export", + "sha256": "323a1b8a6b7050ad692bf684645ec03e518fb0f2b20cbe3ec98fb92df0ceb640", + "bytes": 29877, + "train_step": 19544, + "env_steps": 10006528, + "tensor_count": 15, + "tensor_equality_verified": true, + "removed_fields": [ + "best_performance", + "curr_lr", + "optimizer" + ] + }, + "config_source": "half_cheetah/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/half-cheetah/small-policy/sample-factory-qwenoft-rtx3090-v1/rollout_descriptor.json b/latency-aware/half-cheetah/small-policy/sample-factory-qwenoft-rtx3090-v1/rollout_descriptor.json new file mode 100644 index 0000000000000000000000000000000000000000..9c7cd35d6cf26e4a4a092caaf5203d05d3151b17 --- /dev/null +++ b/latency-aware/half-cheetah/small-policy/sample-factory-qwenoft-rtx3090-v1/rollout_descriptor.json @@ -0,0 +1,310 @@ +{ + "env_name": "half_cheetah_rgb_state", + "integration_name": "gymnasium", + "action_spec": { + "layout": "gymnasium_continuous_v1", + "labels": [ + "back_thigh_torque", + "back_shin_torque", + "back_foot_torque", + "front_thigh_torque", + "front_shin_torque", + "front_foot_torque" + ], + "row_fields": [ + "action", + "action_text" + ], + "reward_field": "raw_reward", + "episode_return_field": "episode_raw_return", + "reward_semantics": "raw_reward" + }, + "base_prompt": "Make the Half Cheetah run forward as fast as possible. Predict six continuous torques in [-1, 1] ordered as back thigh, back shin, back foot, front thigh, front shin, and front foot.", + "checkpoint_experiment_dir": "/mnt/results/latency-sensitive-bench/profile-latency-new/half_cheetah/20260912T023319Z-halfcheetah/small_models/half_cheetah_profile_20260912T023319Z", + "checkpoint_path": "/mnt/results/latency-sensitive-bench/profile-latency-new/half_cheetah/20260912T023319Z-halfcheetah/small_models/half_cheetah_profile_20260912T023319Z/checkpoint_p0/best_000019544_10006528_reward_4867.430.pth", + "flat_cfg": { + "help": false, + "algo": "APPO", + "env": "latency_gymnasium", + "experiment": "half_cheetah_profile_20260912T023319Z", + "train_dir": "/mnt/results/latency-sensitive-bench/profile-latency-new/half_cheetah/20260912T023319Z-halfcheetah/small_models", + "restart_behavior": "overwrite", + "device": "gpu", + "seed": 2222, + "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": 1, + "pixel_format": "CHW", + "use_record_episode_statistics": false, + "with_wandb": false, + "wandb_user": null, + "wandb_project": "sample_factory", + "wandb_group": null, + "wandb_job_type": "SF", + "wandb_tags": [], + "with_pbt": false, + "pbt_mix_policies_in_one_env": true, + "pbt_period_env_steps": 5000000, + "pbt_start_mutation": 20000000, + "pbt_replace_fraction": 0.3, + "pbt_mutation_rate": 0.15, + "pbt_replace_reward_gap": 0.1, + "pbt_replace_reward_gap_absolute": 1e-06, + "pbt_optimize_gamma": false, + "pbt_target_objective": "true_objective", + "pbt_perturb_min": 1.1, + "pbt_perturb_max": 1.5, + "gym_task_name": "half_cheetah_rgb_state", + "gym_env_id": "LatencyBench/HalfCheetahContinuous-v0", + "gym_make_kwargs_json": "{\"base_env_id\": \"HalfCheetah-v4\", \"base_make_kwargs\": {}, \"render_mode\": \"rgb_array\"}", + "gym_registration_imports_json": "[\"latency_bench.envs.gymnasium_half_cheetah\"]", + "gym_action_space_json": "{\"dtype\": \"float32\", \"high\": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0], \"labels\": [\"back_thigh_torque\", \"back_shin_torque\", \"back_foot_torque\", \"front_thigh_torque\", \"front_shin_torque\", \"front_foot_torque\"], \"low\": [-1.0, -1.0, -1.0, -1.0, -1.0, -1.0], \"type\": \"box\"}", + "gym_action_labels_json": null, + "gym_action_values_json": null, + "gym_noop_action_id": null, + "gym_noop_action_json": "[0.0, 0.0, 0.0, 0.0, 0.0, 0.0]", + "gym_base_prompt": "Make the Half Cheetah run forward as fast as possible. Predict six continuous torques in [-1, 1] ordered as back thigh, back shin, back foot, front thigh, front shin, and front foot.", + "gym_state_labels_json": "[\"front_tip_height\", \"front_tip_angle\", \"back_thigh_angle\", \"back_shin_angle\", \"back_foot_angle\", \"front_thigh_angle\", \"front_shin_angle\", \"front_foot_angle\", \"front_tip_x_velocity\", \"front_tip_z_velocity\", \"front_tip_angular_velocity\", \"back_thigh_angular_velocity\", \"back_shin_angular_velocity\", \"back_foot_angular_velocity\", \"front_thigh_angular_velocity\", \"front_shin_angular_velocity\", \"front_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/half_cheetah/20260912T023319Z-halfcheetah/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/half_cheetah/20260912T023319Z-halfcheetah/small_models/half_cheetah_profile_20260912T023319Z/episode_metrics.jsonl", + "command_line": "--mode train --algo APPO --env latency_gymnasium --experiment half_cheetah_profile_20260912T023319Z --train_dir /mnt/results/latency-sensitive-bench/profile-latency-new/half_cheetah/20260912T023319Z-halfcheetah/small_models --restart_behavior overwrite --device gpu --seed 2222 --episode_metrics_path /mnt/results/latency-sensitive-bench/profile-latency-new/half_cheetah/20260912T023319Z-halfcheetah/small_models/half_cheetah_profile_20260912T023319Z/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 1 --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 --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 --encoder_mlp_layers 64 64 --latency-type iid --latency-seed 0 --latency-profile-path /mnt/results/latency-sensitive-bench/profile-latency-new/half_cheetah/20260912T023319Z-halfcheetah/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 half_cheetah_rgb_state --gym-env-id LatencyBench/HalfCheetahContinuous-v0 --gym-make-kwargs-json {\"base_env_id\": \"HalfCheetah-v4\", \"base_make_kwargs\": {}, \"render_mode\": \"rgb_array\"} --gym-registration-imports-json [\"latency_bench.envs.gymnasium_half_cheetah\"] --gym-action-space-json {\"dtype\": \"float32\", \"high\": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0], \"labels\": [\"back_thigh_torque\", \"back_shin_torque\", \"back_foot_torque\", \"front_thigh_torque\", \"front_shin_torque\", \"front_foot_torque\"], \"low\": [-1.0, -1.0, -1.0, -1.0, -1.0, -1.0], \"type\": \"box\"} --gym-noop-action-json [0.0, 0.0, 0.0, 0.0, 0.0, 0.0] --gym-base-prompt Make the Half Cheetah run forward as fast as possible. Predict six continuous torques in [-1, 1] ordered as back thigh, back shin, back foot, front thigh, front shin, and front foot. --gym-state-labels-json [\"front_tip_height\", \"front_tip_angle\", \"back_thigh_angle\", \"back_shin_angle\", \"back_foot_angle\", \"front_thigh_angle\", \"front_shin_angle\", \"front_foot_angle\", \"front_tip_x_velocity\", \"front_tip_z_velocity\", \"front_tip_angular_velocity\", \"back_thigh_angular_velocity\", \"back_shin_angular_velocity\", \"back_foot_angular_velocity\", \"front_thigh_angular_velocity\", \"front_shin_angular_velocity\", \"front_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": "half_cheetah_profile_20260912T023319Z", + "train_dir": "/mnt/results/latency-sensitive-bench/profile-latency-new/half_cheetah/20260912T023319Z-halfcheetah/small_models", + "restart_behavior": "overwrite", + "device": "gpu", + "seed": 2222, + "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, + "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": 1, + "with_wandb": false, + "gym_task_name": "half_cheetah_rgb_state", + "gym_env_id": "LatencyBench/HalfCheetahContinuous-v0", + "gym_make_kwargs_json": "{\"base_env_id\": \"HalfCheetah-v4\", \"base_make_kwargs\": {}, \"render_mode\": \"rgb_array\"}", + "gym_registration_imports_json": "[\"latency_bench.envs.gymnasium_half_cheetah\"]", + "gym_action_space_json": "{\"dtype\": \"float32\", \"high\": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0], \"labels\": [\"back_thigh_torque\", \"back_shin_torque\", \"back_foot_torque\", \"front_thigh_torque\", \"front_shin_torque\", \"front_foot_torque\"], \"low\": [-1.0, -1.0, -1.0, -1.0, -1.0, -1.0], \"type\": \"box\"}", + "gym_noop_action_json": "[0.0, 0.0, 0.0, 0.0, 0.0, 0.0]", + "gym_base_prompt": "Make the Half Cheetah run forward as fast as possible. Predict six continuous torques in [-1, 1] ordered as back thigh, back shin, back foot, front thigh, front shin, and front foot.", + "gym_state_labels_json": "[\"front_tip_height\", \"front_tip_angle\", \"back_thigh_angle\", \"back_shin_angle\", \"back_foot_angle\", \"front_thigh_angle\", \"front_shin_angle\", \"front_foot_angle\", \"front_tip_x_velocity\", \"front_tip_z_velocity\", \"front_tip_angular_velocity\", \"back_thigh_angular_velocity\", \"back_shin_angular_velocity\", \"back_foot_angular_velocity\", \"front_thigh_angular_velocity\", \"front_shin_angular_velocity\", \"front_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/half_cheetah/20260912T023319Z-halfcheetah/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/half_cheetah/20260912T023319Z-halfcheetah/small_models/half_cheetah_profile_20260912T023319Z/episode_metrics.jsonl", + "output_dir": "/mnt/results/latency-sensitive-bench/profile-latency-new/half_cheetah/20260912T023319Z-halfcheetah/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/half_cheetah/20260912T023319Z-halfcheetah/small_train" + }, + "device_override": "gpu", + "latency_override": { + "method": "iid", + "fixed_latency_ms": null, + "profile_path": "/mnt/results/latency-sensitive-bench/profile-latency-new/half_cheetah/20260912T023319Z-halfcheetah/profile/profile.json", + "profile_worker_slot": 0, + "seed": 0, + "add_latency_info": false + } +} diff --git a/latency-aware/half-cheetah/small-policy/sample-factory-qwenoft-rtx3090-v1/rollout_metadata.json b/latency-aware/half-cheetah/small-policy/sample-factory-qwenoft-rtx3090-v1/rollout_metadata.json new file mode 100644 index 0000000000000000000000000000000000000000..17099a26a411665491761c65cb88113439d5a196 --- /dev/null +++ b/latency-aware/half-cheetah/small-policy/sample-factory-qwenoft-rtx3090-v1/rollout_metadata.json @@ -0,0 +1,363 @@ +{ + "checkpoint_experiment_dir": "/mnt/results/latency-sensitive-bench/profile-latency-new/half_cheetah/20260912T023319Z-halfcheetah/small_models/half_cheetah_profile_20260912T023319Z", + "checkpoint_path": "/mnt/results/latency-sensitive-bench/profile-latency-new/half_cheetah/20260912T023319Z-halfcheetah/small_models/half_cheetah_profile_20260912T023319Z/checkpoint_p0/best_000019544_10006528_reward_4867.430.pth", + "config_source": "/mnt/results/latency-sensitive-bench/profile-latency-new/half_cheetah/20260912T023319Z-halfcheetah/small_models/half_cheetah_profile_20260912T023319Z/config.json", + "latency_source": "/mnt/results/latency-sensitive-bench/profile-latency-new/half_cheetah/20260912T023319Z-halfcheetah/h100/small_train.yaml", + "checkpoint_config": { + "help": false, + "algo": "APPO", + "env": "latency_gymnasium", + "experiment": "half_cheetah_profile_20260912T023319Z", + "train_dir": "/mnt/results/latency-sensitive-bench/profile-latency-new/half_cheetah/20260912T023319Z-halfcheetah/small_models", + "restart_behavior": "overwrite", + "device": "gpu", + "seed": 2222, + "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": 1, + "pixel_format": "CHW", + "use_record_episode_statistics": false, + "with_wandb": false, + "wandb_user": null, + "wandb_project": "sample_factory", + "wandb_group": null, + "wandb_job_type": "SF", + "wandb_tags": [], + "with_pbt": false, + "pbt_mix_policies_in_one_env": true, + "pbt_period_env_steps": 5000000, + "pbt_start_mutation": 20000000, + "pbt_replace_fraction": 0.3, + "pbt_mutation_rate": 0.15, + "pbt_replace_reward_gap": 0.1, + "pbt_replace_reward_gap_absolute": 1e-06, + "pbt_optimize_gamma": false, + "pbt_target_objective": "true_objective", + "pbt_perturb_min": 1.1, + "pbt_perturb_max": 1.5, + "gym_task_name": "half_cheetah_rgb_state", + "gym_env_id": "LatencyBench/HalfCheetahContinuous-v0", + "gym_make_kwargs_json": "{\"base_env_id\": \"HalfCheetah-v4\", \"base_make_kwargs\": {}, \"render_mode\": \"rgb_array\"}", + "gym_registration_imports_json": "[\"latency_bench.envs.gymnasium_half_cheetah\"]", + "gym_action_space_json": "{\"dtype\": \"float32\", \"high\": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0], \"labels\": [\"back_thigh_torque\", \"back_shin_torque\", \"back_foot_torque\", \"front_thigh_torque\", \"front_shin_torque\", \"front_foot_torque\"], \"low\": [-1.0, -1.0, -1.0, -1.0, -1.0, -1.0], \"type\": \"box\"}", + "gym_action_labels_json": null, + "gym_action_values_json": null, + "gym_noop_action_id": null, + "gym_noop_action_json": "[0.0, 0.0, 0.0, 0.0, 0.0, 0.0]", + "gym_base_prompt": "Make the Half Cheetah run forward as fast as possible. Predict six continuous torques in [-1, 1] ordered as back thigh, back shin, back foot, front thigh, front shin, and front foot.", + "gym_state_labels_json": "[\"front_tip_height\", \"front_tip_angle\", \"back_thigh_angle\", \"back_shin_angle\", \"back_foot_angle\", \"front_thigh_angle\", \"front_shin_angle\", \"front_foot_angle\", \"front_tip_x_velocity\", \"front_tip_z_velocity\", \"front_tip_angular_velocity\", \"back_thigh_angular_velocity\", \"back_shin_angular_velocity\", \"back_foot_angular_velocity\", \"front_thigh_angular_velocity\", \"front_shin_angular_velocity\", \"front_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/half_cheetah/20260912T023319Z-halfcheetah/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/half_cheetah/20260912T023319Z-halfcheetah/small_models/half_cheetah_profile_20260912T023319Z/episode_metrics.jsonl", + "command_line": "--mode train --algo APPO --env latency_gymnasium --experiment half_cheetah_profile_20260912T023319Z --train_dir /mnt/results/latency-sensitive-bench/profile-latency-new/half_cheetah/20260912T023319Z-halfcheetah/small_models --restart_behavior overwrite --device gpu --seed 2222 --episode_metrics_path /mnt/results/latency-sensitive-bench/profile-latency-new/half_cheetah/20260912T023319Z-halfcheetah/small_models/half_cheetah_profile_20260912T023319Z/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 1 --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 --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 --encoder_mlp_layers 64 64 --latency-type iid --latency-seed 0 --latency-profile-path /mnt/results/latency-sensitive-bench/profile-latency-new/half_cheetah/20260912T023319Z-halfcheetah/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 half_cheetah_rgb_state --gym-env-id LatencyBench/HalfCheetahContinuous-v0 --gym-make-kwargs-json {\"base_env_id\": \"HalfCheetah-v4\", \"base_make_kwargs\": {}, \"render_mode\": \"rgb_array\"} --gym-registration-imports-json [\"latency_bench.envs.gymnasium_half_cheetah\"] --gym-action-space-json {\"dtype\": \"float32\", \"high\": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0], \"labels\": [\"back_thigh_torque\", \"back_shin_torque\", \"back_foot_torque\", \"front_thigh_torque\", \"front_shin_torque\", \"front_foot_torque\"], \"low\": [-1.0, -1.0, -1.0, -1.0, -1.0, -1.0], \"type\": \"box\"} --gym-noop-action-json [0.0, 0.0, 0.0, 0.0, 0.0, 0.0] --gym-base-prompt Make the Half Cheetah run forward as fast as possible. Predict six continuous torques in [-1, 1] ordered as back thigh, back shin, back foot, front thigh, front shin, and front foot. --gym-state-labels-json [\"front_tip_height\", \"front_tip_angle\", \"back_thigh_angle\", \"back_shin_angle\", \"back_foot_angle\", \"front_thigh_angle\", \"front_shin_angle\", \"front_foot_angle\", \"front_tip_x_velocity\", \"front_tip_z_velocity\", \"front_tip_angular_velocity\", \"back_thigh_angular_velocity\", \"back_shin_angular_velocity\", \"back_foot_angular_velocity\", \"front_thigh_angular_velocity\", \"front_shin_angular_velocity\", \"front_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": "half_cheetah_profile_20260912T023319Z", + "train_dir": "/mnt/results/latency-sensitive-bench/profile-latency-new/half_cheetah/20260912T023319Z-halfcheetah/small_models", + "restart_behavior": "overwrite", + "device": "gpu", + "seed": 2222, + "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, + "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": 1, + "with_wandb": false, + "gym_task_name": "half_cheetah_rgb_state", + "gym_env_id": "LatencyBench/HalfCheetahContinuous-v0", + "gym_make_kwargs_json": "{\"base_env_id\": \"HalfCheetah-v4\", \"base_make_kwargs\": {}, \"render_mode\": \"rgb_array\"}", + "gym_registration_imports_json": "[\"latency_bench.envs.gymnasium_half_cheetah\"]", + "gym_action_space_json": "{\"dtype\": \"float32\", \"high\": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0], \"labels\": [\"back_thigh_torque\", \"back_shin_torque\", \"back_foot_torque\", \"front_thigh_torque\", \"front_shin_torque\", \"front_foot_torque\"], \"low\": [-1.0, -1.0, -1.0, -1.0, -1.0, -1.0], \"type\": \"box\"}", + "gym_noop_action_json": "[0.0, 0.0, 0.0, 0.0, 0.0, 0.0]", + "gym_base_prompt": "Make the Half Cheetah run forward as fast as possible. Predict six continuous torques in [-1, 1] ordered as back thigh, back shin, back foot, front thigh, front shin, and front foot.", + "gym_state_labels_json": "[\"front_tip_height\", \"front_tip_angle\", \"back_thigh_angle\", \"back_shin_angle\", \"back_foot_angle\", \"front_thigh_angle\", \"front_shin_angle\", \"front_foot_angle\", \"front_tip_x_velocity\", \"front_tip_z_velocity\", \"front_tip_angular_velocity\", \"back_thigh_angular_velocity\", \"back_shin_angular_velocity\", \"back_foot_angular_velocity\", \"front_thigh_angular_velocity\", \"front_shin_angular_velocity\", \"front_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/half_cheetah/20260912T023319Z-halfcheetah/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/half_cheetah/20260912T023319Z-halfcheetah/small_models/half_cheetah_profile_20260912T023319Z/episode_metrics.jsonl", + "output_dir": "/mnt/results/latency-sensitive-bench/profile-latency-new/half_cheetah/20260912T023319Z-halfcheetah/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/half_cheetah/20260912T023319Z-halfcheetah/small_train" + }, + "latency_override": { + "method": "iid", + "fixed_latency_ms": null, + "profile_path": "/mnt/results/latency-sensitive-bench/profile-latency-new/half_cheetah/20260912T023319Z-halfcheetah/profile/profile.json", + "profile_worker_slot": 0, + "seed": 0, + "add_latency_info": false + }, + "gymnasium_task": { + "task_name": "half_cheetah_rgb_state", + "env_id": "LatencyBench/HalfCheetahContinuous-v0", + "make_kwargs": { + "base_env_id": "HalfCheetah-v4", + "base_make_kwargs": {}, + "render_mode": "rgb_array" + }, + "registration_imports": [ + "latency_bench.envs.gymnasium_half_cheetah" + ], + "base_prompt": "Make the Half Cheetah run forward as fast as possible. Predict six continuous torques in [-1, 1] ordered as back thigh, back shin, back foot, front thigh, front shin, and front foot.", + "env_fps": 10.0, + "obs_fps": 10.0, + "frame_stack": 1, + "state_space": { + "labels": [ + "front_tip_height", + "front_tip_angle", + "back_thigh_angle", + "back_shin_angle", + "back_foot_angle", + "front_thigh_angle", + "front_shin_angle", + "front_foot_angle", + "front_tip_x_velocity", + "front_tip_z_velocity", + "front_tip_angular_velocity", + "back_thigh_angular_velocity", + "back_shin_angular_velocity", + "back_foot_angular_velocity", + "front_thigh_angular_velocity", + "front_shin_angular_velocity", + "front_foot_angular_velocity" + ] + }, + "action_space": { + "dtype": "float32", + "high": [ + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0 + ], + "labels": [ + "back_thigh_torque", + "back_shin_torque", + "back_foot_torque", + "front_thigh_torque", + "front_shin_torque", + "front_foot_torque" + ], + "low": [ + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0 + ], + "type": "box" + }, + "noop_action": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + } +} diff --git a/latency-aware/half-cheetah/small-policy/sample-factory-qwenoft-rtx3090-v1/selection.json b/latency-aware/half-cheetah/small-policy/sample-factory-qwenoft-rtx3090-v1/selection.json new file mode 100644 index 0000000000000000000000000000000000000000..48e5a7f2f0d8f3b982df7cb41acd8a907c7db5c5 --- /dev/null +++ b/latency-aware/half-cheetah/small-policy/sample-factory-qwenoft-rtx3090-v1/selection.json @@ -0,0 +1,14 @@ +{ + "run_id": "20260912T023319Z-halfcheetah", + "source_profile_run_id": "20260911T035806295710Z", + "selection_rule": "best_reward_after_full_training_budget", + "best_checkpoint": "/mnt/results/latency-sensitive-bench/profile-latency-new/half_cheetah/20260912T023319Z-halfcheetah/small_models/half_cheetah_profile_20260912T023319Z/checkpoint_p0/best_000019544_10006528_reward_4867.430.pth", + "best_bundle_path": "checkpoint_p0/best_000019544_10006528_reward_4867.430.pth", + "final_checkpoint": "/mnt/results/latency-sensitive-bench/profile-latency-new/half_cheetah/20260912T023319Z-halfcheetah/small_models/half_cheetah_profile_20260912T023319Z/checkpoint_p0/checkpoint_000019544_10006528.pth", + "final_env_steps": 10006528, + "final_train_step": 19544, + "training_exit_code": 0, + "profile_sha256": "5580e628965a3fa3d29538c2b51dbd9ae3f6adc8662bd2cca8f1ada5a1d8e06e", + "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/half-cheetah/small-policy/sample-factory-qwenoft-rtx3090-v1/small_train.yaml b/latency-aware/half-cheetah/small-policy/sample-factory-qwenoft-rtx3090-v1/small_train.yaml new file mode 100644 index 0000000000000000000000000000000000000000..63e70092f89c62bf08191ed5c7f40eae0c363020 --- /dev/null +++ b/latency-aware/half-cheetah/small-policy/sample-factory-qwenoft-rtx3090-v1/small_train.yaml @@ -0,0 +1,172 @@ +experiment: + name: half_cheetah_profile_20260912T023319Z + seed: 2222 +backend: + type: sample_factory + algo: APPO + device: cuda + train_dir: /mnt/results/latency-sensitive-bench/profile-latency-new/half_cheetah/20260912T023319Z-halfcheetah/small_models + restart_behavior: overwrite + run_mode: train +executor: + mode: simulated +env: + name: gymnasium + action_space: + dtype: float32 + high: + - 1.0 + - 1.0 + - 1.0 + - 1.0 + - 1.0 + - 1.0 + labels: + - back_thigh_torque + - back_shin_torque + - back_foot_torque + - front_thigh_torque + - front_shin_torque + - front_foot_torque + low: + - -1.0 + - -1.0 + - -1.0 + - -1.0 + - -1.0 + - -1.0 + type: box + base_prompt: Make the Half Cheetah run forward as fast as possible. Predict six + continuous torques in [-1, 1] ordered as back thigh, back shin, back foot, front + thigh, front shin, and front foot. + env_fps: 10.0 + env_id: LatencyBench/HalfCheetahContinuous-v0 + frame_stack: 1 + make_kwargs: + base_env_id: HalfCheetah-v4 + base_make_kwargs: {} + render_mode: rgb_array + noop_action: + - 0.0 + - 0.0 + - 0.0 + - 0.0 + - 0.0 + - 0.0 + obs_fps: 10.0 + registration_imports: + - latency_bench.envs.gymnasium_half_cheetah + state_space: + labels: + - front_tip_height + - front_tip_angle + - back_thigh_angle + - back_shin_angle + - back_foot_angle + - front_thigh_angle + - front_shin_angle + - front_foot_angle + - front_tip_x_velocity + - front_tip_z_velocity + - front_tip_angular_velocity + - back_thigh_angular_velocity + - back_shin_angular_velocity + - back_foot_angular_velocity + - front_thigh_angular_velocity + - front_shin_angular_velocity + - front_foot_angular_velocity + task_name: half_cheetah_rgb_state +latency: + method: iid + fixed_latency_ms: null + profile_path: /mnt/results/latency-sensitive-bench/profile-latency-new/half_cheetah/20260912T023319Z-halfcheetah/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 + 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 + env_framestack: 1 + 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 + with_wandb: false +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/half_cheetah/20260912T023319Z-halfcheetah/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/half-cheetah/small-policy/sample-factory-qwenpi_v3-rtx3090-v1/README.md b/latency-aware/half-cheetah/small-policy/sample-factory-qwenpi_v3-rtx3090-v1/README.md new file mode 100644 index 0000000000000000000000000000000000000000..f942a3b43b88b8da45ff60e84b014dba8658c2a4 --- /dev/null +++ b/latency-aware/half-cheetah/small-policy/sample-factory-qwenpi_v3-rtx3090-v1/README.md @@ -0,0 +1,30 @@ +# half-cheetah / sample-factory-appo + +Training condition: `latency-aware`. Run: `pi05_half_cheetah_profile_appo_h1_5fps_g128_20260921`. + +Copied from [Standard-Pipeline](https://huggingface.co/latency-sensitive-bench/Standard-Pipeline/tree/2208875f92b21e3f8ec2363a30a6705dfe56cc04/half_cheetah/profile_latency/small_model/Pi05) at `2208875f92b21e3f8ec2363a30a6705dfe56cc04`. + +- Checkpoint: `checkpoint.pth` +- Selection: final +- Checkpoint SHA256: `e4132118cf179ae368b37aacdc3566b58463b77a8d541647cb45a5b0879010aa` +- 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/half-cheetah/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/half-cheetah/small-policy/sample-factory-qwenpi_v3-rtx3090-v1/checkpoint.pth b/latency-aware/half-cheetah/small-policy/sample-factory-qwenpi_v3-rtx3090-v1/checkpoint.pth new file mode 100644 index 0000000000000000000000000000000000000000..65dc51f4e322e6e811c45822fabda0a8f3a25ab4 --- /dev/null +++ b/latency-aware/half-cheetah/small-policy/sample-factory-qwenpi_v3-rtx3090-v1/checkpoint.pth @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e4132118cf179ae368b37aacdc3566b58463b77a8d541647cb45a5b0879010aa +size 29835 diff --git a/latency-aware/half-cheetah/small-policy/sample-factory-qwenpi_v3-rtx3090-v1/config.json b/latency-aware/half-cheetah/small-policy/sample-factory-qwenpi_v3-rtx3090-v1/config.json new file mode 100644 index 0000000000000000000000000000000000000000..5beff633e5fd6cec7826c122e5ce56d21faaf1df --- /dev/null +++ b/latency-aware/half-cheetah/small-policy/sample-factory-qwenpi_v3-rtx3090-v1/config.json @@ -0,0 +1,286 @@ +{ + "help": false, + "algo": "APPO", + "env": "latency_gymnasium", + "experiment": "pi05_half_cheetah_profile_appo_h1_5fps_g128_20260921", + "train_dir": "${PI05_RUN_DIR}/profile_latency/teacher/checkpoints", + "restart_behavior": "overwrite", + "device": "gpu", + "seed": 2222, + "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": 1, + "pixel_format": "CHW", + "use_record_episode_statistics": false, + "with_wandb": true, + "wandb_user": "dongqianyu99-zhejiang-university", + "wandb_project": "latency-sensitive-bench", + "wandb_group": "pi05-profile-teachers-h1-5fps-20260921", + "wandb_job_type": "profile_teacher", + "wandb_tags": [ + "half_cheetah", + "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": "half_cheetah", + "gym_env_id": "LatencyBench/HalfCheetahContinuous-v0", + "gym_make_kwargs_json": "{\"base_env_id\": \"HalfCheetah-v4\", \"render_mode\": \"rgb_array\", \"base_make_kwargs\": {}}", + "gym_registration_imports_json": "[\"latency_bench.envs.gymnasium_half_cheetah\"]", + "gym_action_space_json": "{\"type\": \"box\", \"labels\": [\"back_thigh_torque\", \"back_shin_torque\", \"back_foot_torque\", \"front_thigh_torque\", \"front_shin_torque\", \"front_foot_torque\"], \"low\": [-1.0, -1.0, -1.0, -1.0, -1.0, -1.0], \"high\": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0], \"dtype\": \"float32\"}", + "gym_action_labels_json": null, + "gym_action_values_json": null, + "gym_noop_action_id": null, + "gym_noop_action_json": "[0.0, 0.0, 0.0, 0.0, 0.0, 0.0]", + "gym_base_prompt": "Make the Half Cheetah run forward as fast as possible. Predict six continuous torques in [-1, 1] ordered as back thigh, back shin, back foot, front thigh, front shin, and front foot.", + "gym_state_labels_json": "[\"front_tip_height\", \"front_tip_angle\", \"back_thigh_angle\", \"back_shin_angle\", \"back_foot_angle\", \"front_thigh_angle\", \"front_shin_angle\", \"front_foot_angle\", \"front_tip_x_velocity\", \"front_tip_z_velocity\", \"front_tip_angular_velocity\", \"back_thigh_angular_velocity\", \"back_shin_angular_velocity\", \"back_foot_angular_velocity\", \"front_thigh_angular_velocity\", \"front_shin_angular_velocity\", \"front_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_half_cheetah_profile_appo_h1_5fps_g128_20260921/episode_metrics.jsonl", + "ppo": null, + "command_line": "--mode train --algo APPO --env latency_gymnasium --experiment pi05_half_cheetah_profile_appo_h1_5fps_g128_20260921 --train_dir ${PI05_RUN_DIR}/profile_latency/teacher/checkpoints --restart_behavior overwrite --device gpu --seed 2222 --episode_metrics_path ${PI05_RUN_DIR}/profile_latency/teacher/checkpoints/pi05_half_cheetah_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 --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 half_cheetah APPO Pi05 measured-IID-profile H1 5FPS --wandb_user dongqianyu99-zhejiang-university --gym-task-name half_cheetah --gym-env-id LatencyBench/HalfCheetahContinuous-v0 --gym-make-kwargs-json {\"base_env_id\": \"HalfCheetah-v4\", \"render_mode\": \"rgb_array\", \"base_make_kwargs\": {}} --gym-registration-imports-json [\"latency_bench.envs.gymnasium_half_cheetah\"] --gym-action-space-json {\"type\": \"box\", \"labels\": [\"back_thigh_torque\", \"back_shin_torque\", \"back_foot_torque\", \"front_thigh_torque\", \"front_shin_torque\", \"front_foot_torque\"], \"low\": [-1.0, -1.0, -1.0, -1.0, -1.0, -1.0], \"high\": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0], \"dtype\": \"float32\"} --gym-noop-action-json [0.0, 0.0, 0.0, 0.0, 0.0, 0.0] --gym-base-prompt Make the Half Cheetah run forward as fast as possible. Predict six continuous torques in [-1, 1] ordered as back thigh, back shin, back foot, front thigh, front shin, and front foot. --gym-state-labels-json [\"front_tip_height\", \"front_tip_angle\", \"back_thigh_angle\", \"back_shin_angle\", \"back_foot_angle\", \"front_thigh_angle\", \"front_shin_angle\", \"front_foot_angle\", \"front_tip_x_velocity\", \"front_tip_z_velocity\", \"front_tip_angular_velocity\", \"back_thigh_angular_velocity\", \"back_shin_angular_velocity\", \"back_foot_angular_velocity\", \"front_thigh_angular_velocity\", \"front_shin_angular_velocity\", \"front_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_half_cheetah_profile_appo_h1_5fps_g128_20260921", + "train_dir": "${PI05_RUN_DIR}/profile_latency/teacher/checkpoints", + "restart_behavior": "overwrite", + "device": "gpu", + "seed": 2222, + "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, + "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": [ + "half_cheetah", + "APPO", + "Pi05", + "measured-IID-profile", + "H1", + "5FPS" + ], + "gym_task_name": "half_cheetah", + "gym_env_id": "LatencyBench/HalfCheetahContinuous-v0", + "gym_make_kwargs_json": "{\"base_env_id\": \"HalfCheetah-v4\", \"render_mode\": \"rgb_array\", \"base_make_kwargs\": {}}", + "gym_registration_imports_json": "[\"latency_bench.envs.gymnasium_half_cheetah\"]", + "gym_action_space_json": "{\"type\": \"box\", \"labels\": [\"back_thigh_torque\", \"back_shin_torque\", \"back_foot_torque\", \"front_thigh_torque\", \"front_shin_torque\", \"front_foot_torque\"], \"low\": [-1.0, -1.0, -1.0, -1.0, -1.0, -1.0], \"high\": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0], \"dtype\": \"float32\"}", + "gym_noop_action_json": "[0.0, 0.0, 0.0, 0.0, 0.0, 0.0]", + "gym_base_prompt": "Make the Half Cheetah run forward as fast as possible. Predict six continuous torques in [-1, 1] ordered as back thigh, back shin, back foot, front thigh, front shin, and front foot.", + "gym_state_labels_json": "[\"front_tip_height\", \"front_tip_angle\", \"back_thigh_angle\", \"back_shin_angle\", \"back_foot_angle\", \"front_thigh_angle\", \"front_shin_angle\", \"front_foot_angle\", \"front_tip_x_velocity\", \"front_tip_z_velocity\", \"front_tip_angular_velocity\", \"back_thigh_angular_velocity\", \"back_shin_angular_velocity\", \"back_foot_angular_velocity\", \"front_thigh_angular_velocity\", \"front_shin_angular_velocity\", \"front_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_half_cheetah_profile_appo_h1_5fps_g128_20260921/episode_metrics.jsonl", + "output_dir": "${PI05_RUN_DIR}/profile_latency/teacher/training", + "wandb_unique_id": "pi05_half_cheetah_profile_appo_h1_5fps_g128_20260921" + }, + "git_hash": "unknown", + "git_repo_name": "not a git repository", + "eval_env_frameskip": 1, + "output_dir": "${PI05_RUN_DIR}/profile_latency/teacher/training", + "wandb_unique_id": "pi05_half_cheetah_profile_appo_h1_5fps_g128_20260921" +} \ No newline at end of file diff --git a/latency-aware/half-cheetah/small-policy/sample-factory-qwenpi_v3-rtx3090-v1/provenance.json b/latency-aware/half-cheetah/small-policy/sample-factory-qwenpi_v3-rtx3090-v1/provenance.json new file mode 100644 index 0000000000000000000000000000000000000000..cd441360da7e0988fff536025c16a9aa3320ca40 --- /dev/null +++ b/latency-aware/half-cheetah/small-policy/sample-factory-qwenpi_v3-rtx3090-v1/provenance.json @@ -0,0 +1,40 @@ +{ + "task": "half-cheetah", + "model": "sample-factory-appo", + "training_condition": "latency-aware", + "training_run_id": "pi05_half_cheetah_profile_appo_h1_5fps_g128_20260921", + "source": { + "repo_id": "latency-sensitive-bench/Standard-Pipeline", + "revision": "2208875f92b21e3f8ec2363a30a6705dfe56cc04", + "path": "half_cheetah/profile_latency/small_model/Pi05", + "url": "https://huggingface.co/latency-sensitive-bench/Standard-Pipeline/tree/2208875f92b21e3f8ec2363a30a6705dfe56cc04/half_cheetah/profile_latency/small_model/Pi05" + }, + "hf_repo_id": "latency-sensitive-bench/benchmark-models", + "hf_path": "latency-aware/half-cheetah/small-policy/sample-factory-qwenpi_v3-rtx3090-v1", + "checkpoint": { + "source_file": "half_cheetah/profile_latency/small_model/Pi05/checkpoint_p0/selected.pth", + "source_sha256": "e4132118cf179ae368b37aacdc3566b58463b77a8d541647cb45a5b0879010aa", + "source_bytes": 29835, + "file": "checkpoint.pth", + "selection_rule": "final", + "method": "copy", + "sha256": "e4132118cf179ae368b37aacdc3566b58463b77a8d541647cb45a5b0879010aa", + "bytes": 29835, + "train_step": 19536, + "env_steps": 10002432, + "tensor_count": 15, + "tensor_equality_verified": true, + "removed_fields": [] + }, + "config_source": "half_cheetah/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/half-cheetah/small-policy/sample-factory-qwenpi_v3-rtx3090-v1/selection.json b/latency-aware/half-cheetah/small-policy/sample-factory-qwenpi_v3-rtx3090-v1/selection.json new file mode 100644 index 0000000000000000000000000000000000000000..3c4b077caded3aa7b8efdf9d86b0b6811e5f5da2 --- /dev/null +++ b/latency-aware/half-cheetah/small-policy/sample-factory-qwenpi_v3-rtx3090-v1/selection.json @@ -0,0 +1,116 @@ +{ + "selected": "final", + "selection_policy": "higher_mean_over_paired_20; ties_select_final", + "candidates": { + "final": { + "checkpoint": "${PI05_RUN_DIR}/profile_latency/teacher/checkpoints/pi05_half_cheetah_profile_appo_h1_5fps_g128_20260921/checkpoint_p0/checkpoint_000019536_10002432.pth", + "sha256": "612abe26382a5e0a2dc2214c10fa5951c1634362231bc6c3d9f09c08bef7471f", + "train_step": 19536, + "env_steps": 10002432, + "episodes": 20, + "seeds": [ + 104771, + 104772, + 104773, + 104774, + 104775, + 104776, + 104777, + 104778, + 104779, + 104780, + 104781, + 104782, + 104783, + 104784, + 104785, + 104786, + 104787, + 104788, + 104789, + 104790 + ], + "returns": [ + 1934.7516296771494, + 1938.8264883092277, + 1856.9064389676444, + 1848.2332719282635, + 1925.866899346379, + 1857.8906012224634, + 1918.515181152549, + 1883.3762460267756, + 1930.4407512044838, + 1892.1105602876216, + 1916.995842807158, + 1893.3043491687142, + 1801.0072653006578, + 1886.5925101871578, + 1835.578970515134, + 1843.513944498525, + 1939.4990761867555, + 1955.8559919207282, + 1918.0007335675198, + 1813.2951639002843 + ], + "mean": 1889.5280958087596, + "population_std": 44.31970628800741, + "strict_gt3000": 0 + }, + "training_best": { + "checkpoint": "${PI05_RUN_DIR}/profile_latency/teacher/checkpoints/pi05_half_cheetah_profile_appo_h1_5fps_g128_20260921/checkpoint_p0/best_000019536_10002432_reward_1810.751.pth", + "sha256": "89b509df718a271df9edfd81136c3e97c8219f6fe31de66afd6e895f621e3694", + "train_step": 19536, + "env_steps": 10002432, + "episodes": 20, + "seeds": [ + 104771, + 104772, + 104773, + 104774, + 104775, + 104776, + 104777, + 104778, + 104779, + 104780, + 104781, + 104782, + 104783, + 104784, + 104785, + 104786, + 104787, + 104788, + 104789, + 104790 + ], + "returns": [ + 1934.7516296771494, + 1938.8264883092277, + 1856.9064389676444, + 1848.2332719282635, + 1925.866899346379, + 1857.8906012224634, + 1918.515181152549, + 1883.3762460267756, + 1930.4407512044838, + 1892.1105602876216, + 1916.995842807158, + 1893.3043491687142, + 1801.0072653006578, + 1886.5925101871578, + 1835.578970515134, + 1843.513944498525, + 1939.4990761867555, + 1955.8559919207282, + 1918.0007335675198, + 1813.2951639002843 + ], + "mean": 1889.5280958087596, + "population_std": 44.31970628800741, + "strict_gt3000": 0 + } + }, + "selected_checkpoint_sha256": "612abe26382a5e0a2dc2214c10fa5951c1634362231bc6c3d9f09c08bef7471f", + "completed_at_utc": "2026-09-21T13:45:30.235292+00:00" +} diff --git a/latency-aware/half-cheetah/small-policy/sample-factory-qwenpi_v3-rtx3090-v1/source/README.md b/latency-aware/half-cheetah/small-policy/sample-factory-qwenpi_v3-rtx3090-v1/source/README.md new file mode 100644 index 0000000000000000000000000000000000000000..3e22f6e67cfa75cc19f421f120ae4fb1a798b088 --- /dev/null +++ b/latency-aware/half-cheetah/small-policy/sample-factory-qwenpi_v3-rtx3090-v1/source/README.md @@ -0,0 +1,7 @@ +# Half Cheetah Pi0.5 profile-latency APPO checkpoint + +Selected `final` at train step 19536 / 10002432 Sample Factory environment steps, seed 2222, native state/action 17/6, and 5/5 FPS. The selected paired evaluation mean was 1889.528096 (population SD 44.319706); 0 of 20 returns were strictly above 3000. P profile SHA256 `0701721a6111157054fb2e374970afbbffafa2355835b1acc23a1d69184bc132` (publication revision `e1a9bdd566b96d5da4703f769d40eab656af491e`). + +Scientific stop: selected mean 1889.5280958087596 and 0/20 returns >3000. This requested checkpoint export is not an accepted teacher; no formal data or VLA stage was run. + +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/half-cheetah/small-policy/sample-factory-qwenpi_v3-rtx3090-v1/source/provenance.json b/latency-aware/half-cheetah/small-policy/sample-factory-qwenpi_v3-rtx3090-v1/source/provenance.json new file mode 100644 index 0000000000000000000000000000000000000000..deb34475b1d33b7aab4ec34b3e95ef7716bb2445 --- /dev/null +++ b/latency-aware/half-cheetah/small-policy/sample-factory-qwenpi_v3-rtx3090-v1/source/provenance.json @@ -0,0 +1,50 @@ +{ + "state": "SELECTED_TEACHER_INFERENCE_EXPORT_CPU_RELOAD_VERIFIED", + "task": "half_cheetah", + "selected": "final", + "source_checkpoint": "checkpoint_000019536_10002432.pth", + "source_checkpoint_sha256": "612abe26382a5e0a2dc2214c10fa5951c1634362231bc6c3d9f09c08bef7471f", + "export_checkpoint": "checkpoint_p0/selected.pth", + "export_checkpoint_sha256": "e4132118cf179ae368b37aacdc3566b58463b77a8d541647cb45a5b0879010aa", + "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": 2222, + "native_state_dim": 17, + "native_action_dim": 6, + "profile_sha256": "0701721a6111157054fb2e374970afbbffafa2355835b1acc23a1d69184bc132", + "profile_publication_revision": "e1a9bdd566b96d5da4703f769d40eab656af491e", + "audit_state": "TEACHER_FULL_TRACE_AUDIT_PASSED_SCIENTIFIC_STOP", + "audit_gate": { + "strict_return_gt": 3000, + "Final_qualifying_episodes": 0, + "training_best_qualifying_episodes": 0, + "state": "SCIENTIFIC_STOP_SELECTED_20_ZERO_QUALIFYING" + }, + "selected_mean": 1889.5280958087596, + "selected_population_std": 44.31970628800741, + "selected_strict_gt3000_episodes": 0, + "data": { + "state": "NOT_EXPORTED_SCIENTIFIC_STOP" + }, + "source_sha256": { + "config.json": "039d7f577f364037803a7180d6be17b52afe4616cd67ae1e7c8508e1e8d0035c", + "teacher.yaml": "aec3c58d02f16c28205998505bd0c5e52b9fb2dc62f3fc931735d71edf8425e7", + "selection.json": "17c14c4a945815653e476edd32b266fb679941d1a897c1867e66ea3df94cfe2b", + "verification.json": "d99341e9c1b21a1a3e82c0affbb2ef59aa31fc672cb0ba13b70b26e4cab15df5" + } +} diff --git a/latency-aware/half-cheetah/small-policy/sample-factory-qwenpi_v3-rtx3090-v1/teacher.yaml b/latency-aware/half-cheetah/small-policy/sample-factory-qwenpi_v3-rtx3090-v1/teacher.yaml new file mode 100644 index 0000000000000000000000000000000000000000..34c528efb1aaff3b4e929fe56c44394007541258 --- /dev/null +++ b/latency-aware/half-cheetah/small-policy/sample-factory-qwenpi_v3-rtx3090-v1/teacher.yaml @@ -0,0 +1,170 @@ +experiment: + name: pi05_half_cheetah_profile_appo_h1_5fps_g128_20260921 + seed: 2222 +backend: + type: sample_factory + algo: APPO + device: cuda + train_dir: ${PI05_RUN_DIR}/profile_latency/teacher/checkpoints + restart_behavior: overwrite + run_mode: train + extra_args: + - --wandb_user + - dongqianyu99-zhejiang-university +executor: + mode: simulated +env: + name: gymnasium + task_name: half_cheetah + env_id: LatencyBench/HalfCheetahContinuous-v0 + registration_imports: + - latency_bench.envs.gymnasium_half_cheetah + make_kwargs: + base_env_id: HalfCheetah-v4 + render_mode: rgb_array + base_make_kwargs: {} + env_fps: 5 + obs_fps: 5 + frame_stack: 1 + state_space: + labels: + - front_tip_height + - front_tip_angle + - back_thigh_angle + - back_shin_angle + - back_foot_angle + - front_thigh_angle + - front_shin_angle + - front_foot_angle + - front_tip_x_velocity + - front_tip_z_velocity + - front_tip_angular_velocity + - back_thigh_angular_velocity + - back_shin_angular_velocity + - back_foot_angular_velocity + - front_thigh_angular_velocity + - front_shin_angular_velocity + - front_foot_angular_velocity + action_space: + type: box + labels: + - back_thigh_torque + - back_shin_torque + - back_foot_torque + - front_thigh_torque + - front_shin_torque + - front_foot_torque + low: + - -1.0 + - -1.0 + - -1.0 + - -1.0 + - -1.0 + - -1.0 + high: + - 1.0 + - 1.0 + - 1.0 + - 1.0 + - 1.0 + - 1.0 + dtype: float32 + noop_action: + - 0.0 + - 0.0 + - 0.0 + - 0.0 + - 0.0 + - 0.0 + base_prompt: Make the Half Cheetah run forward as fast as possible. Predict six + continuous torques in [-1, 1] ordered as back thigh, back shin, back foot, front + thigh, front shin, and front foot. +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 + 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_half_cheetah_profile_appo_h1_5fps_g128_20260921 + wandb_job_type: profile_teacher + wandb_tags: + - half_cheetah + - APPO + - Pi05 + - measured-IID-profile + - H1 + - 5FPS diff --git a/latency-aware/half-cheetah/small-policy/sample-factory-qwenpi_v3-rtx3090-v1/verification.json b/latency-aware/half-cheetah/small-policy/sample-factory-qwenpi_v3-rtx3090-v1/verification.json new file mode 100644 index 0000000000000000000000000000000000000000..37fe8399b341281eebc2742ca1b9a7337dd950d1 --- /dev/null +++ b/latency-aware/half-cheetah/small-policy/sample-factory-qwenpi_v3-rtx3090-v1/verification.json @@ -0,0 +1,226 @@ +{ + "state": "TEACHER_FULL_TRACE_AUDIT_PASSED_SCIENTIFIC_STOP", + "task": "half_cheetah", + "verified_at_utc": "2026-09-21T14:04:36.819491+00:00", + "source_recipe_sha256": "d59c0024205f97b0b5b78ec59f4a65040e6c4357fbd08c9b72f755848a5cad58", + "teacher_config_sha256": "aec3c58d02f16c28205998505bd0c5e52b9fb2dc62f3fc931735d71edf8425e7", + "P_profile_sha256": "0701721a6111157054fb2e374970afbbffafa2355835b1acc23a1d69184bc132", + "P_audit_state": "PROFILE_PROTOCOL_AND_CANONICAL_SAMPLES_VERIFIED", + "P_verification_sha256": "3ecbf4de38927ec065e19967edec0286eeaef2115515ffe75c3e1075af0afa24", + "P_profile_files": { + "profile.json": { + "size": 2034, + "sha256": "0701721a6111157054fb2e374970afbbffafa2355835b1acc23a1d69184bc132" + }, + "latency_distribution.json": { + "size": 25468, + "sha256": "2b2eae3acc378eeccaf728de1ff52e8530dd4e801b43f89ef491d86239ef74dd" + }, + "latency_burst_model.json": { + "size": 22349, + "sha256": "501daf6afa2979b2a43e6f951b29e14a0a7d2c7d538e83899879ac80b41acde2" + }, + "hardware.json": { + "size": 988, + "sha256": "96bcbee23e82762b3c411a16fcd96227c3029457b5a40114bb4070ff7fda910e" + }, + "latency_profile.png": { + "size": 61409, + "sha256": "732411a4b5921074aa77cb65bbd4a9e033eb9065e55d6128d6215d4c6f6b4f53" + }, + "DONE": { + "size": 0, + "sha256": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855" + } + }, + "training": { + "source_seed": 2222, + "target_env_steps": 10000000, + "Final_checkpoint": "${PI05_RUN_DIR}/profile_latency/teacher/checkpoints/pi05_half_cheetah_profile_appo_h1_5fps_g128_20260921/checkpoint_p0/checkpoint_000019536_10002432.pth", + "Final_checkpoint_sha256": "612abe26382a5e0a2dc2214c10fa5951c1634362231bc6c3d9f09c08bef7471f", + "Final_checkpoint_train_step": 19536, + "Final_checkpoint_env_steps": 10002432, + "training_best_checkpoint": "${PI05_RUN_DIR}/profile_latency/teacher/checkpoints/pi05_half_cheetah_profile_appo_h1_5fps_g128_20260921/checkpoint_p0/best_000019536_10002432_reward_1810.751.pth", + "training_best_checkpoint_sha256": "89b509df718a271df9edfd81136c3e97c8219f6fe31de66afd6e895f621e3694", + "training_best_train_step": 19536, + "training_best_env_steps": 10002432, + "unit_exit_code_proven": false, + "unit_evidence_boundary": "Transient service is LoadState=not-found; completion is established by user report plus actual Final checkpoint env_steps, not default systemd status fields." + }, + "selection": { + "selected": "final", + "policy": "higher_mean_over_paired_20; ties_select_final", + "paired_seeds": [ + 104771, + 104772, + 104773, + 104774, + 104775, + 104776, + 104777, + 104778, + 104779, + 104780, + 104781, + 104782, + 104783, + 104784, + 104785, + 104786, + 104787, + 104788, + 104789, + 104790 + ], + "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": [ + 1934.7516296771494, + 1938.8264883092277, + 1856.9064389676444, + 1848.2332719282635, + 1925.866899346379, + 1857.8906012224634, + 1918.515181152549, + 1883.3762460267756, + 1930.4407512044838, + 1892.1105602876216, + 1916.995842807158, + 1893.3043491687142, + 1801.0072653006578, + 1886.5925101871578, + 1835.578970515134, + 1843.513944498525, + 1939.4990761867555, + 1955.8559919207282, + 1918.0007335675198, + 1813.2951639002843 + ], + "mean_return": 1889.5280958087596, + "population_std_return": 44.31970628800741, + "strict_gt3000": 0, + "raw_steps": 20000, + "actions": 20000, + "invalid_actions": 0, + "dropped_actions": 0, + "submitted_observations": 20000, + "dropped_observations": 0, + "mean_action_latency_ms": 145.74751463552934, + "action_latency_p95_ms": 151.4791488018123, + "raw_action_min": -1.0, + "raw_action_max": 1.0, + "raw_actions_requiring_native_clip": 0, + "config_sha256": "751b98c49ddecc84c69b46733fcfbc678fcd60671f1ed86116ef80d5a6985772", + "episode_metrics_sha256": "fad97d4b8f43609968239ce03d56155bec56e5ef3614338f16bea4dba4349805", + "steps_sha256": "b2a95ff5c1efe37687ac8bc244de04dd6545d52870b28ef842de4c61261f9cd9", + "actions_sha256": "5f63fae02a1a1e83420ef304bcd828f483aec94f3d0236c88cfc99814ca11ea9", + "raw_step_indices_clock_and_reward_sums_verified": true, + "profile_action_and_latency_provenance_verified": true + }, + "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": [ + 1934.7516296771494, + 1938.8264883092277, + 1856.9064389676444, + 1848.2332719282635, + 1925.866899346379, + 1857.8906012224634, + 1918.515181152549, + 1883.3762460267756, + 1930.4407512044838, + 1892.1105602876216, + 1916.995842807158, + 1893.3043491687142, + 1801.0072653006578, + 1886.5925101871578, + 1835.578970515134, + 1843.513944498525, + 1939.4990761867555, + 1955.8559919207282, + 1918.0007335675198, + 1813.2951639002843 + ], + "mean_return": 1889.5280958087596, + "population_std_return": 44.31970628800741, + "strict_gt3000": 0, + "raw_steps": 20000, + "actions": 20000, + "invalid_actions": 0, + "dropped_actions": 0, + "submitted_observations": 20000, + "dropped_observations": 0, + "mean_action_latency_ms": 145.74751463552934, + "action_latency_p95_ms": 151.4791488018123, + "raw_action_min": -1.0, + "raw_action_max": 1.0, + "raw_actions_requiring_native_clip": 0, + "config_sha256": "5fbaacdb03a541c3e8aaf7dab93bd512930c463b400ecaaa5521ce12f767f51c", + "episode_metrics_sha256": "06ba6b7853dc671731cbcdc173047c961a59f71b74e8cb128ee037f71f67593a", + "steps_sha256": "b2a95ff5c1efe37687ac8bc244de04dd6545d52870b28ef842de4c61261f9cd9", + "actions_sha256": "5f63fae02a1a1e83420ef304bcd828f483aec94f3d0236c88cfc99814ca11ea9", + "raw_step_indices_clock_and_reward_sums_verified": true, + "profile_action_and_latency_provenance_verified": true + } + } + }, + "gate": { + "strict_return_gt": 3000, + "Final_qualifying_episodes": 0, + "training_best_qualifying_episodes": 0, + "state": "SCIENTIFIC_STOP_SELECTED_20_ZERO_QUALIFYING" + }, + "later_stages": { + "E10": "NOT_RUN_AFTER_SELECTED_ZERO_GATE", + "probe12": "NOT_RUN_AFTER_SELECTED_ZERO_GATE", + "data_rollout": "NOT_RUN_AFTER_SELECTED_ZERO_GATE", + "VLA_training": "NOT_RUN", + "HF_publication": "PENDING_MAIN_DATASET_WRITER_SLOT_FOR_STOP_EVIDENCE" + }, + "scientific_stop_sha256": "34c4ff7be8a9d09bfbfe08c969fd454f5afad04b3f942aca509627a22bcb3cbb" +} diff --git a/latency-aware/half-cheetah/vla/starvla-qwenoft-h1/half_cheetah_rgb_state_profile_20260912T023319Z_openvla_native_sft_5k/README.md b/latency-aware/half-cheetah/vla/starvla-qwenoft-h1/half_cheetah_rgb_state_profile_20260912T023319Z_openvla_native_sft_5k/README.md new file mode 100644 index 0000000000000000000000000000000000000000..c5f892943d6a14707f1acbcb72ca1589ca9527d0 --- /dev/null +++ b/latency-aware/half-cheetah/vla/starvla-qwenoft-h1/half_cheetah_rgb_state_profile_20260912T023319Z_openvla_native_sft_5k/README.md @@ -0,0 +1,29 @@ +# half-cheetah / qwenoft + +Training condition: `latency-aware`. Run: `half_cheetah_rgb_state_profile_20260912T023319Z_openvla_native_sft_5k`. + +Copied from [Standard-Pipeline](https://huggingface.co/latency-sensitive-bench/Standard-Pipeline/tree/2208875f92b21e3f8ec2363a30a6705dfe56cc04/half_cheetah/profile_latency/OpenVLA) at `2208875f92b21e3f8ec2363a30a6705dfe56cc04`. + +- Checkpoint: `checkpoints/model.pt` +- Selection: only published VLA checkpoint +- Checkpoint SHA256: `e3c3c8b9fe59d7e20f19c70c584cd1181f65d1caba8079452e7fc3f82c226f52` +- 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/half-cheetah/vla/starvla-qwenoft-h1/half_cheetah_rgb_state_profile_20260912T023319Z_openvla_native_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/half-cheetah/vla/starvla-qwenoft-h1/half_cheetah_rgb_state_profile_20260912T023319Z_openvla_native_sft_5k/checkpoints/model.pt b/latency-aware/half-cheetah/vla/starvla-qwenoft-h1/half_cheetah_rgb_state_profile_20260912T023319Z_openvla_native_sft_5k/checkpoints/model.pt new file mode 100644 index 0000000000000000000000000000000000000000..12ef07c324a6c9e728af8a2c3a6ddeb3bee5ac4f --- /dev/null +++ b/latency-aware/half-cheetah/vla/starvla-qwenoft-h1/half_cheetah_rgb_state_profile_20260912T023319Z_openvla_native_sft_5k/checkpoints/model.pt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e3c3c8b9fe59d7e20f19c70c584cd1181f65d1caba8079452e7fc3f82c226f52 +size 9785049835 diff --git a/latency-aware/half-cheetah/vla/starvla-qwenoft-h1/half_cheetah_rgb_state_profile_20260912T023319Z_openvla_native_sft_5k/config.full.yaml b/latency-aware/half-cheetah/vla/starvla-qwenoft-h1/half_cheetah_rgb_state_profile_20260912T023319Z_openvla_native_sft_5k/config.full.yaml new file mode 100644 index 0000000000000000000000000000000000000000..8a8b229e3964fe8ac8cb726242ac902b76fd5817 --- /dev/null +++ b/latency-aware/half-cheetah/vla/starvla-qwenoft-h1/half_cheetah_rgb_state_profile_20260912T023319Z_openvla_native_sft_5k/config.full.yaml @@ -0,0 +1,384 @@ +framework: + name: QwenOFT + qwenvl: + base_vlm: /mnt/checkpoints/latency-sensitive-bench/openvla/Qwen3-VL-4B-Instruct + attn_implementation: flash_attention_2 + enable_gradient_checkpointing: true + action_model: + action_model_type: MLP + action_dim: 6 + state_dim: 17 + action_hidden_dim: 2560 + future_action_window_size: 0 + past_action_window_size: 0 + loss_type: l1 + state_encoding: discretized_text + task_objective: null + action_horizon: 1 + action_env_dim: 6 + kv_memory: + enabled: false + window: 4 + rollout_len: 8 + packed_train: false + rebased_sink: true +datasets: + vla_data: + dataset_py: lerobot_datasets + include_state: true + data_root_dir: /mnt/results/latency-sensitive-bench/profile-latency-new/half_cheetah/20260912T023319Z-halfcheetah/converted + data_mix: half_cheetah_rgb_state_profile_20260912T023319Z + eval_data_mix: half_cheetah_rgb_state_profile_20260912T023319Z__val + custom_mixtures_path: /mnt/results/latency-sensitive-bench/profile-latency-new/half_cheetah/20260912T023319Z-halfcheetah/converted/_generated_mixtures/half_cheetah_rgb_state_profile_20260912T023319Z.json + action_type: discrete + sequential_step_sampling: false + eval_sequential_step_sampling: null + num_workers: 8 + eval_num_workers: 8 + prefetch_factor: 4 + persistent_workers: true + pin_memory: true + shuffle: true + action_balance: + enabled: false + strategy: balanced_epoch + action_key: action_id + target_flap_fraction: 0.3 + noop_id: 0 + flap_id: 1 + latency_curriculum: + enabled: false + strategy: exclusive + latencies: null + phase_steps: null + phase_distributions: null + new_latency_passes: 1.0 + replay_passes: 0.25 + target_total_passes: 2.0 + final_equalization: true + step_budget_mode: auto + eval_at_phase_end: false + save_at_phase_end: false + computed_plan: null + per_device_batch_size: 16 + load_all_data_for_training: true + num_obs_frames: 1 + image_mode: single + prompt_mode: raw + stitch_grid: + - 2 + - 2 + obs_image_size: null + video_backend: torchvision_av + gymnasium_task_contract: + action_space: + dtype: float32 + high: + - 1.0 + - 1.0 + - 1.0 + - 1.0 + - 1.0 + - 1.0 + labels: + - back_thigh_torque + - back_shin_torque + - back_foot_torque + - front_thigh_torque + - front_shin_torque + - front_foot_torque + low: + - -1.0 + - -1.0 + - -1.0 + - -1.0 + - -1.0 + - -1.0 + type: box + base_prompt: Make the Half Cheetah run forward as fast as possible. Predict + six continuous torques in [-1, 1] ordered as back thigh, back shin, back foot, + front thigh, front shin, and front foot. + env_fps: 10.0 + env_id: LatencyBench/HalfCheetahContinuous-v0 + frame_stack: 1 + make_kwargs: + base_env_id: HalfCheetah-v4 + base_make_kwargs: {} + render_mode: rgb_array + noop_action: + - 0.0 + - 0.0 + - 0.0 + - 0.0 + - 0.0 + - 0.0 + obs_fps: 10.0 + registration_imports: + - latency_bench.envs.gymnasium_half_cheetah + state_space: + labels: + - front_tip_height + - front_tip_angle + - back_thigh_angle + - back_shin_angle + - back_foot_angle + - front_thigh_angle + - front_shin_angle + - front_foot_angle + - front_tip_x_velocity + - front_tip_z_velocity + - front_tip_angular_velocity + - back_thigh_angular_velocity + - back_shin_angular_velocity + - back_foot_angular_velocity + - front_thigh_angular_velocity + - front_shin_angular_velocity + - front_foot_angular_velocity + task_name: half_cheetah_rgb_state + active_action_dim: 6 +dataset: + source_hf: '' + config_name: null + source_subdir: null + converted_name: half_cheetah_rgb_state_profile_20260912T023319Z + single_source_hf: '' + mixed_source_hf: '' + single_converted_name: half_cheetah_rgb_state_profile_20260912T023319Z + mixed_converted_name: half_cheetah_rgb_state_profile_20260912T023319Z + 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: StarVLA/Qwen3-VL-4B-Instruct-Action +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: WORKSPACE_DIR +run_root_dir: /mnt/results/latency-sensitive-bench/profile-latency-new/half_cheetah/20260912T023319Z-halfcheetah/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/half_cheetah/20260912T023319Z-halfcheetah/vla + dataset_local_dir: /mnt/results/latency-sensitive-bench/profile-latency-new/half_cheetah/20260912T023319Z-halfcheetah/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/half_cheetah/20260912T023319Z-halfcheetah/converted/half_cheetah_rgb_state_profile_20260912T023319Z/latency_prompt_map.json + mode: single + values: + - 1 + - 2 + task: gymnasium + gymnasium: + task_contract: + action_space: + dtype: float32 + high: + - 1.0 + - 1.0 + - 1.0 + - 1.0 + - 1.0 + - 1.0 + labels: + - back_thigh_torque + - back_shin_torque + - back_foot_torque + - front_thigh_torque + - front_shin_torque + - front_foot_torque + low: + - -1.0 + - -1.0 + - -1.0 + - -1.0 + - -1.0 + - -1.0 + type: box + base_prompt: Make the Half Cheetah run forward as fast as possible. Predict + six continuous torques in [-1, 1] ordered as back thigh, back shin, back foot, + front thigh, front shin, and front foot. + env_fps: 10.0 + env_id: LatencyBench/HalfCheetahContinuous-v0 + frame_stack: 1 + make_kwargs: + base_env_id: HalfCheetah-v4 + base_make_kwargs: {} + render_mode: rgb_array + noop_action: + - 0.0 + - 0.0 + - 0.0 + - 0.0 + - 0.0 + - 0.0 + obs_fps: 10.0 + registration_imports: + - latency_bench.envs.gymnasium_half_cheetah + state_space: + labels: + - front_tip_height + - front_tip_angle + - back_thigh_angle + - back_shin_angle + - back_foot_angle + - front_thigh_angle + - front_shin_angle + - front_foot_angle + - front_tip_x_velocity + - front_tip_z_velocity + - front_tip_angular_velocity + - back_thigh_angular_velocity + - back_shin_angular_velocity + - back_foot_angular_velocity + - front_thigh_angular_velocity + - front_shin_angular_velocity + - front_foot_angular_velocity + task_name: half_cheetah_rgb_state + task_name: half_cheetah_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: half_cheetah_rgb_state_profile_20260912T023319Z_openvla_native_sft_5k +output_dir: /mnt/results/latency-sensitive-bench/profile-latency-new/half_cheetah/20260912T023319Z-halfcheetah/vla/half_cheetah_rgb_state_profile_20260912T023319Z_openvla_native_sft_5k +config_yaml: /mnt/results/latency-sensitive-bench/profile-latency-new/half_cheetah/20260912T023319Z-halfcheetah/h100/vla_train.yaml +is_debug: false +version_id: '0.21' diff --git a/latency-aware/half-cheetah/vla/starvla-qwenoft-h1/half_cheetah_rgb_state_profile_20260912T023319Z_openvla_native_sft_5k/config.yaml b/latency-aware/half-cheetah/vla/starvla-qwenoft-h1/half_cheetah_rgb_state_profile_20260912T023319Z_openvla_native_sft_5k/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..8a8b229e3964fe8ac8cb726242ac902b76fd5817 --- /dev/null +++ b/latency-aware/half-cheetah/vla/starvla-qwenoft-h1/half_cheetah_rgb_state_profile_20260912T023319Z_openvla_native_sft_5k/config.yaml @@ -0,0 +1,384 @@ +framework: + name: QwenOFT + qwenvl: + base_vlm: /mnt/checkpoints/latency-sensitive-bench/openvla/Qwen3-VL-4B-Instruct + attn_implementation: flash_attention_2 + enable_gradient_checkpointing: true + action_model: + action_model_type: MLP + action_dim: 6 + state_dim: 17 + action_hidden_dim: 2560 + future_action_window_size: 0 + past_action_window_size: 0 + loss_type: l1 + state_encoding: discretized_text + task_objective: null + action_horizon: 1 + action_env_dim: 6 + kv_memory: + enabled: false + window: 4 + rollout_len: 8 + packed_train: false + rebased_sink: true +datasets: + vla_data: + dataset_py: lerobot_datasets + include_state: true + data_root_dir: /mnt/results/latency-sensitive-bench/profile-latency-new/half_cheetah/20260912T023319Z-halfcheetah/converted + data_mix: half_cheetah_rgb_state_profile_20260912T023319Z + eval_data_mix: half_cheetah_rgb_state_profile_20260912T023319Z__val + custom_mixtures_path: /mnt/results/latency-sensitive-bench/profile-latency-new/half_cheetah/20260912T023319Z-halfcheetah/converted/_generated_mixtures/half_cheetah_rgb_state_profile_20260912T023319Z.json + action_type: discrete + sequential_step_sampling: false + eval_sequential_step_sampling: null + num_workers: 8 + eval_num_workers: 8 + prefetch_factor: 4 + persistent_workers: true + pin_memory: true + shuffle: true + action_balance: + enabled: false + strategy: balanced_epoch + action_key: action_id + target_flap_fraction: 0.3 + noop_id: 0 + flap_id: 1 + latency_curriculum: + enabled: false + strategy: exclusive + latencies: null + phase_steps: null + phase_distributions: null + new_latency_passes: 1.0 + replay_passes: 0.25 + target_total_passes: 2.0 + final_equalization: true + step_budget_mode: auto + eval_at_phase_end: false + save_at_phase_end: false + computed_plan: null + per_device_batch_size: 16 + load_all_data_for_training: true + num_obs_frames: 1 + image_mode: single + prompt_mode: raw + stitch_grid: + - 2 + - 2 + obs_image_size: null + video_backend: torchvision_av + gymnasium_task_contract: + action_space: + dtype: float32 + high: + - 1.0 + - 1.0 + - 1.0 + - 1.0 + - 1.0 + - 1.0 + labels: + - back_thigh_torque + - back_shin_torque + - back_foot_torque + - front_thigh_torque + - front_shin_torque + - front_foot_torque + low: + - -1.0 + - -1.0 + - -1.0 + - -1.0 + - -1.0 + - -1.0 + type: box + base_prompt: Make the Half Cheetah run forward as fast as possible. Predict + six continuous torques in [-1, 1] ordered as back thigh, back shin, back foot, + front thigh, front shin, and front foot. + env_fps: 10.0 + env_id: LatencyBench/HalfCheetahContinuous-v0 + frame_stack: 1 + make_kwargs: + base_env_id: HalfCheetah-v4 + base_make_kwargs: {} + render_mode: rgb_array + noop_action: + - 0.0 + - 0.0 + - 0.0 + - 0.0 + - 0.0 + - 0.0 + obs_fps: 10.0 + registration_imports: + - latency_bench.envs.gymnasium_half_cheetah + state_space: + labels: + - front_tip_height + - front_tip_angle + - back_thigh_angle + - back_shin_angle + - back_foot_angle + - front_thigh_angle + - front_shin_angle + - front_foot_angle + - front_tip_x_velocity + - front_tip_z_velocity + - front_tip_angular_velocity + - back_thigh_angular_velocity + - back_shin_angular_velocity + - back_foot_angular_velocity + - front_thigh_angular_velocity + - front_shin_angular_velocity + - front_foot_angular_velocity + task_name: half_cheetah_rgb_state + active_action_dim: 6 +dataset: + source_hf: '' + config_name: null + source_subdir: null + converted_name: half_cheetah_rgb_state_profile_20260912T023319Z + single_source_hf: '' + mixed_source_hf: '' + single_converted_name: half_cheetah_rgb_state_profile_20260912T023319Z + mixed_converted_name: half_cheetah_rgb_state_profile_20260912T023319Z + 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: StarVLA/Qwen3-VL-4B-Instruct-Action +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: WORKSPACE_DIR +run_root_dir: /mnt/results/latency-sensitive-bench/profile-latency-new/half_cheetah/20260912T023319Z-halfcheetah/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/half_cheetah/20260912T023319Z-halfcheetah/vla + dataset_local_dir: /mnt/results/latency-sensitive-bench/profile-latency-new/half_cheetah/20260912T023319Z-halfcheetah/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/half_cheetah/20260912T023319Z-halfcheetah/converted/half_cheetah_rgb_state_profile_20260912T023319Z/latency_prompt_map.json + mode: single + values: + - 1 + - 2 + task: gymnasium + gymnasium: + task_contract: + action_space: + dtype: float32 + high: + - 1.0 + - 1.0 + - 1.0 + - 1.0 + - 1.0 + - 1.0 + labels: + - back_thigh_torque + - back_shin_torque + - back_foot_torque + - front_thigh_torque + - front_shin_torque + - front_foot_torque + low: + - -1.0 + - -1.0 + - -1.0 + - -1.0 + - -1.0 + - -1.0 + type: box + base_prompt: Make the Half Cheetah run forward as fast as possible. Predict + six continuous torques in [-1, 1] ordered as back thigh, back shin, back foot, + front thigh, front shin, and front foot. + env_fps: 10.0 + env_id: LatencyBench/HalfCheetahContinuous-v0 + frame_stack: 1 + make_kwargs: + base_env_id: HalfCheetah-v4 + base_make_kwargs: {} + render_mode: rgb_array + noop_action: + - 0.0 + - 0.0 + - 0.0 + - 0.0 + - 0.0 + - 0.0 + obs_fps: 10.0 + registration_imports: + - latency_bench.envs.gymnasium_half_cheetah + state_space: + labels: + - front_tip_height + - front_tip_angle + - back_thigh_angle + - back_shin_angle + - back_foot_angle + - front_thigh_angle + - front_shin_angle + - front_foot_angle + - front_tip_x_velocity + - front_tip_z_velocity + - front_tip_angular_velocity + - back_thigh_angular_velocity + - back_shin_angular_velocity + - back_foot_angular_velocity + - front_thigh_angular_velocity + - front_shin_angular_velocity + - front_foot_angular_velocity + task_name: half_cheetah_rgb_state + task_name: half_cheetah_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: half_cheetah_rgb_state_profile_20260912T023319Z_openvla_native_sft_5k +output_dir: /mnt/results/latency-sensitive-bench/profile-latency-new/half_cheetah/20260912T023319Z-halfcheetah/vla/half_cheetah_rgb_state_profile_20260912T023319Z_openvla_native_sft_5k +config_yaml: /mnt/results/latency-sensitive-bench/profile-latency-new/half_cheetah/20260912T023319Z-halfcheetah/h100/vla_train.yaml +is_debug: false +version_id: '0.21' diff --git a/latency-aware/half-cheetah/vla/starvla-qwenoft-h1/half_cheetah_rgb_state_profile_20260912T023319Z_openvla_native_sft_5k/dataset_statistics.json b/latency-aware/half-cheetah/vla/starvla-qwenoft-h1/half_cheetah_rgb_state_profile_20260912T023319Z_openvla_native_sft_5k/dataset_statistics.json new file mode 100644 index 0000000000000000000000000000000000000000..b765195abd4472cb4bcba69b8e38b20001fbac48 --- /dev/null +++ b/latency-aware/half-cheetah/vla/starvla-qwenoft-h1/half_cheetah_rgb_state_profile_20260912T023319Z_openvla_native_sft_5k/dataset_statistics.json @@ -0,0 +1,180 @@ +{ + "new_embodiment": { + "action": { + "mean": [ + 0.10770988464355469, + -0.22736652195453644, + 0.10839294642210007, + 0.3289090692996979, + 0.3570038080215454, + -0.23193077743053436 + ], + "std": [ + 0.9823699593544006, + 0.8835218548774719, + 0.8981277346611023, + 0.7482634782791138, + 0.6214219331741333, + 0.7868151664733887 + ], + "max": [ + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0 + ], + "min": [ + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0 + ], + "q01": [ + -1.0, + -1.0, + -1.0, + -1.0, + -0.9372798657417297, + -1.0 + ], + "q99": [ + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0 + ], + "mask": [ + true, + true, + true, + true, + true, + true + ] + }, + "state": { + "mean": [ + -0.10728839784860611, + -0.5624207854270935, + -0.04555436596274376, + -0.1454501748085022, + -0.11111028492450714, + 0.421887069940567, + 0.13110347092151642, + -0.1150883361697197, + 0.4590475261211395, + 0.026997756212949753, + 0.01987910084426403, + 0.15814617276191711, + -0.057850856333971024, + 0.1238384023308754, + -0.013364912942051888, + 0.016042476519942284, + -0.1159631609916687 + ], + "std": [ + 0.11889327317476273, + 0.08449525386095047, + 0.5472133755683899, + 0.49602779746055603, + 0.5866584181785583, + 0.3321983516216278, + 0.4677083492279053, + 0.5666539072990417, + 0.22859491407871246, + 0.2627943754196167, + 0.26507872343063354, + 0.5673505663871765, + 0.3777157962322235, + 0.5484620928764343, + 0.3395417034626007, + 0.4824333190917969, + 0.5185849666595459 + ], + "max": [ + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0 + ], + "min": [ + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0 + ], + "q01": [ + -0.33906219065189364, + -0.6886490625143051, + -0.9395584189891815, + -0.9324191778898239, + -0.9777716618776321, + -0.31617344081401827, + -0.6568031430244445, + -0.9545256400108337, + -0.3354434877634048, + -0.6166862392425537, + -0.7010972607135773, + -0.8671129012107849, + -0.8751532965898514, + -0.8775771206617355, + -0.7591884690523147, + -0.82644451379776, + -0.9271202409267425 + ], + "q99": [ + 0.20573618650436362, + -0.3561570644378664, + 0.8709963262081146, + 0.7403977537155135, + 0.8409473288059234, + 0.8861571657657623, + 0.8961218094825741, + 0.797587602138519, + 0.8380255770683288, + 0.6127554690837859, + 0.5165472745895382, + 0.902275366783142, + 0.594376227855682, + 0.945205647945404, + 0.6881502497196197, + 0.781806918382644, + 0.7805549800395957 + ] + }, + "num_transitions": 90000, + "num_trajectories": 90 + } +} \ No newline at end of file diff --git a/latency-aware/half-cheetah/vla/starvla-qwenoft-h1/half_cheetah_rgb_state_profile_20260912T023319Z_openvla_native_sft_5k/dataset_statistics_eval.json b/latency-aware/half-cheetah/vla/starvla-qwenoft-h1/half_cheetah_rgb_state_profile_20260912T023319Z_openvla_native_sft_5k/dataset_statistics_eval.json new file mode 100644 index 0000000000000000000000000000000000000000..1a3140c94f712cb4b2fffe894f2b60ab8e514fb0 --- /dev/null +++ b/latency-aware/half-cheetah/vla/starvla-qwenoft-h1/half_cheetah_rgb_state_profile_20260912T023319Z_openvla_native_sft_5k/dataset_statistics_eval.json @@ -0,0 +1,180 @@ +{ + "new_embodiment": { + "action": { + "mean": [ + 0.11145979166030884, + -0.214508056640625, + 0.11074115335941315, + 0.327595055103302, + 0.3518350422382355, + -0.23154450953006744 + ], + "std": [ + 0.9813211560249329, + 0.8875551819801331, + 0.8911529183387756, + 0.7460984587669373, + 0.620377242565155, + 0.7883656620979309 + ], + "max": [ + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0 + ], + "min": [ + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0 + ], + "q01": [ + -1.0, + -1.0, + -1.0, + -1.0, + -0.9503968459367752, + -1.0 + ], + "q99": [ + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0 + ], + "mask": [ + true, + true, + true, + true, + true, + true + ] + }, + "state": { + "mean": [ + -0.10398252308368683, + -0.5657179951667786, + -0.041583992540836334, + -0.13961662352085114, + -0.10986241698265076, + 0.41965508460998535, + 0.1287042796611786, + -0.12071704119443893, + 0.435773640871048, + 0.026984786614775658, + 0.021505804732441902, + 0.15837563574314117, + -0.057794880121946335, + 0.12203305959701538, + -0.014990058727562428, + 0.016157714650034904, + -0.11708204448223114 + ], + "std": [ + 0.1244068369269371, + 0.07347161322832108, + 0.5536452531814575, + 0.4919765293598175, + 0.5864324569702148, + 0.33482614159584045, + 0.47133588790893555, + 0.5699070692062378, + 0.2429482787847519, + 0.26402586698532104, + 0.2640051245689392, + 0.5619649291038513, + 0.37026268243789673, + 0.5385760068893433, + 0.33654022216796875, + 0.4757087230682373, + 0.5125513076782227 + ], + "max": [ + 0.8230167627334595, + -0.0014048218727111816, + 0.9493592977523804, + 0.9882220029830933, + 1.0, + 0.9731433391571045, + 0.9668114185333252, + 0.9491677284240723, + 0.9851869344711304, + 0.8379104137420654, + 1.0, + 0.9966773986816406, + 0.7800441980361938, + 0.9713037014007568, + 0.8712660074234009, + 0.9649665355682373, + 0.984099268913269 + ], + "min": [ + -0.48008835315704346, + -0.8673253655433655, + -0.9924408793449402, + -0.97048419713974, + -1.0, + -0.8882791996002197, + -0.9792449474334717, + -0.9918810129165649, + -0.6799987554550171, + -0.9922735095024109, + -0.9111155867576599, + -1.0, + -0.9412549138069153, + -1.0, + -0.9556921720504761, + -0.9734961986541748, + -0.9983583688735962 + ], + "q01": [ + -0.3424838989973068, + -0.7226481068134307, + -0.9416929757595062, + -0.9292510569095611, + -0.9770489847660064, + -0.33539102911949153, + -0.6675442630052566, + -0.9519808453321457, + -0.3450982230901718, + -0.6134166467189789, + -0.7010219645500183, + -0.8668506604433059, + -0.8667900818586349, + -0.8788304537534714, + -0.7496186256408691, + -0.8328363233804703, + -0.9277301490306854 + ], + "q99": [ + 0.27186036467552194, + -0.3383111631870269, + 0.871550648212433, + 0.7503189408779147, + 0.841783242225647, + 0.8875537037849427, + 0.8962072730064392, + 0.8053963160514832, + 0.8340123009681703, + 0.6146782374382019, + 0.5173484361171725, + 0.9027680957317352, + 0.595805132389069, + 0.9427596497535706, + 0.681271653175354, + 0.7714509844779969, + 0.760542529821396 + ] + }, + "num_transitions": 10000, + "num_trajectories": 10 + } +} \ No newline at end of file diff --git a/latency-aware/half-cheetah/vla/starvla-qwenoft-h1/half_cheetah_rgb_state_profile_20260912T023319Z_openvla_native_sft_5k/latency_prompt_map.json b/latency-aware/half-cheetah/vla/starvla-qwenoft-h1/half_cheetah_rgb_state_profile_20260912T023319Z_openvla_native_sft_5k/latency_prompt_map.json new file mode 100644 index 0000000000000000000000000000000000000000..1e5dd7db213425961474c33800d566f3cc41283f --- /dev/null +++ b/latency-aware/half-cheetah/vla/starvla-qwenoft-h1/half_cheetah_rgb_state_profile_20260912T023319Z_openvla_native_sft_5k/latency_prompt_map.json @@ -0,0 +1,12 @@ +{ + "1": { + "prompt": "Make the Half Cheetah run forward as fast as possible. Predict six continuous torques in [-1, 1] ordered as back thigh, back shin, back foot, front thigh, front shin, and front 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": "Make the Half Cheetah run forward as fast as possible. Predict six continuous torques in [-1, 1] ordered as back thigh, back shin, back foot, front thigh, front shin, and front 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/half-cheetah/vla/starvla-qwenoft-h1/half_cheetah_rgb_state_profile_20260912T023319Z_openvla_native_sft_5k/manifest.json b/latency-aware/half-cheetah/vla/starvla-qwenoft-h1/half_cheetah_rgb_state_profile_20260912T023319Z_openvla_native_sft_5k/manifest.json new file mode 100644 index 0000000000000000000000000000000000000000..3e5b64146f34b83f621fbbc5ddc05b3af0bd6b74 --- /dev/null +++ b/latency-aware/half-cheetah/vla/starvla-qwenoft-h1/half_cheetah_rgb_state_profile_20260912T023319Z_openvla_native_sft_5k/manifest.json @@ -0,0 +1,178 @@ +{ + "dataset_name": "half_cheetah_rgb_state_profile_20260912T023319Z", + "env_name": "half_cheetah_rgb_state", + "episodes": 90, + "frames": 90000, + "task_prompts": [ + "Make the Half Cheetah run forward as fast as possible. Predict six continuous torques in [-1, 1] ordered as back thigh, back shin, back foot, front thigh, front shin, and front 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.", + "Make the Half Cheetah run forward as fast as possible. Predict six continuous torques in [-1, 1] ordered as back thigh, back shin, back foot, front thigh, front shin, and front 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/half_cheetah/20260912T023319Z-halfcheetah/raw", + "integration_name": "gymnasium", + "task_name": "half_cheetah_rgb_state", + "action_layout": "gymnasium_continuous_v1", + "action_labels": [ + "back_thigh_torque", + "back_shin_torque", + "back_foot_torque", + "front_thigh_torque", + "front_shin_torque", + "front_foot_torque" + ], + "carrier_action_labels": [ + "back_thigh_torque", + "back_shin_torque", + "back_foot_torque", + "front_thigh_torque", + "front_shin_torque", + "front_foot_torque" + ], + "action_dim": 6, + "active_action_dim": 6, + "action_carrier": "native", + "rows_unit": "decision_step", + "obs_fps": 10.0, + "obs_stride_raw_frames": 1, + "uses_state": true, + "state_dim": 17, + "state_labels": [ + "front_tip_height", + "front_tip_angle", + "back_thigh_angle", + "back_shin_angle", + "back_foot_angle", + "front_thigh_angle", + "front_shin_angle", + "front_foot_angle", + "front_tip_x_velocity", + "front_tip_z_velocity", + "front_tip_angular_velocity", + "back_thigh_angular_velocity", + "back_shin_angular_velocity", + "back_foot_angular_velocity", + "front_thigh_angular_velocity", + "front_shin_angular_velocity", + "front_foot_angular_velocity" + ], + "state_normalization": { + "type": "min_max", + "min": [ + -0.5772290229797363, + -0.911627471446991, + -0.7220134139060974, + -0.8680613040924072, + -0.5131105184555054, + -0.9485418200492859, + -0.9157634377479553, + -0.658291757106781, + -3.6468114852905273, + -3.9537510871887207, + -6.7527337074279785, + -26.907974243164062, + -25.66440200805664, + -22.458248138427734, + -22.975290298461914, + -26.29881477355957, + -18.896934509277344 + ], + "max": [ + 0.6679929494857788, + 3.516580104827881, + 1.0322039127349854, + 0.8538315296173096, + 0.8270522356033325, + 0.8448535799980164, + 1.0135595798492432, + 0.7175425291061401, + 9.11288070678711, + 3.604336977005005, + 6.252868175506592, + 19.51249122619629, + 28.126272201538086, + 17.998640060424805, + 24.064804077148438, + 25.965932846069336, + 24.8583927154541 + ] + }, + "robot_type": "rl_games_gymnasium", + "latency_prompt_map_path": "/mnt/results/latency-sensitive-bench/profile-latency-new/half_cheetah/20260912T023319Z-halfcheetah/converted/half_cheetah_rgb_state_profile_20260912T023319Z/latency_prompt_map.json", + "custom_mixtures_path": "/mnt/results/latency-sensitive-bench/profile-latency-new/half_cheetah/20260912T023319Z-halfcheetah/converted/_generated_mixtures/half_cheetah_rgb_state_profile_20260912T023319Z.json", + "gymnasium_task": { + "action_space": { + "dtype": "float32", + "high": [ + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0 + ], + "labels": [ + "back_thigh_torque", + "back_shin_torque", + "back_foot_torque", + "front_thigh_torque", + "front_shin_torque", + "front_foot_torque" + ], + "low": [ + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0 + ], + "type": "box" + }, + "base_prompt": "Make the Half Cheetah run forward as fast as possible. Predict six continuous torques in [-1, 1] ordered as back thigh, back shin, back foot, front thigh, front shin, and front foot.", + "env_fps": 10.0, + "env_id": "LatencyBench/HalfCheetahContinuous-v0", + "frame_stack": 1, + "make_kwargs": { + "base_env_id": "HalfCheetah-v4", + "base_make_kwargs": {}, + "render_mode": "rgb_array" + }, + "noop_action": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ], + "obs_fps": 10.0, + "registration_imports": [ + "latency_bench.envs.gymnasium_half_cheetah" + ], + "state_space": { + "labels": [ + "front_tip_height", + "front_tip_angle", + "back_thigh_angle", + "back_shin_angle", + "back_foot_angle", + "front_thigh_angle", + "front_shin_angle", + "front_foot_angle", + "front_tip_x_velocity", + "front_tip_z_velocity", + "front_tip_angular_velocity", + "back_thigh_angular_velocity", + "back_shin_angular_velocity", + "back_foot_angular_velocity", + "front_thigh_angular_velocity", + "front_shin_angular_velocity", + "front_foot_angular_velocity" + ] + }, + "task_name": "half_cheetah_rgb_state" + }, + "validation_dataset_name": "half_cheetah_rgb_state_profile_20260912T023319Z__val", + "validation_episodes": 10, + "validation_frames": 10000 +} \ No newline at end of file diff --git a/latency-aware/half-cheetah/vla/starvla-qwenoft-h1/half_cheetah_rgb_state_profile_20260912T023319Z_openvla_native_sft_5k/profile/DONE b/latency-aware/half-cheetah/vla/starvla-qwenoft-h1/half_cheetah_rgb_state_profile_20260912T023319Z_openvla_native_sft_5k/profile/DONE new file mode 100644 index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391 diff --git a/latency-aware/half-cheetah/vla/starvla-qwenoft-h1/half_cheetah_rgb_state_profile_20260912T023319Z_openvla_native_sft_5k/profile/hardware.json b/latency-aware/half-cheetah/vla/starvla-qwenoft-h1/half_cheetah_rgb_state_profile_20260912T023319Z_openvla_native_sft_5k/profile/hardware.json new file mode 100644 index 0000000000000000000000000000000000000000..e8204e81807044975c0e0a0953d16d6835192f31 --- /dev/null +++ b/latency-aware/half-cheetah/vla/starvla-qwenoft-h1/half_cheetah_rgb_state_profile_20260912T023319Z_openvla_native_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/half-cheetah/vla/starvla-qwenoft-h1/half_cheetah_rgb_state_profile_20260912T023319Z_openvla_native_sft_5k/profile/latency_burst_model.json b/latency-aware/half-cheetah/vla/starvla-qwenoft-h1/half_cheetah_rgb_state_profile_20260912T023319Z_openvla_native_sft_5k/profile/latency_burst_model.json new file mode 100644 index 0000000000000000000000000000000000000000..aa5e6d42f5c6a718266afa349f01bef857224531 --- /dev/null +++ b/latency-aware/half-cheetah/vla/starvla-qwenoft-h1/half_cheetah_rgb_state_profile_20260912T023319Z_openvla_native_sft_5k/profile/latency_burst_model.json @@ -0,0 +1,1329 @@ +{ + "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, + 1 + ], + "burst_merge_gap_records": 30, + "burst_rank_processes": [ + { + "draw_count": 4, + "dwell_length_spearman_rho": 0.0, + "level_ms": { + "distribution_type": "inverse_cdf", + "latency_ms": [ + 91.29252696037292, + 91.29252696037292, + 91.29252696037292, + 91.29252696037292, + 91.29252696037292, + 91.29252696037292, + 91.29252696037292, + 91.29252696037292, + 91.29252696037292, + 91.29252696037292, + 91.29252696037292, + 91.29252696037292, + 91.29252696037292, + 91.29252696037292, + 91.29252696037292, + 91.29252696037292, + 91.29252696037292, + 91.29252696037292, + 91.29252696037292, + 91.29252696037292, + 91.29252696037292, + 91.29252696037292, + 91.29252696037292, + 91.29252696037292, + 91.48080760002136, + 91.85736887931824, + 92.23393015861511, + 92.61049143791199, + 92.98705271720887, + 93.36361399650573, + 93.74017527580261, + 94.1167365550995, + 94.49329783439636, + 94.86985911369324, + 95.24642039299012, + 95.62298167228698, + 95.99954295158386, + 96.37610423088074, + 96.75266551017761, + 97.12922678947449, + 97.50578806877137, + 97.88234934806823, + 98.25891062736511, + 98.63547190666199, + 99.01203318595887, + 99.38859446525574, + 99.76515574455262, + 100.14171702384948, + 100.51827830314636, + 100.78247780323028, + 100.93431552410125, + 101.08615324497222, + 101.2379909658432, + 101.38982868671417, + 101.54166640758514, + 101.69350412845611, + 101.84534184932708, + 101.99717957019806, + 102.14901729106903, + 102.30085501194, + 102.45269273281097, + 102.60453045368195, + 102.75636817455292, + 102.90820589542389, + 103.06004361629486, + 103.21188133716583, + 103.3637190580368, + 103.51555677890778, + 103.66739449977875, + 103.81923222064972, + 103.9710699415207, + 104.12290766239167, + 104.27474538326264, + 104.42658310413361, + 104.64846988677978, + 104.94040573120117, + 105.23234157562256, + 105.52427742004394, + 105.81621326446533, + 106.10814910888672, + 106.40008495330811, + 106.6920207977295, + 106.98395664215087, + 107.27589248657226, + 107.56782833099365, + 107.85976417541504, + 108.15170001983643, + 108.44363586425781, + 108.7355717086792, + 109.02750755310059, + 109.31944339752198, + 109.61137924194335, + 109.90331508636474, + 110.19525093078613, + 110.48718677520752, + 110.77912261962891, + 111.0710584640503, + 111.36299430847168, + 111.65493015289307, + 111.80089807510376, + 111.80089807510376, + 111.80089807510376, + 111.80089807510376, + 111.80089807510376, + 111.80089807510376, + 111.80089807510376, + 111.80089807510376, + 111.80089807510376, + 111.80089807510376, + 111.80089807510376, + 111.80089807510376, + 111.80089807510376, + 111.80089807510376, + 111.80089807510376, + 111.80089807510376, + 111.80089807510376, + 111.80089807510376, + 111.80089807510376, + 111.80089807510376, + 111.80089807510376, + 111.80089807510376, + 111.80089807510376, + 111.80089807510376 + ], + "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": 4 + } + ], + "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 + } + }, + { + "dwell_length": 1, + "slot_to_rank": { + "0": 0 + } + } + ], + "model_type": "hidden_regime", + "pre_worker_time_ms": { + "distribution_type": "inverse_cdf", + "latency_ms": [ + 0.28966736793518066, + 0.29661628777980803, + 0.30831514048576353, + 0.31727498722076414, + 0.338593578338623, + 0.3461133544445038, + 0.3511757125854492, + 0.3563822603225708, + 0.3618279318809509, + 0.3671014323234558, + 0.3725587911605835, + 0.3779346108436584, + 0.38324137926101687, + 0.4556529235839844, + 0.509889326095581, + 0.5281284952163696, + 0.5403126955032349, + 0.5479220390319824, + 0.5539253664016723, + 0.5594458866119385, + 0.5640100693702698, + 0.5677289962768555, + 0.5709687566757202, + 0.5738528633117675, + 0.5767292833328247, + 0.5795547199249268, + 0.5822810411453248, + 0.5850470447540284, + 0.5876750636100769, + 0.5903967952728272, + 0.592800943851471, + 0.5953300476074219, + 0.5980806803703308, + 0.6004024124145508, + 0.6025746560096741, + 0.6047077465057373, + 0.6067898273468018, + 0.6086850166320801, + 0.6106685948371887, + 0.612722282409668, + 0.6146113967895508, + 0.6165520191192627, + 0.6182403135299682, + 0.6199717569351196, + 0.6218879985809326, + 0.6237829494476318, + 0.6255278944969177, + 0.6274885368347168, + 0.6292352557182312, + 0.6305854225158691, + 0.6320900011062622, + 0.6335686922073365, + 0.6350811290740966, + 0.6365709590911866, + 0.6379385328292847, + 0.6395668697357177, + 0.640962290763855, + 0.642304630279541, + 0.6439359402656555, + 0.645436725616455, + 0.6472484803199768, + 0.6488471031188965, + 0.6504889988899231, + 0.6521338987350463, + 0.6537751984596253, + 0.655202226638794, + 0.6567079544067383, + 0.6582039451599122, + 0.6598467111587524, + 0.6612679958343506, + 0.6625553750991822, + 0.6639010190963746, + 0.6652205157279968, + 0.6667079925537109, + 0.6680989265441895, + 0.6693964767456054, + 0.6709147572517395, + 0.6722634792327881, + 0.6739575719833374, + 0.6756732177734375, + 0.6773415374755859, + 0.6790953636169433, + 0.6808587408065796, + 0.6830189752578736, + 0.6850442719459534, + 0.687199592590332, + 0.68963623046875, + 0.6920534992218017, + 0.6945298147201538, + 0.6971553516387939, + 0.6997041010856628, + 0.7023355007171631, + 0.7050988674163818, + 0.7076102638244629, + 0.7099640583992004, + 0.7127210712432861, + 0.7153138279914856, + 0.7181663990020751, + 0.7208955335617065, + 0.7237677955627442, + 0.7270624494552612, + 0.7306365489959716, + 0.7349855732917786, + 0.739589581489563, + 0.7454291129112245, + 0.7518837356567382, + 0.7594980955123901, + 0.7684542274475097, + 0.7794348359107971, + 0.799935255050659, + 0.8361522078514098, + 0.841301315307617, + 0.8511774196624763, + 0.858581780910492, + 0.8691914596557617, + 0.8781117153167721, + 0.8919882164001455, + 0.9130556282997128, + 0.9367391667366031, + 0.9679305844306949, + 1.0121961189508684, + 1.2404262654781233, + 1.388310194015503 + ], + "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": 4, + "calm": 29993 + }, + "regime_transition_counts": { + "burst": { + "burst": 0, + "calm": 4 + }, + "calm": { + "burst": 4, + "calm": 29984 + } + }, + "reset_scope": "session", + "schema_version": 12, + "spike_median_multiplier": 1.25, + "spike_threshold_ms_by_worker_slot": { + "0": 88.3233892917633 + }, + "worker_count": 1 +} diff --git a/latency-aware/half-cheetah/vla/starvla-qwenoft-h1/half_cheetah_rgb_state_profile_20260912T023319Z_openvla_native_sft_5k/profile/latency_distribution.json b/latency-aware/half-cheetah/vla/starvla-qwenoft-h1/half_cheetah_rgb_state_profile_20260912T023319Z_openvla_native_sft_5k/profile/latency_distribution.json new file mode 100644 index 0000000000000000000000000000000000000000..b5dc4b58602cca48b6383606d7ca634ea28cec4e --- /dev/null +++ b/latency-aware/half-cheetah/vla/starvla-qwenoft-h1/half_cheetah_rgb_state_profile_20260912T023319Z_openvla_native_sft_5k/profile/latency_distribution.json @@ -0,0 +1,1027 @@ +{ + "schema_version": 3, + "worker_slots": { + "0": { + "all": { + "count": 29997, + "observation_to_action_latency_ms": { + "distribution_type": "inverse_cdf", + "latency_ms": [ + 65.39556503295898, + 65.90319257864952, + 66.34795825266838, + 66.52524732637406, + 66.76708299684525, + 67.02418623113633, + 67.17869547843934, + 67.26114100694656, + 67.40597652959823, + 67.50079720020294, + 67.60124465465546, + 67.68513790631295, + 67.77476126432418, + 69.4245816898346, + 70.02091380119323, + 70.28190787315368, + 70.46180428266526, + 70.5905613231659, + 70.6786233663559, + 70.76199055194854, + 70.82302750349045, + 70.87952952384948, + 70.93279217720031, + 70.97876193523408, + 71.02215361356735, + 71.05534663200379, + 71.08903501033782, + 71.12099472522736, + 71.14889664411545, + 71.17835633277893, + 71.20590738773346, + 71.23764076232911, + 71.26494863510132, + 71.29445910453796, + 71.31955314636231, + 71.3448329114914, + 71.370949447155, + 71.39764987945557, + 71.42063557147979, + 71.44203779697418, + 71.46336093187332, + 71.48176407814026, + 71.50155672311783, + 71.52068632602692, + 71.53738250017166, + 71.55426707267762, + 71.5741131901741, + 71.59390948295594, + 71.61351494789123, + 71.63126242637634, + 71.65004153251648, + 71.6698139667511, + 71.68906754732132, + 71.7093546485901, + 71.72822314500809, + 71.74877698421479, + 71.77002807855607, + 71.79165102005005, + 71.81265570163727, + 71.83448543548585, + 71.85390951871872, + 71.872318983078, + 71.88918613672257, + 71.90831651687623, + 71.92598719120025, + 71.94410138130188, + 71.96174719333649, + 71.97731454849243, + 71.99444168806076, + 72.01067091941833, + 72.02784201383591, + 72.04347264766693, + 72.06015002250672, + 72.07703278541565, + 72.09355116605758, + 72.11092311382293, + 72.12674032449722, + 72.14582781791687, + 72.1675558257103, + 72.18711300849914, + 72.20661722660064, + 72.22612509727477, + 72.24460446119309, + 72.26352282524108, + 72.28893879890443, + 72.3113342666626, + 72.33358442783356, + 72.35914145946502, + 72.38095099925995, + 72.40431155204773, + 72.42851207256317, + 72.45446176528931, + 72.48137830018997, + 72.5092474937439, + 72.53562726974488, + 72.56389167308808, + 72.59069626331329, + 72.62336647033692, + 72.65377742528915, + 72.68993539810181, + 72.72776058673858, + 72.770436668396, + 72.82028721570968, + 72.8761532688141, + 72.93663378715515, + 73.01012805938721, + 73.0942009806633, + 73.18615770816803, + 73.34689612150193, + 73.59420322418212, + 74.26737931489944, + 74.39098821926116, + 74.46888185977936, + 74.59232448458671, + 74.7213032426834, + 74.85509558796882, + 75.02051337623595, + 75.32228709888452, + 75.84211740064622, + 78.21195779466629, + 80.96420217287572, + 96.67231945006563, + 112.48595118522644 + ], + "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": 0.9963489496805269, + "worker_service_time_ms": { + "distribution_type": "inverse_cdf", + "latency_ms": [ + 65.03665804862976, + 65.55401456711293, + 65.98787849330903, + 66.16043975687028, + 66.39510622596741, + 66.65682306718826, + 66.81381133651733, + 66.91495643734932, + 67.01536854743958, + 67.10630764722823, + 67.22047565555573, + 67.31252860784531, + 67.39065796613693, + 68.88474659919738, + 69.45113959550858, + 69.69205457687377, + 69.84725253582, + 69.97218856811523, + 70.05851617097855, + 70.12988825798034, + 70.19403784513473, + 70.24813051223755, + 70.29781345844269, + 70.34338568687438, + 70.38068360805511, + 70.41919645309449, + 70.44898892641068, + 70.4820789384842, + 70.50974248409271, + 70.53648761749268, + 70.5667546749115, + 70.59351196289063, + 70.61864386320114, + 70.64630803108216, + 70.6727023935318, + 70.69872972011567, + 70.72493976354599, + 70.7476874256134, + 70.77046714782715, + 70.79256957054139, + 70.81106184720993, + 70.83241534233093, + 70.8515542602539, + 70.86944600582123, + 70.89099317789078, + 70.90784223556518, + 70.926378428936, + 70.94538230895996, + 70.96381083726882, + 70.98396788597107, + 71.00406824111938, + 71.02347581386566, + 71.04286310195923, + 71.06384484291077, + 71.08473001003266, + 71.10399283409119, + 71.12429804801941, + 71.14437225341797, + 71.1641768026352, + 71.1816377735138, + 71.20087228775024, + 71.21854376792908, + 71.23626034259796, + 71.25274047851562, + 71.27102743148804, + 71.28886438369751, + 71.30608870983124, + 71.32378898143769, + 71.33962363481521, + 71.35465386390686, + 71.371551156044, + 71.3889540195465, + 71.40592097759247, + 71.42115729331971, + 71.43740501642228, + 71.45302256584168, + 71.47052863836288, + 71.4882806968689, + 71.50748840570449, + 71.52952627658844, + 71.55059865713119, + 71.57056875228882, + 71.58898196697236, + 71.60883347988128, + 71.6305475282669, + 71.65208287239075, + 71.67324340343475, + 71.69656970024108, + 71.71778207540513, + 71.7423820400238, + 71.76660402297973, + 71.79135773181915, + 71.8179384636879, + 71.84612865447998, + 71.87385843992233, + 71.90183074951172, + 71.92967116832733, + 71.96011421203613, + 71.99151601076126, + 72.02691771984101, + 72.06483766317368, + 72.10874972343444, + 72.15644047737122, + 72.21186874389649, + 72.2747768831253, + 72.34141691207886, + 72.41462646722793, + 72.5150628232956, + 72.65761122465133, + 72.89338689804077, + 73.48060830593109, + 73.56621590280531, + 73.65627765464784, + 73.81442364430428, + 73.93480096769333, + 74.05676769256591, + 74.22994810676575, + 74.53286233878123, + 75.0761570343972, + 77.40913844251634, + 80.2655718396903, + 96.00236716117574, + 111.80089807510376 + ], + "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": 29993, + "observation_to_action_latency_ms": { + "distribution_type": "inverse_cdf", + "latency_ms": [ + 65.39556503295898, + 65.90317262949944, + 66.34794480299949, + 66.52523650598526, + 66.76706639719009, + 67.02408304548264, + 67.17864943504334, + 67.26108434915542, + 67.40594648313522, + 67.5005348443985, + 67.6011360502243, + 67.68505244469642, + 67.77472558736801, + 69.4243199634552, + 70.02058695793151, + 70.28187300682067, + 70.46149944067001, + 70.59055726051331, + 70.67848117351532, + 70.76189371585846, + 70.82301428794861, + 70.87942266464233, + 70.93270378589631, + 70.97858119487762, + 71.02201527357101, + 71.05530756950378, + 71.0888930439949, + 71.12097437858581, + 71.14887702703476, + 71.17826912879944, + 71.2058483171463, + 71.23736724853515, + 71.26493341445922, + 71.2944363117218, + 71.31952383756638, + 71.34473480701446, + 71.3709145784378, + 71.39759300231934, + 71.42056230068206, + 71.44201333999634, + 71.46333529710769, + 71.4817370414734, + 71.50154428243637, + 71.52057462692261, + 71.5372909116745, + 71.55419871330261, + 71.5740326166153, + 71.59386352539063, + 71.61349762916565, + 71.6311550617218, + 71.64978809833526, + 71.6697144985199, + 71.68892177343369, + 71.70919153213501, + 71.72815980434417, + 71.74859975814819, + 71.7698770403862, + 71.79157061576844, + 71.81258868217468, + 71.83437560558319, + 71.8536979842186, + 71.87217211723328, + 71.88915539979935, + 71.90820856571197, + 71.92582542896271, + 71.94395114898681, + 71.96166801452637, + 71.97716438293457, + 71.99436103343963, + 72.01059600830078, + 72.02775737524033, + 72.04330060482025, + 72.0600642299652, + 72.07685089111328, + 72.09341137647628, + 72.11067230224609, + 72.12661106586457, + 72.14578086853027, + 72.16744079828263, + 72.18705845355987, + 72.20654620409012, + 72.22568397521972, + 72.2445086312294, + 72.26336144447326, + 72.28845200538635, + 72.3110949420929, + 72.33329623937607, + 72.35886358737946, + 72.38054457426071, + 72.40417922973633, + 72.42825816869735, + 72.45425968170166, + 72.48099433660508, + 72.50895282745361, + 72.53547874927521, + 72.56302463531495, + 72.59042900800705, + 72.62289143562317, + 72.65363789081573, + 72.68952176094055, + 72.72733676195145, + 72.76983900070191, + 72.81949627876281, + 72.87578072071075, + 72.93594701766968, + 73.00869238853454, + 73.09237577915192, + 73.1850743484497, + 73.34378555059433, + 73.58954241752625, + 74.25404654741287, + 74.377984126091, + 74.45795891284942, + 74.56692840909957, + 74.68882406949997, + 74.83917311906815, + 74.99441411018371, + 75.2778499314785, + 75.79855277395247, + 77.6527411787511, + 79.8600849787, + 83.74479532060701, + 87.96991801261902 + ], + "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": 0.9963474887247316, + "worker_service_time_ms": { + "distribution_type": "inverse_cdf", + "latency_ms": [ + 65.03665804862976, + 65.55398055746555, + 65.9878485212326, + 66.16043788576125, + 66.39510260963439, + 66.65668116617202, + 66.81379850769044, + 66.91483943581581, + 67.0153457736969, + 67.10624616384506, + 67.22031239414216, + 67.31242878675461, + 67.39064136266708, + 68.884574842453, + 69.4511189675331, + 69.69204290390014, + 69.84723336696625, + 69.97216024398804, + 70.05845548868179, + 70.1298413324356, + 70.19400545597077, + 70.24808297157287, + 70.29777021169663, + 70.34335553169251, + 70.3806136393547, + 70.41918155670166, + 70.448968064785, + 70.48199189662934, + 70.50970908641816, + 70.53647972106934, + 70.56663146018983, + 70.59342155456542, + 70.61852590322495, + 70.64628098487854, + 70.67265600442886, + 70.69867686271668, + 70.72488385438919, + 70.7476558971405, + 70.7703990483284, + 70.79254083633423, + 70.81092375040055, + 70.83221793174744, + 70.85147241830826, + 70.8694316148758, + 70.89091994285583, + 70.9078179550171, + 70.92635341882706, + 70.94532262325286, + 70.9637482881546, + 70.98387693405151, + 71.00404205799103, + 71.02344303131103, + 71.04280935287476, + 71.06381531715392, + 71.08464890480042, + 71.10389989852905, + 71.12422089576721, + 71.14431077003479, + 71.16399006843567, + 71.18156445026398, + 71.2007286643982, + 71.21847820281982, + 71.23595291376114, + 71.25258999824524, + 71.27089961290359, + 71.28884199142456, + 71.3059007525444, + 71.32369264602661, + 71.33935357570648, + 71.35447284698486, + 71.37133013010025, + 71.38878374099731, + 71.40563994407654, + 71.42104347229004, + 71.43730006217956, + 71.4528238105774, + 71.47043471336364, + 71.48799346923828, + 71.50737279176712, + 71.52930597782135, + 71.55040246009827, + 71.57041568756104, + 71.58885224580764, + 71.60867823123932, + 71.63035848617554, + 71.65184557914733, + 71.67304146289825, + 71.69649241924286, + 71.71739944934845, + 71.74205615997315, + 71.76628437042237, + 71.79092388153076, + 71.81759873390197, + 71.84595065116882, + 71.87376455783844, + 71.9015589761734, + 71.92951682806014, + 71.95996829032897, + 71.99123099327088, + 72.02651390075684, + 72.06419330596924, + 72.10861415863037, + 72.15558389186859, + 72.21108134746551, + 72.27408450841904, + 72.34028102874755, + 72.4134205698967, + 72.51264576911926, + 72.65399513483048, + 72.89117924690247, + 73.4721660399437, + 73.55530694293975, + 73.64461615943908, + 73.80222330904007, + 73.91535479927065, + 74.04386379599572, + 74.20484306907652, + 74.46073522233964, + 75.0037772812843, + 76.84873060560238, + 79.33595724523077, + 82.89885435822055, + 87.28808784484863 + ], + "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/half-cheetah/vla/starvla-qwenoft-h1/half_cheetah_rgb_state_profile_20260912T023319Z_openvla_native_sft_5k/profile/latency_profile.png b/latency-aware/half-cheetah/vla/starvla-qwenoft-h1/half_cheetah_rgb_state_profile_20260912T023319Z_openvla_native_sft_5k/profile/latency_profile.png new file mode 100644 index 0000000000000000000000000000000000000000..935d1345a0421a6f634df4db3f9010ec284e1bd4 Binary files /dev/null and b/latency-aware/half-cheetah/vla/starvla-qwenoft-h1/half_cheetah_rgb_state_profile_20260912T023319Z_openvla_native_sft_5k/profile/latency_profile.png differ diff --git a/latency-aware/half-cheetah/vla/starvla-qwenoft-h1/half_cheetah_rgb_state_profile_20260912T023319Z_openvla_native_sft_5k/profile/profile.json b/latency-aware/half-cheetah/vla/starvla-qwenoft-h1/half_cheetah_rgb_state_profile_20260912T023319Z_openvla_native_sft_5k/profile/profile.json new file mode 100644 index 0000000000000000000000000000000000000000..7dc2634bd101f638e0549818d7b64c92bb028e57 --- /dev/null +++ b/latency-aware/half-cheetah/vla/starvla-qwenoft-h1/half_cheetah_rgb_state_profile_20260912T023319Z_openvla_native_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": 29997, + "n_capacity_drops": 3, + "n_observation_attempts": 30000, + "per_slot_summary": { + "0": { + "admitted_count": 29997, + "mean_observation_to_action_latency_ms": 71.8201196935092, + "mean_worker_service_time_ms": 71.17275555820454, + "p95_observation_to_action_latency_ms": 73.09400281906129, + "p95_worker_service_time_ms": 72.4146143436432, + "p99_worker_service_time_ms": 73.4799683380127 + } + }, + "provenance": { + "base_config": "/workspace/tasks/20260911T023128Z-p-only4/half_cheetah/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/half_cheetah_rgb_state_l0_return_gt3000_100ep_openvla_native_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": "20260911T035806295710Z", + "summary": { + "frame_ms": 100.0, + "max_ms": 112.48595118522644, + "mean_effective_frames": 0.718201196935092, + "mean_ms": 71.8201196935092, + "min_ms": 65.39556503295898, + "n_samples": 29997, + "p50_frames": 0.71872318983078, + "p50_ms": 71.872318983078, + "p90_frames": 0.7277030391693116, + "p90_ms": 72.77030391693116, + "p95_frames": 0.7309400281906128, + "p95_ms": 73.09400281906127, + "p99_frames": 0.7426679273605348, + "p99_ms": 74.26679273605347, + "prob_latency_gt_1_frame": 0.00010001000100010001, + "prob_latency_gt_2_frames": 0.0, + "prob_latency_gt_3_frames": 0.0, + "std_ms": 1.0660114911848162 + }, + "visualization_path": "latency_profile.png", + "workload_id": "half_cheetah" +} diff --git a/latency-aware/half-cheetah/vla/starvla-qwenoft-h1/half_cheetah_rgb_state_profile_20260912T023319Z_openvla_native_sft_5k/provenance.json b/latency-aware/half-cheetah/vla/starvla-qwenoft-h1/half_cheetah_rgb_state_profile_20260912T023319Z_openvla_native_sft_5k/provenance.json new file mode 100644 index 0000000000000000000000000000000000000000..4f1a03173e59a69adecf48d30c617d51d733dbe7 --- /dev/null +++ b/latency-aware/half-cheetah/vla/starvla-qwenoft-h1/half_cheetah_rgb_state_profile_20260912T023319Z_openvla_native_sft_5k/provenance.json @@ -0,0 +1,34 @@ +{ + "task": "half-cheetah", + "model": "qwenoft", + "training_condition": "latency-aware", + "training_run_id": "half_cheetah_rgb_state_profile_20260912T023319Z_openvla_native_sft_5k", + "source": { + "repo_id": "latency-sensitive-bench/Standard-Pipeline", + "revision": "2208875f92b21e3f8ec2363a30a6705dfe56cc04", + "path": "half_cheetah/profile_latency/OpenVLA", + "url": "https://huggingface.co/latency-sensitive-bench/Standard-Pipeline/tree/2208875f92b21e3f8ec2363a30a6705dfe56cc04/half_cheetah/profile_latency/OpenVLA" + }, + "hf_repo_id": "latency-sensitive-bench/benchmark-models", + "hf_path": "latency-aware/half-cheetah/vla/starvla-qwenoft-h1/half_cheetah_rgb_state_profile_20260912T023319Z_openvla_native_sft_5k", + "checkpoint": { + "source_file": "half_cheetah/profile_latency/OpenVLA/checkpoints/steps_5000_pytorch_model.pt", + "source_sha256": "e3c3c8b9fe59d7e20f19c70c584cd1181f65d1caba8079452e7fc3f82c226f52", + "source_bytes": 9785049835, + "file": "checkpoints/model.pt", + "selection_rule": "only published VLA checkpoint", + "method": "copy", + "sha256": "e3c3c8b9fe59d7e20f19c70c584cd1181f65d1caba8079452e7fc3f82c226f52", + "bytes": 9785049835 + }, + "config_source": "half_cheetah/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/half-cheetah/vla/starvla-qwenoft-h1/half_cheetah_rgb_state_profile_20260912T023319Z_openvla_native_sft_5k/source/config.yaml b/latency-aware/half-cheetah/vla/starvla-qwenoft-h1/half_cheetah_rgb_state_profile_20260912T023319Z_openvla_native_sft_5k/source/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..5fc43e3a4d6a0a4cd6f20132a31f542dd612ed46 --- /dev/null +++ b/latency-aware/half-cheetah/vla/starvla-qwenoft-h1/half_cheetah_rgb_state_profile_20260912T023319Z_openvla_native_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: half_cheetah_rgb_state_profile_20260912T023319Z + dataset_py: lerobot_datasets + eval_data_mix: half_cheetah_rgb_state_profile_20260912T023319Z__val + latency_curriculum: + enabled: false + per_device_batch_size: 16 +framework: + action_model: + action_dim: 6 + action_env_dim: 6 + action_hidden_dim: 2560 + action_horizon: 1 + action_model_type: MLP + loss_type: l1 + state_encoding: discretized_text + 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/half_cheetah/20260912T023319Z-halfcheetah/vla/half_cheetah_rgb_state_profile_20260912T023319Z_openvla_native_sft_5k +rl_games: + env_eval: + enabled: false + task: gymnasium +run_id: half_cheetah_rgb_state_profile_20260912T023319Z_openvla_native_sft_5k +run_root_dir: /mnt/results/latency-sensitive-bench/profile-latency-new/half_cheetah/20260912T023319Z-halfcheetah/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/half-cheetah/vla/starvla-qwenoft-h1/half_cheetah_rgb_state_profile_20260912T023319Z_openvla_native_sft_5k/source/provenance.json b/latency-aware/half-cheetah/vla/starvla-qwenoft-h1/half_cheetah_rgb_state_profile_20260912T023319Z_openvla_native_sft_5k/source/provenance.json new file mode 100644 index 0000000000000000000000000000000000000000..dbc19456d1d9755be080c1d8748c2f46cd0f77f2 --- /dev/null +++ b/latency-aware/half-cheetah/vla/starvla-qwenoft-h1/half_cheetah_rgb_state_profile_20260912T023319Z_openvla_native_sft_5k/source/provenance.json @@ -0,0 +1,875 @@ +{ + "run_id": "20260912T023319Z-halfcheetah", + "final_step": 5000, + "exit_code": 0, + "checkpoint_name": "steps_5000_pytorch_model.pt", + "checkpoint_size": 9785049835, + "checkpoint_sha256": "e3c3c8b9fe59d7e20f19c70c584cd1181f65d1caba8079452e7fc3f82c226f52", + "state_dict_entries": 730, + "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": 13675.306737235, + "_step": 5000, + "_timestamp": 1789200543.729739, + "_wandb.runtime": 13675, + "batch/effective_tokens": 3935, + "batch/image_count": 16, + "batch/input_len_max": 248, + "batch/input_len_mean": 245.9375, + "batch/padding_ratio": 0.008316532258064502, + "batch/pixel_values_rows": 4096, + "batch/size": 16, + "epoch": 0.89, + "eval/action_loss/samples": 3200, + "eval/action_loss/seconds": 35.30825001600897, + "eval/half_cheetah_rgb_state/latency_1/loss": 0.05284370481967926, + "eval/half_cheetah_rgb_state/loss": 0.05284370481967926, + "eval/latency_1/loss": 0.05284370481967926, + "eval/loss": 0.05284370481967926, + "learning_rate/action_model": 5e-06, + "learning_rate/qwen_vl_interface": 5e-07, + "throughput/effective_tokens_per_sec": 1493.134857907947, + "throughput/samples_per_sec": 6.071196372687967, + "timing/action_head_loss": 0.0014461719838436693, + "timing/backward": 0.8761777500039898, + "timing/checkpoint_total": 13.609502172999782, + "timing/data": 0.000588834984228015, + "timing/dataloader_next": 0.0006844820163678378, + "timing/eval_action_loss_total": 35.31088111799909, + "timing/forward": 0.18807735299924389, + "timing/log_metrics_total": 0.005098304012790322, + "timing/lr_scheduler": 9.150899131782352e-05, + "timing/model": 2.635211137996521, + "timing/optimizer_step": 1.567223363002995, + "timing/qwen_h2d": 0.0033301259973086417, + "timing/qwen_input_build_total": 0.02363571999012493, + "timing/qwen_processor": 0.01972937601385638, + "timing/train_step_total": 2.6353949070035014, + "timing/vlm_forward": 0.16187168197939172, + "train/grad_norm_pre_clip": 13.34566593170166, + "train/loss": 0.019473642110824585 + }, + "url": "https://wandb.ai/dongqianyu99-zhejiang-university/latency-sensitive-bench/runs/e25knikj", + "exit_code": 0 + }, + "dataset_upload": [ + { + "repo_id": "latency-sensitive-bench/extra-envs-rollouts", + "repo_type": "dataset", + "prefix": "profile-latency-new/half_cheetah/20260912T023319Z-halfcheetah/raw", + "commit": "8cfca81fc4219500ca1d3e669f0e875edcf767e4", + "url": "https://huggingface.co/datasets/latency-sensitive-bench/extra-envs-rollouts/commit/8cfca81fc4219500ca1d3e669f0e875edcf767e4", + "verified_files": 6, + "bytes": 1029841880, + "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/half_cheetah/20260912T023319Z-halfcheetah/converted", + "commit": "11c353e661a9bf53d817401307b32077a8d6ef09", + "url": "https://huggingface.co/datasets/latency-sensitive-bench/extra-envs-rollouts/commit/11c353e661a9bf53d817401307b32077a8d6ef09", + "verified_files": 117, + "bytes": 1027680444, + "verification": "Every file size; Git blob SHA1 for regular files and SHA256 for LFS files" + } + ], + "data_validation": { + "episodes": [ + { + "split": "train", + "episode_idx": 0, + "seed": 2222, + "actual_rows": 1000, + "actual_return": 5244.513095088303 + }, + { + "split": "train", + "episode_idx": 2, + "seed": 2224, + "actual_rows": 1000, + "actual_return": 5293.220133725554 + }, + { + "split": "train", + "episode_idx": 4, + "seed": 2226, + "actual_rows": 1000, + "actual_return": 5442.571658849716 + }, + { + "split": "train", + "episode_idx": 5, + "seed": 2227, + "actual_rows": 1000, + "actual_return": 5211.861645676196 + }, + { + "split": "train", + "episode_idx": 6, + "seed": 2228, + "actual_rows": 1000, + "actual_return": 5359.001422204077 + }, + { + "split": "train", + "episode_idx": 8, + "seed": 2231, + "actual_rows": 1000, + "actual_return": 5174.373645864427 + }, + { + "split": "train", + "episode_idx": 9, + "seed": 2232, + "actual_rows": 1000, + "actual_return": 5056.3356936611235 + }, + { + "split": "train", + "episode_idx": 10, + "seed": 2233, + "actual_rows": 1000, + "actual_return": 5255.268740549684 + }, + { + "split": "train", + "episode_idx": 11, + "seed": 2234, + "actual_rows": 1000, + "actual_return": 5143.60781172663 + }, + { + "split": "train", + "episode_idx": 12, + "seed": 2235, + "actual_rows": 1000, + "actual_return": 4971.32707776688 + }, + { + "split": "train", + "episode_idx": 13, + "seed": 2236, + "actual_rows": 1000, + "actual_return": 5243.369085617363 + }, + { + "split": "train", + "episode_idx": 14, + "seed": 2237, + "actual_rows": 1000, + "actual_return": 5470.776911958121 + }, + { + "split": "train", + "episode_idx": 15, + "seed": 2238, + "actual_rows": 1000, + "actual_return": 5416.247515305877 + }, + { + "split": "train", + "episode_idx": 16, + "seed": 2239, + "actual_rows": 1000, + "actual_return": 5323.359701700509 + }, + { + "split": "train", + "episode_idx": 18, + "seed": 2241, + "actual_rows": 1000, + "actual_return": 5430.164633862674 + }, + { + "split": "train", + "episode_idx": 19, + "seed": 2242, + "actual_rows": 1000, + "actual_return": 3849.5855917446315 + }, + { + "split": "train", + "episode_idx": 20, + "seed": 2243, + "actual_rows": 1000, + "actual_return": 4564.361369084567 + }, + { + "split": "train", + "episode_idx": 21, + "seed": 2244, + "actual_rows": 1000, + "actual_return": 5198.675570689142 + }, + { + "split": "train", + "episode_idx": 22, + "seed": 2245, + "actual_rows": 1000, + "actual_return": 5343.187917873263 + }, + { + "split": "train", + "episode_idx": 23, + "seed": 2246, + "actual_rows": 1000, + "actual_return": 5273.128447249532 + }, + { + "split": "train", + "episode_idx": 24, + "seed": 2247, + "actual_rows": 1000, + "actual_return": 5393.604007827118 + }, + { + "split": "train", + "episode_idx": 26, + "seed": 2249, + "actual_rows": 1000, + "actual_return": 5103.907880667597 + }, + { + "split": "train", + "episode_idx": 27, + "seed": 2250, + "actual_rows": 1000, + "actual_return": 5109.762326894328 + }, + { + "split": "train", + "episode_idx": 28, + "seed": 2251, + "actual_rows": 1000, + "actual_return": 4419.583458398934 + }, + { + "split": "train", + "episode_idx": 30, + "seed": 2253, + "actual_rows": 1000, + "actual_return": 5318.688621792942 + }, + { + "split": "train", + "episode_idx": 31, + "seed": 2254, + "actual_rows": 1000, + "actual_return": 5071.568065676838 + }, + { + "split": "train", + "episode_idx": 32, + "seed": 2255, + "actual_rows": 1000, + "actual_return": 5113.272311076522 + }, + { + "split": "train", + "episode_idx": 33, + "seed": 2256, + "actual_rows": 1000, + "actual_return": 5205.3343274742365 + }, + { + "split": "train", + "episode_idx": 34, + "seed": 2257, + "actual_rows": 1000, + "actual_return": 5445.97093359381 + }, + { + "split": "train", + "episode_idx": 35, + "seed": 2258, + "actual_rows": 1000, + "actual_return": 5249.648839783855 + }, + { + "split": "train", + "episode_idx": 36, + "seed": 2259, + "actual_rows": 1000, + "actual_return": 5386.2273238524795 + }, + { + "split": "train", + "episode_idx": 37, + "seed": 2260, + "actual_rows": 1000, + "actual_return": 5269.797439053655 + }, + { + "split": "train", + "episode_idx": 38, + "seed": 2262, + "actual_rows": 1000, + "actual_return": 5309.328847865574 + }, + { + "split": "train", + "episode_idx": 39, + "seed": 2263, + "actual_rows": 1000, + "actual_return": 5225.462058153003 + }, + { + "split": "train", + "episode_idx": 40, + "seed": 2264, + "actual_rows": 1000, + "actual_return": 5291.524852637202 + }, + { + "split": "train", + "episode_idx": 41, + "seed": 2265, + "actual_rows": 1000, + "actual_return": 5221.125392388552 + }, + { + "split": "train", + "episode_idx": 42, + "seed": 2267, + "actual_rows": 1000, + "actual_return": 5324.138521671295 + }, + { + "split": "train", + "episode_idx": 43, + "seed": 2268, + "actual_rows": 1000, + "actual_return": 5227.976520098746 + }, + { + "split": "train", + "episode_idx": 44, + "seed": 2269, + "actual_rows": 1000, + "actual_return": 5103.029847417958 + }, + { + "split": "train", + "episode_idx": 45, + "seed": 2270, + "actual_rows": 1000, + "actual_return": 5169.398928850889 + }, + { + "split": "train", + "episode_idx": 46, + "seed": 2271, + "actual_rows": 1000, + "actual_return": 5295.467996349558 + }, + { + "split": "train", + "episode_idx": 48, + "seed": 2273, + "actual_rows": 1000, + "actual_return": 4990.449957065284 + }, + { + "split": "train", + "episode_idx": 49, + "seed": 2274, + "actual_rows": 1000, + "actual_return": 5157.177657991648 + }, + { + "split": "train", + "episode_idx": 50, + "seed": 2275, + "actual_rows": 1000, + "actual_return": 5145.647473279387 + }, + { + "split": "train", + "episode_idx": 51, + "seed": 2276, + "actual_rows": 1000, + "actual_return": 5140.257968872786 + }, + { + "split": "train", + "episode_idx": 52, + "seed": 2277, + "actual_rows": 1000, + "actual_return": 5366.840925060213 + }, + { + "split": "train", + "episode_idx": 53, + "seed": 2278, + "actual_rows": 1000, + "actual_return": 5246.116013586521 + }, + { + "split": "train", + "episode_idx": 54, + "seed": 2279, + "actual_rows": 1000, + "actual_return": 5282.868002966978 + }, + { + "split": "train", + "episode_idx": 55, + "seed": 2280, + "actual_rows": 1000, + "actual_return": 5199.930402893573 + }, + { + "split": "train", + "episode_idx": 56, + "seed": 2281, + "actual_rows": 1000, + "actual_return": 5418.548827514052 + }, + { + "split": "train", + "episode_idx": 57, + "seed": 2282, + "actual_rows": 1000, + "actual_return": 5346.238312704489 + }, + { + "split": "train", + "episode_idx": 59, + "seed": 2284, + "actual_rows": 1000, + "actual_return": 5396.301347769797 + }, + { + "split": "train", + "episode_idx": 60, + "seed": 2285, + "actual_rows": 1000, + "actual_return": 5275.902740761638 + }, + { + "split": "train", + "episode_idx": 61, + "seed": 2286, + "actual_rows": 1000, + "actual_return": 3199.60966812633 + }, + { + "split": "train", + "episode_idx": 62, + "seed": 2287, + "actual_rows": 1000, + "actual_return": 4981.688836708665 + }, + { + "split": "train", + "episode_idx": 63, + "seed": 2288, + "actual_rows": 1000, + "actual_return": 5204.507361498661 + }, + { + "split": "train", + "episode_idx": 64, + "seed": 2289, + "actual_rows": 1000, + "actual_return": 5208.518176048994 + }, + { + "split": "train", + "episode_idx": 65, + "seed": 2291, + "actual_rows": 1000, + "actual_return": 5438.322305753827 + }, + { + "split": "train", + "episode_idx": 66, + "seed": 2292, + "actual_rows": 1000, + "actual_return": 5286.404946647584 + }, + { + "split": "train", + "episode_idx": 67, + "seed": 2293, + "actual_rows": 1000, + "actual_return": 5332.273515948269 + }, + { + "split": "train", + "episode_idx": 68, + "seed": 2294, + "actual_rows": 1000, + "actual_return": 5330.759604662657 + }, + { + "split": "train", + "episode_idx": 69, + "seed": 2295, + "actual_rows": 1000, + "actual_return": 5208.749480232596 + }, + { + "split": "train", + "episode_idx": 70, + "seed": 2296, + "actual_rows": 1000, + "actual_return": 5292.029105883092 + }, + { + "split": "train", + "episode_idx": 71, + "seed": 2297, + "actual_rows": 1000, + "actual_return": 5481.1443650871515 + }, + { + "split": "train", + "episode_idx": 72, + "seed": 2298, + "actual_rows": 1000, + "actual_return": 5205.774039665237 + }, + { + "split": "train", + "episode_idx": 73, + "seed": 2299, + "actual_rows": 1000, + "actual_return": 5093.480286929756 + }, + { + "split": "train", + "episode_idx": 74, + "seed": 2300, + "actual_rows": 1000, + "actual_return": 5233.802292553708 + }, + { + "split": "train", + "episode_idx": 75, + "seed": 2301, + "actual_rows": 1000, + "actual_return": 5118.425067208707 + }, + { + "split": "train", + "episode_idx": 76, + "seed": 2302, + "actual_rows": 1000, + "actual_return": 5172.481769055128 + }, + { + "split": "train", + "episode_idx": 78, + "seed": 2304, + "actual_rows": 1000, + "actual_return": 5381.566341370344 + }, + { + "split": "train", + "episode_idx": 79, + "seed": 2305, + "actual_rows": 1000, + "actual_return": 5563.690887295641 + }, + { + "split": "train", + "episode_idx": 80, + "seed": 2306, + "actual_rows": 1000, + "actual_return": 5343.528509296477 + }, + { + "split": "train", + "episode_idx": 82, + "seed": 2308, + "actual_rows": 1000, + "actual_return": 5117.677170692012 + }, + { + "split": "train", + "episode_idx": 83, + "seed": 2309, + "actual_rows": 1000, + "actual_return": 5131.077620420605 + }, + { + "split": "train", + "episode_idx": 84, + "seed": 2310, + "actual_rows": 1000, + "actual_return": 4979.46707751276 + }, + { + "split": "train", + "episode_idx": 85, + "seed": 2311, + "actual_rows": 1000, + "actual_return": 5329.2891001701355 + }, + { + "split": "train", + "episode_idx": 86, + "seed": 2312, + "actual_rows": 1000, + "actual_return": 5411.920866906643 + }, + { + "split": "train", + "episode_idx": 87, + "seed": 2314, + "actual_rows": 1000, + "actual_return": 5146.85342657275 + }, + { + "split": "train", + "episode_idx": 88, + "seed": 2315, + "actual_rows": 1000, + "actual_return": 5418.382528834045 + }, + { + "split": "train", + "episode_idx": 89, + "seed": 2316, + "actual_rows": 1000, + "actual_return": 5225.123071342707 + }, + { + "split": "train", + "episode_idx": 90, + "seed": 2317, + "actual_rows": 1000, + "actual_return": 5248.501497633755 + }, + { + "split": "train", + "episode_idx": 91, + "seed": 2318, + "actual_rows": 1000, + "actual_return": 5313.211864814162 + }, + { + "split": "train", + "episode_idx": 92, + "seed": 2319, + "actual_rows": 1000, + "actual_return": 5229.680067633861 + }, + { + "split": "train", + "episode_idx": 93, + "seed": 2320, + "actual_rows": 1000, + "actual_return": 5159.047764834017 + }, + { + "split": "train", + "episode_idx": 94, + "seed": 2321, + "actual_rows": 1000, + "actual_return": 5282.597278513014 + }, + { + "split": "train", + "episode_idx": 95, + "seed": 2322, + "actual_rows": 1000, + "actual_return": 5397.530524291098 + }, + { + "split": "train", + "episode_idx": 96, + "seed": 2323, + "actual_rows": 1000, + "actual_return": 5186.173977211118 + }, + { + "split": "train", + "episode_idx": 97, + "seed": 2325, + "actual_rows": 1000, + "actual_return": 5244.024773940444 + }, + { + "split": "train", + "episode_idx": 98, + "seed": 2326, + "actual_rows": 1000, + "actual_return": 5073.73131226562 + }, + { + "split": "train", + "episode_idx": 99, + "seed": 2327, + "actual_rows": 1000, + "actual_return": 5119.739166211337 + }, + { + "split": "val", + "episode_idx": 1, + "seed": 2223, + "actual_rows": 1000, + "actual_return": 5091.6459036022425 + }, + { + "split": "val", + "episode_idx": 3, + "seed": 2225, + "actual_rows": 1000, + "actual_return": 5215.420770222321 + }, + { + "split": "val", + "episode_idx": 7, + "seed": 2230, + "actual_rows": 1000, + "actual_return": 5137.353235999122 + }, + { + "split": "val", + "episode_idx": 17, + "seed": 2240, + "actual_rows": 1000, + "actual_return": 5216.283385582268 + }, + { + "split": "val", + "episode_idx": 25, + "seed": 2248, + "actual_rows": 1000, + "actual_return": 5315.44768577069 + }, + { + "split": "val", + "episode_idx": 29, + "seed": 2252, + "actual_rows": 1000, + "actual_return": 5309.680116742849 + }, + { + "split": "val", + "episode_idx": 47, + "seed": 2272, + "actual_rows": 1000, + "actual_return": 5360.585786227137 + }, + { + "split": "val", + "episode_idx": 58, + "seed": 2283, + "actual_rows": 1000, + "actual_return": 3338.4295101789758 + }, + { + "split": "val", + "episode_idx": 77, + "seed": 2303, + "actual_rows": 1000, + "actual_return": 5284.776956349611 + }, + { + "split": "val", + "episode_idx": 81, + "seed": 2307, + "actual_rows": 1000, + "actual_return": 5223.671133276075 + } + ], + "total_episodes": 100, + "total_rows": 100000, + "pass_gate_count": 100, + "mean_return": 5180.70115935914, + "minimum_return": 3199.60966812633, + "probe_train_rows": 90000, + "probe_val_rows": 10000, + "actual_train_rows": 90000, + "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_threshold": 3000, + "effective_threshold": 3000, + "changed": false, + "reason": "Original IID exporter probe accepts11/12 and independentE10/10 at gate3000; retain source gate.", + "probe": { + "status": "diagnostic_budget_reached", + "attempts": 12, + "completed_reset_seeds": [ + 2222, + 2223, + 2224, + 2225, + 2226, + 2227, + 2228, + 2229, + 2230, + 2231, + 2232, + 2233 + ], + "accepted_original_gate": 11, + "min_return": 2773.7933985423297, + "max_return": 5442.571658849716, + "mean_return": 5021.279945332091, + "gate": "episode_raw_return > 3000", + "original_dataset_target": 100, + "checkpoint": "/mnt/results/latency-sensitive-bench/profile-latency-new/half_cheetah/20260912T023319Z-halfcheetah/small_models/half_cheetah_profile_20260912T023319Z/checkpoint_p0/best_000019544_10006528_reward_4867.430.pth", + "deterministic": true, + "env_fps": 10, + "obs_fps": 10, + "latency_semantics": "action-delay IID exporter, no worker occupancy replay", + "note": "Intentional 12-attempt diagnostic cap; no dataset materialized; not exporter failure" + }, + "deterministic_evidence": { + "source_metadata_flag": "omitted", + "source_teacher_eval_deterministic": true, + "actual_exporter_default": true, + "exporter_sha256": "a311714c474e9dbac97e4ddb15d4b126ee89e8bb7b57e5b8e202ff666b4e4254", + "line": 66, + "new_command": "explicit --deterministic" + }, + "independent_E": { + "status": "parent_verified_exit0", + "completed_utc": "2026-09-12T03:06:10Z", + "episodes": 10, + "raw_ticks_each": 1000, + "actual_seeds": [ + 106951, + 106952, + 106953, + 106954, + 106955, + 106956, + 106957, + 106958, + 106959, + 106960 + ], + "mean_return": 5217.471104, + "std_return": 108.445224, + "pass_gt3000": 10 + } + }, + "normalization_source": "new HalfCheetah converted dataset and actual VLA training dataset_statistics.json" +} diff --git a/latency-aware/half-cheetah/vla/starvla-qwenoft-h1/half_cheetah_rgb_state_profile_20260912T023319Z_openvla_native_sft_5k/task_contract.json b/latency-aware/half-cheetah/vla/starvla-qwenoft-h1/half_cheetah_rgb_state_profile_20260912T023319Z_openvla_native_sft_5k/task_contract.json new file mode 100644 index 0000000000000000000000000000000000000000..82bf97a66c02605bdf812e25013052b5ae6b5728 --- /dev/null +++ b/latency-aware/half-cheetah/vla/starvla-qwenoft-h1/half_cheetah_rgb_state_profile_20260912T023319Z_openvla_native_sft_5k/task_contract.json @@ -0,0 +1,73 @@ +{ + "action_space": { + "dtype": "float32", + "high": [ + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0 + ], + "labels": [ + "back_thigh_torque", + "back_shin_torque", + "back_foot_torque", + "front_thigh_torque", + "front_shin_torque", + "front_foot_torque" + ], + "low": [ + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0 + ], + "type": "box" + }, + "base_prompt": "Make the Half Cheetah run forward as fast as possible. Predict six continuous torques in [-1, 1] ordered as back thigh, back shin, back foot, front thigh, front shin, and front foot.", + "env_fps": 10.0, + "env_id": "LatencyBench/HalfCheetahContinuous-v0", + "frame_stack": 1, + "make_kwargs": { + "base_env_id": "HalfCheetah-v4", + "base_make_kwargs": {}, + "render_mode": "rgb_array" + }, + "noop_action": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ], + "obs_fps": 10.0, + "registration_imports": [ + "latency_bench.envs.gymnasium_half_cheetah" + ], + "state_space": { + "labels": [ + "front_tip_height", + "front_tip_angle", + "back_thigh_angle", + "back_shin_angle", + "back_foot_angle", + "front_thigh_angle", + "front_shin_angle", + "front_foot_angle", + "front_tip_x_velocity", + "front_tip_z_velocity", + "front_tip_angular_velocity", + "back_thigh_angular_velocity", + "back_shin_angular_velocity", + "back_foot_angular_velocity", + "front_thigh_angular_velocity", + "front_shin_angular_velocity", + "front_foot_angular_velocity" + ] + }, + "task_name": "half_cheetah_rgb_state" +} diff --git a/zero-latency/half-cheetah/small-policy/sample-factory-v1/README.md b/zero-latency/half-cheetah/small-policy/sample-factory-v1/README.md new file mode 100644 index 0000000000000000000000000000000000000000..305065a3de6e7a19eef10275b9ec2c92f72874a2 --- /dev/null +++ b/zero-latency/half-cheetah/small-policy/sample-factory-v1/README.md @@ -0,0 +1,26 @@ +# half-cheetah / sample-factory-appo + +Training condition: `zero-latency`. Run: `gymnasium_half_cheetah_continuous_zero_latency_10m`. + +Copied from [Standard-Pipeline](https://huggingface.co/latency-sensitive-bench/Standard-Pipeline/tree/2208875f92b21e3f8ec2363a30a6705dfe56cc04/half_cheetah/zero_latency/small_model) at `2208875f92b21e3f8ec2363a30a6705dfe56cc04`. + +- Checkpoint: `checkpoint.pth` +- Selection: existing zero-latency training best +- Checkpoint SHA256: `86b4640d192f7d1e78cda34a10945cb22bde03347f5025484f3eba8bc519b173` +- 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/half-cheetah/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/half-cheetah/small-policy/sample-factory-v1/checkpoint.pth b/zero-latency/half-cheetah/small-policy/sample-factory-v1/checkpoint.pth new file mode 100644 index 0000000000000000000000000000000000000000..b079f383b4e0a19d2361446329255632810f622a --- /dev/null +++ b/zero-latency/half-cheetah/small-policy/sample-factory-v1/checkpoint.pth @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:86b4640d192f7d1e78cda34a10945cb22bde03347f5025484f3eba8bc519b173 +size 29877 diff --git a/zero-latency/half-cheetah/small-policy/sample-factory-v1/config.json b/zero-latency/half-cheetah/small-policy/sample-factory-v1/config.json new file mode 100644 index 0000000000000000000000000000000000000000..02dfd1829a6c71d2ac9dccdd190432a6c6d85859 --- /dev/null +++ b/zero-latency/half-cheetah/small-policy/sample-factory-v1/config.json @@ -0,0 +1,244 @@ +{ + "help": false, + "algo": "APPO", + "env": "latency_gymnasium", + "experiment": "gymnasium_half_cheetah_continuous_zero_latency_10m", + "train_dir": "/mnt/checkpoints/latency-sensitive-bench/small_models/half_cheetah", + "restart_behavior": "overwrite", + "device": "gpu", + "seed": 2222, + "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, + "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": 1, + "pixel_format": "CHW", + "use_record_episode_statistics": false, + "with_wandb": false, + "wandb_user": null, + "wandb_project": "sample_factory", + "wandb_group": null, + "wandb_job_type": "SF", + "wandb_tags": [], + "with_pbt": false, + "pbt_mix_policies_in_one_env": true, + "pbt_period_env_steps": 5000000, + "pbt_start_mutation": 20000000, + "pbt_replace_fraction": 0.3, + "pbt_mutation_rate": 0.15, + "pbt_replace_reward_gap": 0.1, + "pbt_replace_reward_gap_absolute": 1e-06, + "pbt_optimize_gamma": false, + "pbt_target_objective": "true_objective", + "pbt_perturb_min": 1.1, + "pbt_perturb_max": 1.5, + "gym_task_name": "half_cheetah_continuous", + "gym_env_id": "LatencyBench/HalfCheetahContinuous-v0", + "gym_make_kwargs_json": "{\"base_env_id\": \"HalfCheetah-v4\", \"render_mode\": \"rgb_array\", \"base_make_kwargs\": {}}", + "gym_registration_imports_json": "[\"latency_bench.envs.gymnasium_half_cheetah\"]", + "gym_action_space_json": "{\"type\": \"box\", \"labels\": [\"back_thigh_torque\", \"back_shin_torque\", \"back_foot_torque\", \"front_thigh_torque\", \"front_shin_torque\", \"front_foot_torque\"], \"low\": [-1.0, -1.0, -1.0, -1.0, -1.0, -1.0], \"high\": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0], \"dtype\": \"float32\"}", + "gym_action_labels_json": null, + "gym_action_values_json": null, + "gym_noop_action_id": null, + "gym_noop_action_json": "[0.0, 0.0, 0.0, 0.0, 0.0, 0.0]", + "gym_base_prompt": "Make the Half Cheetah run forward as fast as possible. Predict six continuous torques in [-1, 1] ordered as back thigh, back shin, back foot, front thigh, front shin, and front foot.", + "gym_state_labels_json": "", + "env_fps": 20.0, + "obs_fps": 20.0, + "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/half_cheetah/gymnasium_half_cheetah_continuous_zero_latency_10m/episode_metrics.jsonl", + "command_line": "--mode train --algo APPO --env latency_gymnasium --experiment gymnasium_half_cheetah_continuous_zero_latency_10m --train_dir /mnt/checkpoints/latency-sensitive-bench/small_models/half_cheetah --restart_behavior overwrite --device gpu --seed 2222 --episode_metrics_path /mnt/checkpoints/latency-sensitive-bench/small_models/half_cheetah/gymnasium_half_cheetah_continuous_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 --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 half_cheetah_continuous --gym-env-id LatencyBench/HalfCheetahContinuous-v0 --gym-make-kwargs-json {\"base_env_id\": \"HalfCheetah-v4\", \"render_mode\": \"rgb_array\", \"base_make_kwargs\": {}} --gym-registration-imports-json [\"latency_bench.envs.gymnasium_half_cheetah\"] --gym-action-space-json {\"type\": \"box\", \"labels\": [\"back_thigh_torque\", \"back_shin_torque\", \"back_foot_torque\", \"front_thigh_torque\", \"front_shin_torque\", \"front_foot_torque\"], \"low\": [-1.0, -1.0, -1.0, -1.0, -1.0, -1.0], \"high\": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0], \"dtype\": \"float32\"} --gym-noop-action-json [0.0, 0.0, 0.0, 0.0, 0.0, 0.0] --gym-base-prompt Make the Half Cheetah run forward as fast as possible. Predict six continuous torques in [-1, 1] ordered as back thigh, back shin, back foot, front thigh, front shin, and front foot. --env-fps 20.0 --obs-fps 20.0 --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_half_cheetah_continuous_zero_latency_10m", + "train_dir": "/mnt/checkpoints/latency-sensitive-bench/small_models/half_cheetah", + "restart_behavior": "overwrite", + "device": "gpu", + "seed": 2222, + "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, + "gym_task_name": "half_cheetah_continuous", + "gym_env_id": "LatencyBench/HalfCheetahContinuous-v0", + "gym_make_kwargs_json": "{\"base_env_id\": \"HalfCheetah-v4\", \"render_mode\": \"rgb_array\", \"base_make_kwargs\": {}}", + "gym_registration_imports_json": "[\"latency_bench.envs.gymnasium_half_cheetah\"]", + "gym_action_space_json": "{\"type\": \"box\", \"labels\": [\"back_thigh_torque\", \"back_shin_torque\", \"back_foot_torque\", \"front_thigh_torque\", \"front_shin_torque\", \"front_foot_torque\"], \"low\": [-1.0, -1.0, -1.0, -1.0, -1.0, -1.0], \"high\": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0], \"dtype\": \"float32\"}", + "gym_noop_action_json": "[0.0, 0.0, 0.0, 0.0, 0.0, 0.0]", + "gym_base_prompt": "Make the Half Cheetah run forward as fast as possible. Predict six continuous torques in [-1, 1] ordered as back thigh, back shin, back foot, front thigh, front shin, and front foot.", + "env_fps": 20.0, + "obs_fps": 20.0, + "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/half_cheetah/gymnasium_half_cheetah_continuous_zero_latency_10m/episode_metrics.jsonl" + }, + "git_hash": "1679dcaedbe465741be1156addfdc17dfe87d61f", + "git_repo_name": "https://github.com/ZihanWang314/latency-sensitive-bench.git", + "eval_env_frameskip": 1, + "output_dir": "/mnt/results/latency-sensitive-bench/tasks/half_cheetah/small_model_train" +} \ No newline at end of file diff --git a/zero-latency/half-cheetah/small-policy/sample-factory-v1/provenance.json b/zero-latency/half-cheetah/small-policy/sample-factory-v1/provenance.json new file mode 100644 index 0000000000000000000000000000000000000000..05eda69b4308907ad6a779e2313527439da07574 --- /dev/null +++ b/zero-latency/half-cheetah/small-policy/sample-factory-v1/provenance.json @@ -0,0 +1,44 @@ +{ + "task": "half-cheetah", + "model": "sample-factory-appo", + "training_condition": "zero-latency", + "training_run_id": "gymnasium_half_cheetah_continuous_zero_latency_10m", + "source": { + "repo_id": "latency-sensitive-bench/Standard-Pipeline", + "revision": "2208875f92b21e3f8ec2363a30a6705dfe56cc04", + "path": "half_cheetah/zero_latency/small_model", + "url": "https://huggingface.co/latency-sensitive-bench/Standard-Pipeline/tree/2208875f92b21e3f8ec2363a30a6705dfe56cc04/half_cheetah/zero_latency/small_model" + }, + "hf_repo_id": "latency-sensitive-bench/benchmark-models", + "hf_path": "zero-latency/half-cheetah/small-policy/sample-factory-v1", + "checkpoint": { + "source_file": "half_cheetah/zero_latency/small_model/checkpoint_p0/best_000019008_9732096_reward_5331.187.pth", + "source_sha256": "58dbf0560f256f88831836867c1b62256ba71d34e47fe813ee454bb8c2b42212", + "source_bytes": 83973, + "file": "checkpoint.pth", + "selection_rule": "existing zero-latency training best", + "method": "inference_export", + "sha256": "86b4640d192f7d1e78cda34a10945cb22bde03347f5025484f3eba8bc519b173", + "bytes": 29877, + "train_step": 19008, + "env_steps": 9732096, + "tensor_count": 15, + "tensor_equality_verified": true, + "removed_fields": [ + "best_performance", + "curr_lr", + "optimizer" + ] + }, + "config_source": "half_cheetah/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/half-cheetah/vla/starvla-qwengr00t-h1/half_cheetah_l0_gr00t_h1_5fps_2h200_20260917/README.md b/zero-latency/half-cheetah/vla/starvla-qwengr00t-h1/half_cheetah_l0_gr00t_h1_5fps_2h200_20260917/README.md new file mode 100644 index 0000000000000000000000000000000000000000..d97f24a0a47d902f6e8a323b01ca08acb62aef9e --- /dev/null +++ b/zero-latency/half-cheetah/vla/starvla-qwengr00t-h1/half_cheetah_l0_gr00t_h1_5fps_2h200_20260917/README.md @@ -0,0 +1,31 @@ +# half-cheetah / qwengr00t + +Training condition: `zero-latency`. Run: `half_cheetah_l0_gr00t_h1_5fps_2h200_20260917`. + +Copied from [Standard-Pipeline](https://huggingface.co/latency-sensitive-bench/Standard-Pipeline/tree/2208875f92b21e3f8ec2363a30a6705dfe56cc04/half_cheetah/zero_latency/GR00T) at `2208875f92b21e3f8ec2363a30a6705dfe56cc04`. + +- Checkpoint: `checkpoints/model.pt` +- Selection: only published VLA checkpoint +- Checkpoint SHA256: `8f007f801c45ed8476546ebf7cc96e60283d2ffd05a6d30feaa529d7ea7f658f` +- 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/half-cheetah/vla/starvla-qwengr00t-h1/half_cheetah_l0_gr00t_h1_5fps_2h200_20260917" +snapshot_download("latency-sensitive-bench/benchmark-models", revision="", + allow_patterns=[prefix + "/**"]) +``` + +Use the verified import commit listed in the repository's import report. diff --git a/zero-latency/half-cheetah/vla/starvla-qwengr00t-h1/half_cheetah_l0_gr00t_h1_5fps_2h200_20260917/checkpoints/model.pt b/zero-latency/half-cheetah/vla/starvla-qwengr00t-h1/half_cheetah_l0_gr00t_h1_5fps_2h200_20260917/checkpoints/model.pt new file mode 100644 index 0000000000000000000000000000000000000000..798fa60de8bd1fd3cdd51a32f280c7b95e5bec5e --- /dev/null +++ b/zero-latency/half-cheetah/vla/starvla-qwengr00t-h1/half_cheetah_l0_gr00t_h1_5fps_2h200_20260917/checkpoints/model.pt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8f007f801c45ed8476546ebf7cc96e60283d2ffd05a6d30feaa529d7ea7f658f +size 9976860963 diff --git a/zero-latency/half-cheetah/vla/starvla-qwengr00t-h1/half_cheetah_l0_gr00t_h1_5fps_2h200_20260917/config.full.yaml b/zero-latency/half-cheetah/vla/starvla-qwengr00t-h1/half_cheetah_l0_gr00t_h1_5fps_2h200_20260917/config.full.yaml new file mode 100644 index 0000000000000000000000000000000000000000..d9f243273a4490f9a508cdcd9632ea3413305039 --- /dev/null +++ b/zero-latency/half-cheetah/vla/starvla-qwengr00t-h1/half_cheetah_l0_gr00t_h1_5fps_2h200_20260917/config.full.yaml @@ -0,0 +1,255 @@ +framework: + name: QwenGR00T + qwenvl: + base_vlm: ${oc.env:GR00T_BACKBONE_DIR} + attn_implementation: flash_attention_2 + vl_hidden_dim: 2048 + enable_gradient_checkpointing: true + action_model: + action_model_type: DiT-B + action_hidden_dim: 1024 + hidden_size: 1024 + add_pos_embed: true + max_seq_len: 1024 + action_dim: 6 + state_dim: 17 + task_objective: null + action_horizon: 1 + repeated_diffusion_steps: 8 + noise_beta_alpha: 1.5 + noise_beta_beta: 1.0 + noise_s: 0.999 + num_timestep_buckets: 1000 + num_inference_timesteps: 4 + num_target_vision_tokens: 32 + diffusion_model_cfg: + cross_attention_dim: 2560 + dropout: 0.2 + final_dropout: true + interleave_self_attention: true + norm_type: ada_norm + num_layers: 16 + output_dim: 1024 + positional_embeddings: null + action_env_dim: 6 + future_action_window_size: 0 + past_action_window_size: 0 +datasets: + vla_data: + dataset_py: lerobot_datasets + include_state: true + data_root_dir: / + data_mix: train + eval_data_mix: validation + custom_mixtures_path: /mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/half_cheetah/zero_latency/vla/mixture.json + action_type: continuous + sequential_step_sampling: true + eval_sequential_step_sampling: null + num_workers: 4 + eval_num_workers: 0 + prefetch_factor: 2 + persistent_workers: true + pin_memory: true + shuffle: true + action_balance: + enabled: false + strategy: balanced_epoch + action_key: action_id + target_flap_fraction: 0.3 + noop_id: 0 + flap_id: 1 + latency_curriculum: + enabled: false + strategy: exclusive + latencies: null + phase_steps: null + phase_distributions: null + new_latency_passes: 1.0 + replay_passes: 0.25 + target_total_passes: 2.0 + final_equalization: true + step_budget_mode: auto + eval_at_phase_end: false + save_at_phase_end: false + computed_plan: null + per_device_batch_size: 32 + load_all_data_for_training: true + num_obs_frames: 1 + image_mode: single + prompt_mode: raw + stitch_grid: + - 2 + - 2 + obs_image_size: + - 224 + - 224 + video_backend: pyav + lerobot_version: v2.0 + gymnasium_task_contract: + action_space: + dtype: float32 + high: + - 1.0 + - 1.0 + - 1.0 + - 1.0 + - 1.0 + - 1.0 + labels: + - back_thigh_torque + - back_shin_torque + - back_foot_torque + - front_thigh_torque + - front_shin_torque + - front_foot_torque + low: + - -1.0 + - -1.0 + - -1.0 + - -1.0 + - -1.0 + - -1.0 + type: box + base_prompt: Make the Half Cheetah run forward as fast as possible. Predict + six continuous torques in [-1, 1] ordered as back thigh, back shin, back foot, + front thigh, front shin, and front foot. + env_fps: 5 + env_id: LatencyBench/HalfCheetahContinuous-v0 + frame_stack: 1 + make_kwargs: + base_env_id: HalfCheetah-v4 + base_make_kwargs: {} + render_mode: rgb_array + noop_action: + - 0.0 + - 0.0 + - 0.0 + - 0.0 + - 0.0 + - 0.0 + obs_fps: 5 + registration_imports: + - latency_bench.envs.gymnasium_half_cheetah + state_space: + labels: + - front_tip_height + - front_tip_angle + - back_thigh_angle + - back_shin_angle + - back_foot_angle + - front_thigh_angle + - front_shin_angle + - front_foot_angle + - front_tip_x_velocity + - front_tip_z_velocity + - front_tip_angular_velocity + - back_thigh_angular_velocity + - back_shin_angular_velocity + - back_foot_angular_velocity + - front_thigh_angular_velocity + - front_shin_angular_velocity + - front_foot_angular_velocity + task_name: half_cheetah_rgb_state + active_action_dim: 6 +dataset: + source_hf: '' + config_name: null + source_subdir: null + converted_name: flappy_train + single_source_hf: '' + mixed_source_hf: '' + single_converted_name: flappy_train + mixed_converted_name: flappy_mixed_latency_train + single_latency_filter: null + mixed_latency_filter: null + force_download: false + setup_force: false + skip_verification: false + target_latency_unit: observation_steps + verify_rows: 200 + max_episodes: null + episodes_per_latency: null + latency_filter: null + debug_subset: + enabled: false + max_episodes: 5 + suffix: debug +base_model: + repo_id: Qwen/Qwen3-VL-4B-Instruct +initialization: + checkpoint_local_dir: null + checkpoint_hf_repo_id: null + checkpoint_filename: null +trainer: + max_train_steps: 5000 + num_warmup_steps: 100 + save_interval: 500 + eval_interval: 500 + eval_num_batches: 200 + per_latency_eval_num_batches: null + eval_action_classification: false + eval_action_classification_interval: null + cc_f1_tolerance: 1 + learning_rate: + base: 2.0e-05 + qwen_vl_interface: 1.0e-05 + action_model: 0.0001 + lr_scheduler_type: cosine_with_min_lr + scheduler_specific_kwargs: + min_lr: 1.0e-06 + freeze_modules: '' + freeze_vit: false + freeze_tied_embedding: false + freeze_llm_layers: [] + loss_scale: + vla: 1.0 + vlm: 0.1 + max_grad_norm: 1.0 + weight_decay: 0.0 + logging_frequency: 1 + profile_timing: + enabled: false + log_interval: 10 + gradient_clipping: 1.0 + gradient_accumulation_steps: 1 + distributed_backend: deepspeed + is_resume: false + pretrained_checkpoint: null + resume_step: 0 + reload_modules: null + optimizer: + name: AdamW + betas: + - 0.9 + - 0.95 + eps: 1.0e-08 + weight_decay: 1.0e-08 + fused: true + save_format: pt +run_id: half_cheetah_l0_gr00t_h1_5fps_2h200_20260917 +run_root_dir: /mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/half_cheetah/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: half_cheetah_l0_gr00t_h1_5fps_2h200_20260917 +wandb_group: gr00t-six-env-h1-5fps +wandb_tags: +- half_cheetah +- 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/half_cheetah/zero_latency/train.yaml +output_dir: /mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/half_cheetah/zero_latency/vla/training/half_cheetah_l0_gr00t_h1_5fps_2h200_20260917 diff --git a/zero-latency/half-cheetah/vla/starvla-qwengr00t-h1/half_cheetah_l0_gr00t_h1_5fps_2h200_20260917/config.yaml b/zero-latency/half-cheetah/vla/starvla-qwengr00t-h1/half_cheetah_l0_gr00t_h1_5fps_2h200_20260917/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..6de86bcf5dec25d75a694285e7901734924da6dc --- /dev/null +++ b/zero-latency/half-cheetah/vla/starvla-qwengr00t-h1/half_cheetah_l0_gr00t_h1_5fps_2h200_20260917/config.yaml @@ -0,0 +1,132 @@ +framework: + name: QwenGR00T + qwenvl: + base_vlm: ${oc.env:GR00T_BACKBONE_DIR} + attn_implementation: flash_attention_2 + vl_hidden_dim: 2048 + enable_gradient_checkpointing: true + action_model: + action_model_type: DiT-B + action_hidden_dim: 1024 + hidden_size: 1024 + add_pos_embed: true + max_seq_len: 1024 + action_dim: 6 + state_dim: 17 + task_objective: null + action_horizon: 1 + repeated_diffusion_steps: 8 + noise_beta_alpha: 1.5 + noise_beta_beta: 1.0 + noise_s: 0.999 + num_timestep_buckets: 1000 + num_inference_timesteps: 4 + num_target_vision_tokens: 32 + diffusion_model_cfg: + cross_attention_dim: 2560 + dropout: 0.2 + final_dropout: true + interleave_self_attention: true + norm_type: ada_norm + num_layers: 16 + output_dim: 1024 + positional_embeddings: null + action_env_dim: 6 + future_action_window_size: 0 + past_action_window_size: 0 +version_id: '0.21' +datasets: + vla_data: + include_state: true + obs_image_size: + - 224 + - 224 + image_mode: single + stitch_grid: + - 2 + - 2 + num_obs_frames: 1 + gymnasium_task_contract: &id001 + action_space: + dtype: float32 + high: + - 1.0 + - 1.0 + - 1.0 + - 1.0 + - 1.0 + - 1.0 + labels: + - back_thigh_torque + - back_shin_torque + - back_foot_torque + - front_thigh_torque + - front_shin_torque + - front_foot_torque + low: + - -1.0 + - -1.0 + - -1.0 + - -1.0 + - -1.0 + - -1.0 + type: box + base_prompt: Make the Half Cheetah run forward as fast as possible. Predict + six continuous torques in [-1, 1] ordered as back thigh, back shin, back foot, + front thigh, front shin, and front foot. + env_fps: 5 + env_id: LatencyBench/HalfCheetahContinuous-v0 + frame_stack: 1 + make_kwargs: + base_env_id: HalfCheetah-v4 + base_make_kwargs: {} + render_mode: rgb_array + noop_action: + - 0.0 + - 0.0 + - 0.0 + - 0.0 + - 0.0 + - 0.0 + obs_fps: 5 + registration_imports: + - latency_bench.envs.gymnasium_half_cheetah + state_space: + labels: + - front_tip_height + - front_tip_angle + - back_thigh_angle + - back_shin_angle + - back_foot_angle + - front_thigh_angle + - front_shin_angle + - front_foot_angle + - front_tip_x_velocity + - front_tip_z_velocity + - front_tip_angular_velocity + - back_thigh_angular_velocity + - back_shin_angular_velocity + - back_foot_angular_velocity + - front_thigh_angular_velocity + - front_shin_angular_velocity + - front_foot_angular_velocity + task_name: half_cheetah_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/half-cheetah/vla/starvla-qwengr00t-h1/half_cheetah_l0_gr00t_h1_5fps_2h200_20260917/dataset_statistics.json b/zero-latency/half-cheetah/vla/starvla-qwengr00t-h1/half_cheetah_l0_gr00t_h1_5fps_2h200_20260917/dataset_statistics.json new file mode 100644 index 0000000000000000000000000000000000000000..e0a2563a2c7d2978230a2ba367603622a32b0d04 --- /dev/null +++ b/zero-latency/half-cheetah/vla/starvla-qwengr00t-h1/half_cheetah_l0_gr00t_h1_5fps_2h200_20260917/dataset_statistics.json @@ -0,0 +1,180 @@ +{ + "new_embodiment": { + "action": { + "mean": [ + 0.04852694645524025, + -0.07289591431617737, + -0.16553527116775513, + 0.3926909863948822, + 0.4024640917778015, + -0.15250922739505768 + ], + "std": [ + 0.9902900457382202, + 0.9729189872741699, + 0.92887282371521, + 0.7155725359916687, + 0.5879771113395691, + 0.8828322291374207 + ], + "max": [ + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0 + ], + "min": [ + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0 + ], + "q01": [ + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0 + ], + "q99": [ + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0 + ], + "mask": [ + true, + true, + true, + true, + true, + true + ] + }, + "state": { + "mean": [ + -0.16455742716789246, + -0.6733709573745728, + -0.07721110433340073, + -0.1043669581413269, + -0.20796853303909302, + 0.531654417514801, + 0.23800496757030487, + -0.08999839425086975, + 0.4533954858779907, + 0.03410167247056961, + -0.009492190554738045, + 0.04943292587995529, + -0.05999681353569031, + 0.11326351761817932, + -0.022488147020339966, + -0.027951883152127266, + -0.10890804976224899 + ], + "std": [ + 0.1308881938457489, + 0.11011626571416855, + 0.5556989908218384, + 0.4154678285121918, + 0.592194676399231, + 0.2803356647491455, + 0.41065067052841187, + 0.5386135578155518, + 0.2725602984428406, + 0.18254218995571136, + 0.26073697209358215, + 0.6522123217582703, + 0.29445716738700867, + 0.5733875632286072, + 0.2830783426761627, + 0.4337209165096283, + 0.5481678247451782 + ], + "max": [ + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0 + ], + "min": [ + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0 + ], + "q01": [ + -0.8618907463550568, + -0.7782192385196686, + -0.963983958363533, + -0.7420942264795304, + -0.9641105020046234, + -0.15345445454120635, + -0.5540750181674957, + -0.9644814884662628, + -0.5328765374422073, + -0.4222735470533371, + -0.5928484404087067, + -0.8578835678100586, + -0.7154878216981888, + -0.918103654384613, + -0.5789287960529328, + -0.8079770970344543, + -0.8612057662010193 + ], + "q99": [ + 0.11475664258003207, + 0.09281709313392594, + 0.7771777260303496, + 0.785496914386749, + 0.8948100674152373, + 0.9008247101306913, + 0.8394721925258635, + 0.8582286417484283, + 0.8433885538578033, + 0.42764286875724783, + 0.4806103372573851, + 0.8169456958770751, + 0.4850697159767148, + 0.9674746108055114, + 0.5811240792274471, + 0.7913007164001464, + 0.7998981952667232 + ] + }, + "num_transitions": 90000, + "num_trajectories": 90 + } +} \ No newline at end of file diff --git a/zero-latency/half-cheetah/vla/starvla-qwengr00t-h1/half_cheetah_l0_gr00t_h1_5fps_2h200_20260917/evaluation/l0_sim20/config.yaml b/zero-latency/half-cheetah/vla/starvla-qwengr00t-h1/half_cheetah_l0_gr00t_h1_5fps_2h200_20260917/evaluation/l0_sim20/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..f3d6a9668a87be5e48504d555675d70573544d2b --- /dev/null +++ b/zero-latency/half-cheetah/vla/starvla-qwengr00t-h1/half_cheetah_l0_gr00t_h1_5fps_2h200_20260917/evaluation/l0_sim20/config.yaml @@ -0,0 +1,169 @@ +experiment: + name: gr00t_half_cheetah_zero + seed: 42 +backend: + type: sample_factory + algo: APPO + device: cuda + train_dir: /mnt/checkpoints/latency-sensitive-bench/small_models/half_cheetah + restart_behavior: overwrite + run_mode: eval +executor: + mode: simulated + inference_devices: + - cuda:0 + inference_batch_size: 1 + simulated_inference_pool: true + simulated_worker_capacity: 1 +env: + name: gymnasium + task_name: half_cheetah + env_id: LatencyBench/HalfCheetahContinuous-v0 + registration_imports: + - latency_bench.envs.gymnasium_half_cheetah + make_kwargs: + base_env_id: HalfCheetah-v4 + base_make_kwargs: {} + render_mode: rgb_array + env_fps: 5 + obs_fps: 5 + frame_stack: 1 + state_space: + labels: + - front_tip_height + - front_tip_angle + - back_thigh_angle + - back_shin_angle + - back_foot_angle + - front_thigh_angle + - front_shin_angle + - front_foot_angle + - front_tip_x_velocity + - front_tip_z_velocity + - front_tip_angular_velocity + - back_thigh_angular_velocity + - back_shin_angular_velocity + - back_foot_angular_velocity + - front_thigh_angular_velocity + - front_shin_angular_velocity + - front_foot_angular_velocity + action_space: + dtype: float32 + high: + - 1.0 + - 1.0 + - 1.0 + - 1.0 + - 1.0 + - 1.0 + labels: + - back_thigh_torque + - back_shin_torque + - back_foot_torque + - front_thigh_torque + - front_shin_torque + - front_foot_torque + low: + - -1.0 + - -1.0 + - -1.0 + - -1.0 + - -1.0 + - -1.0 + type: box + noop_action: + - 0.0 + - 0.0 + - 0.0 + - 0.0 + - 0.0 + - 0.0 + base_prompt: Make the Half Cheetah run forward as fast as possible. Predict six + continuous torques in [-1, 1] ordered as back thigh, back shin, back foot, front + thigh, front shin, and front foot. + obs_resize: + - 224 + - 224 +latency: + method: zero + sync_cuda: false +scheduler: + hold_policy: hold + ordering_policy: issue_order_fifo +policy: + type: starvla + checkpoint_path: /mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/half_cheetah/zero_latency/bundle/checkpoints/model.pt + model_config_path: /mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/half_cheetah/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/half_cheetah/zero_latency/bundle/manifest.json + latency_prompt_map_path: /mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/half_cheetah/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 + 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/half_cheetah/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/half-cheetah/vla/starvla-qwengr00t-h1/half_cheetah_l0_gr00t_h1_5fps_2h200_20260917/evaluation/l0_sim20/episode_metrics.jsonl b/zero-latency/half-cheetah/vla/starvla-qwengr00t-h1/half_cheetah_l0_gr00t_h1_5fps_2h200_20260917/evaluation/l0_sim20/episode_metrics.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..f3e3068227579164d2031f33e9a3cee28ea2c0e9 --- /dev/null +++ b/zero-latency/half-cheetah/vla/starvla-qwengr00t-h1/half_cheetah_l0_gr00t_h1_5fps_2h200_20260917/evaluation/l0_sim20/episode_metrics.jsonl @@ -0,0 +1,20 @@ +{"episode_id": 0, "episode_return": 4947.113453190952, "episode_return_env": 4947.113453190952, "game_score": null, "mean_latency_ms": 0.0, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/half_cheetah/zero_latency/bundle/checkpoints/model.pt", "config_name": "gr00t_half_cheetah_zero", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/HalfCheetahContinuous-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/half_cheetah/zero_latency/l0_sim20", "policy_id": "starvla", "profile_ref": null, "run_name": "gr00t_half_cheetah_zero", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_run_id": null, "submitted_observation_frames": 1000, "workload_id": null}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 0.0, "p99_latency_ms": 0.0, "return_raw": null, "survival_steps": 1000} +{"episode_id": 1, "episode_return": 5491.527318723737, "episode_return_env": 5491.527318723737, "game_score": null, "mean_latency_ms": 0.0, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/half_cheetah/zero_latency/bundle/checkpoints/model.pt", "config_name": "gr00t_half_cheetah_zero", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/HalfCheetahContinuous-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/half_cheetah/zero_latency/l0_sim20", "policy_id": "starvla", "profile_ref": null, "run_name": "gr00t_half_cheetah_zero", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_run_id": null, "submitted_observation_frames": 1000, "workload_id": null}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 0.0, "p99_latency_ms": 0.0, "return_raw": null, "survival_steps": 1000} +{"episode_id": 2, "episode_return": 5262.7505395850785, "episode_return_env": 5262.7505395850785, "game_score": null, "mean_latency_ms": 0.0, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/half_cheetah/zero_latency/bundle/checkpoints/model.pt", "config_name": "gr00t_half_cheetah_zero", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/HalfCheetahContinuous-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/half_cheetah/zero_latency/l0_sim20", "policy_id": "starvla", "profile_ref": null, "run_name": "gr00t_half_cheetah_zero", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_run_id": null, "submitted_observation_frames": 1000, "workload_id": null}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 0.0, "p99_latency_ms": 0.0, "return_raw": null, "survival_steps": 1000} +{"episode_id": 3, "episode_return": 5384.72348148733, "episode_return_env": 5384.72348148733, "game_score": null, "mean_latency_ms": 0.0, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/half_cheetah/zero_latency/bundle/checkpoints/model.pt", "config_name": "gr00t_half_cheetah_zero", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/HalfCheetahContinuous-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/half_cheetah/zero_latency/l0_sim20", "policy_id": "starvla", "profile_ref": null, "run_name": "gr00t_half_cheetah_zero", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_run_id": null, "submitted_observation_frames": 1000, "workload_id": null}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 0.0, "p99_latency_ms": 0.0, "return_raw": null, "survival_steps": 1000} +{"episode_id": 4, "episode_return": 5261.018543073299, "episode_return_env": 5261.018543073299, "game_score": null, "mean_latency_ms": 0.0, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/half_cheetah/zero_latency/bundle/checkpoints/model.pt", "config_name": "gr00t_half_cheetah_zero", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/HalfCheetahContinuous-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/half_cheetah/zero_latency/l0_sim20", "policy_id": "starvla", "profile_ref": null, "run_name": "gr00t_half_cheetah_zero", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_run_id": null, "submitted_observation_frames": 1000, "workload_id": null}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 0.0, "p99_latency_ms": 0.0, "return_raw": null, "survival_steps": 1000} +{"episode_id": 5, "episode_return": 4506.0295300219295, "episode_return_env": 4506.0295300219295, "game_score": null, "mean_latency_ms": 0.0, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/half_cheetah/zero_latency/bundle/checkpoints/model.pt", "config_name": "gr00t_half_cheetah_zero", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/HalfCheetahContinuous-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/half_cheetah/zero_latency/l0_sim20", "policy_id": "starvla", "profile_ref": null, "run_name": "gr00t_half_cheetah_zero", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_run_id": null, "submitted_observation_frames": 1000, "workload_id": null}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 0.0, "p99_latency_ms": 0.0, "return_raw": null, "survival_steps": 1000} +{"episode_id": 6, "episode_return": 5261.747848030333, "episode_return_env": 5261.747848030333, "game_score": null, "mean_latency_ms": 0.0, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/half_cheetah/zero_latency/bundle/checkpoints/model.pt", "config_name": "gr00t_half_cheetah_zero", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/HalfCheetahContinuous-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/half_cheetah/zero_latency/l0_sim20", "policy_id": "starvla", "profile_ref": null, "run_name": "gr00t_half_cheetah_zero", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_run_id": null, "submitted_observation_frames": 1000, "workload_id": null}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 0.0, "p99_latency_ms": 0.0, "return_raw": null, "survival_steps": 1000} +{"episode_id": 7, "episode_return": 5174.639195548416, "episode_return_env": 5174.639195548416, "game_score": null, "mean_latency_ms": 0.0, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/half_cheetah/zero_latency/bundle/checkpoints/model.pt", "config_name": "gr00t_half_cheetah_zero", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/HalfCheetahContinuous-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/half_cheetah/zero_latency/l0_sim20", "policy_id": "starvla", "profile_ref": null, "run_name": "gr00t_half_cheetah_zero", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_run_id": null, "submitted_observation_frames": 1000, "workload_id": null}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 0.0, "p99_latency_ms": 0.0, "return_raw": null, "survival_steps": 1000} +{"episode_id": 8, "episode_return": 4687.507325634152, "episode_return_env": 4687.507325634152, "game_score": null, "mean_latency_ms": 0.0, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/half_cheetah/zero_latency/bundle/checkpoints/model.pt", "config_name": "gr00t_half_cheetah_zero", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/HalfCheetahContinuous-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/half_cheetah/zero_latency/l0_sim20", "policy_id": "starvla", "profile_ref": null, "run_name": "gr00t_half_cheetah_zero", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_run_id": null, "submitted_observation_frames": 1000, "workload_id": null}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 0.0, "p99_latency_ms": 0.0, "return_raw": null, "survival_steps": 1000} +{"episode_id": 9, "episode_return": 4982.420554479696, "episode_return_env": 4982.420554479696, "game_score": null, "mean_latency_ms": 0.0, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/half_cheetah/zero_latency/bundle/checkpoints/model.pt", "config_name": "gr00t_half_cheetah_zero", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/HalfCheetahContinuous-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/half_cheetah/zero_latency/l0_sim20", "policy_id": "starvla", "profile_ref": null, "run_name": "gr00t_half_cheetah_zero", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_run_id": null, "submitted_observation_frames": 1000, "workload_id": null}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 0.0, "p99_latency_ms": 0.0, "return_raw": null, "survival_steps": 1000} +{"episode_id": 10, "episode_return": 4018.293038783439, "episode_return_env": 4018.293038783439, "game_score": null, "mean_latency_ms": 0.0, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/half_cheetah/zero_latency/bundle/checkpoints/model.pt", "config_name": "gr00t_half_cheetah_zero", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/HalfCheetahContinuous-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/half_cheetah/zero_latency/l0_sim20", "policy_id": "starvla", "profile_ref": null, "run_name": "gr00t_half_cheetah_zero", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_run_id": null, "submitted_observation_frames": 1000, "workload_id": null}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 0.0, "p99_latency_ms": 0.0, "return_raw": null, "survival_steps": 1000} +{"episode_id": 11, "episode_return": 5510.720264117774, "episode_return_env": 5510.720264117774, "game_score": null, "mean_latency_ms": 0.0, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/half_cheetah/zero_latency/bundle/checkpoints/model.pt", "config_name": "gr00t_half_cheetah_zero", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/HalfCheetahContinuous-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/half_cheetah/zero_latency/l0_sim20", "policy_id": "starvla", "profile_ref": null, "run_name": "gr00t_half_cheetah_zero", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_run_id": null, "submitted_observation_frames": 1000, "workload_id": null}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 0.0, "p99_latency_ms": 0.0, "return_raw": null, "survival_steps": 1000} +{"episode_id": 12, "episode_return": 2509.0270585296953, "episode_return_env": 2509.0270585296953, "game_score": null, "mean_latency_ms": 0.0, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/half_cheetah/zero_latency/bundle/checkpoints/model.pt", "config_name": "gr00t_half_cheetah_zero", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/HalfCheetahContinuous-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/half_cheetah/zero_latency/l0_sim20", "policy_id": "starvla", "profile_ref": null, "run_name": "gr00t_half_cheetah_zero", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_run_id": null, "submitted_observation_frames": 1000, "workload_id": null}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 0.0, "p99_latency_ms": 0.0, "return_raw": null, "survival_steps": 1000} +{"episode_id": 13, "episode_return": 5560.689573549269, "episode_return_env": 5560.689573549269, "game_score": null, "mean_latency_ms": 0.0, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/half_cheetah/zero_latency/bundle/checkpoints/model.pt", "config_name": "gr00t_half_cheetah_zero", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/HalfCheetahContinuous-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/half_cheetah/zero_latency/l0_sim20", "policy_id": "starvla", "profile_ref": null, "run_name": "gr00t_half_cheetah_zero", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_run_id": null, "submitted_observation_frames": 1000, "workload_id": null}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 0.0, "p99_latency_ms": 0.0, "return_raw": null, "survival_steps": 1000} +{"episode_id": 14, "episode_return": 5383.194311117947, "episode_return_env": 5383.194311117947, "game_score": null, "mean_latency_ms": 0.0, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/half_cheetah/zero_latency/bundle/checkpoints/model.pt", "config_name": "gr00t_half_cheetah_zero", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/HalfCheetahContinuous-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/half_cheetah/zero_latency/l0_sim20", "policy_id": "starvla", "profile_ref": null, "run_name": "gr00t_half_cheetah_zero", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_run_id": null, "submitted_observation_frames": 1000, "workload_id": null}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 0.0, "p99_latency_ms": 0.0, "return_raw": null, "survival_steps": 1000} +{"episode_id": 15, "episode_return": 5377.5802411971345, "episode_return_env": 5377.5802411971345, "game_score": null, "mean_latency_ms": 0.0, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/half_cheetah/zero_latency/bundle/checkpoints/model.pt", "config_name": "gr00t_half_cheetah_zero", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/HalfCheetahContinuous-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/half_cheetah/zero_latency/l0_sim20", "policy_id": "starvla", "profile_ref": null, "run_name": "gr00t_half_cheetah_zero", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_run_id": null, "submitted_observation_frames": 1000, "workload_id": null}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 0.0, "p99_latency_ms": 0.0, "return_raw": null, "survival_steps": 1000} +{"episode_id": 16, "episode_return": 5524.619208565534, "episode_return_env": 5524.619208565534, "game_score": null, "mean_latency_ms": 0.0, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/half_cheetah/zero_latency/bundle/checkpoints/model.pt", "config_name": "gr00t_half_cheetah_zero", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/HalfCheetahContinuous-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/half_cheetah/zero_latency/l0_sim20", "policy_id": "starvla", "profile_ref": null, "run_name": "gr00t_half_cheetah_zero", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_run_id": null, "submitted_observation_frames": 1000, "workload_id": null}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 0.0, "p99_latency_ms": 0.0, "return_raw": null, "survival_steps": 1000} +{"episode_id": 17, "episode_return": 5691.982912712619, "episode_return_env": 5691.982912712619, "game_score": null, "mean_latency_ms": 0.0, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/half_cheetah/zero_latency/bundle/checkpoints/model.pt", "config_name": "gr00t_half_cheetah_zero", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/HalfCheetahContinuous-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/half_cheetah/zero_latency/l0_sim20", "policy_id": "starvla", "profile_ref": null, "run_name": "gr00t_half_cheetah_zero", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_run_id": null, "submitted_observation_frames": 1000, "workload_id": null}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 0.0, "p99_latency_ms": 0.0, "return_raw": null, "survival_steps": 1000} +{"episode_id": 18, "episode_return": 4074.5527029930076, "episode_return_env": 4074.5527029930076, "game_score": null, "mean_latency_ms": 0.0, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/half_cheetah/zero_latency/bundle/checkpoints/model.pt", "config_name": "gr00t_half_cheetah_zero", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/HalfCheetahContinuous-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/half_cheetah/zero_latency/l0_sim20", "policy_id": "starvla", "profile_ref": null, "run_name": "gr00t_half_cheetah_zero", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_run_id": null, "submitted_observation_frames": 1000, "workload_id": null}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 0.0, "p99_latency_ms": 0.0, "return_raw": null, "survival_steps": 1000} +{"episode_id": 19, "episode_return": 5639.3189698612805, "episode_return_env": 5639.3189698612805, "game_score": null, "mean_latency_ms": 0.0, "metadata": {"checkpoint_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/half_cheetah/zero_latency/bundle/checkpoints/model.pt", "config_name": "gr00t_half_cheetah_zero", "dropped_observation_count": 0, "env_fps": 5.0, "env_id": "LatencyBench/HalfCheetahContinuous-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/half_cheetah/zero_latency/l0_sim20", "policy_id": "starvla", "profile_ref": null, "run_name": "gr00t_half_cheetah_zero", "sim_inference_pool_workers": 1, "simulated_worker_capacity": 1, "source_run_id": null, "submitted_observation_frames": 1000, "workload_id": null}, "num_actions": 1000, "num_dropped_actions": 0, "num_invalid_actions": 0, "p90_latency_ms": 0.0, "p99_latency_ms": 0.0, "return_raw": null, "survival_steps": 1000} diff --git a/zero-latency/half-cheetah/vla/starvla-qwengr00t-h1/half_cheetah_l0_gr00t_h1_5fps_2h200_20260917/evaluation/l0_sim20/verification.json b/zero-latency/half-cheetah/vla/starvla-qwengr00t-h1/half_cheetah_l0_gr00t_h1_5fps_2h200_20260917/evaluation/l0_sim20/verification.json new file mode 100644 index 0000000000000000000000000000000000000000..bf5c5b745f27ed9d0b594dcf3b0b06e24774d636 --- /dev/null +++ b/zero-latency/half-cheetah/vla/starvla-qwengr00t-h1/half_cheetah_l0_gr00t_h1_5fps_2h200_20260917/evaluation/l0_sim20/verification.json @@ -0,0 +1,95 @@ +{ + "task": "half_cheetah", + "state": "L0_SIM20_COMPLETED_AND_VERIFIED", + "verified_at": "2026-09-17T15:02:09.534292+00:00", + "evaluation_finished_at": "2026-09-17T14:50:07.672102+00:00", + "episodes": 20, + "seeds": [ + 42, + 43, + 44, + 45, + 46, + 47, + 48, + 49, + 50, + 51, + 52, + 53, + 54, + 55, + 56, + 57, + 58, + 59, + 60, + 61 + ], + "env_fps": 5, + "obs_fps": 5, + "latency": "zero", + "mode": "simulated", + "max_raw_steps": 1000, + "returns": [ + 4947.113453190952, + 5491.527318723737, + 5262.7505395850785, + 5384.72348148733, + 5261.018543073299, + 4506.0295300219295, + 5261.747848030333, + 5174.639195548416, + 4687.507325634152, + 4982.420554479696, + 4018.293038783439, + 5510.720264117774, + 2509.0270585296953, + 5560.689573549269, + 5383.194311117947, + 5377.5802411971345, + 5524.619208565534, + 5691.982912712619, + 4074.5527029930076, + 5639.3189698612805 + ], + "mean_return": 5012.472803560131, + "population_std_return": 742.2936832677323, + "lengths": [ + 1000, + 1000, + 1000, + 1000, + 1000, + 1000, + 1000, + 1000, + 1000, + 1000, + 1000, + 1000, + 1000, + 1000, + 1000, + 1000, + 1000, + 1000, + 1000, + 1000 + ], + "mean_length": 1000, + "total_raw_steps": 20000, + "raw_step_rewards_match_episode_returns": true, + "raw_step_indices_contiguous": true, + "dropped_actions": 0, + "dropped_observations": 0, + "invalid_actions": 0, + "action_events": 20000, + "all_action_latencies_zero": true, + "training_steps": 5000, + "global_batch": 64, + "checkpoint_sha256": "8f007f801c45ed8476546ebf7cc96e60283d2ffd05a6d30feaa529d7ea7f658f", + "command_semantics": "Generic adapter applied_action may be preclip; native environment clipping unchanged. No preclip-bound assertion was imposed.", + "remaining": "RTX3090 P/B, actual-profile teacher/data/profile-VLA, F; not yet complete", + "acceptance": "Full pipeline acceptance not established" +} diff --git a/zero-latency/half-cheetah/vla/starvla-qwengr00t-h1/half_cheetah_l0_gr00t_h1_5fps_2h200_20260917/latency_prompt_map.json b/zero-latency/half-cheetah/vla/starvla-qwengr00t-h1/half_cheetah_l0_gr00t_h1_5fps_2h200_20260917/latency_prompt_map.json new file mode 100644 index 0000000000000000000000000000000000000000..7c2db5bd52d0a369ac248a1189b89958c14f3dd7 --- /dev/null +++ b/zero-latency/half-cheetah/vla/starvla-qwengr00t-h1/half_cheetah_l0_gr00t_h1_5fps_2h200_20260917/latency_prompt_map.json @@ -0,0 +1,7 @@ +{ + "0": { + "prompt": "Make the Half Cheetah run forward as fast as possible. Predict six continuous torques in [-1, 1] ordered as back thigh, back shin, back foot, front thigh, front shin, and front 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/half-cheetah/vla/starvla-qwengr00t-h1/half_cheetah_l0_gr00t_h1_5fps_2h200_20260917/manifest.json b/zero-latency/half-cheetah/vla/starvla-qwengr00t-h1/half_cheetah_l0_gr00t_h1_5fps_2h200_20260917/manifest.json new file mode 100644 index 0000000000000000000000000000000000000000..fd5dd0a9da7d543c51bf6b3fbdd2956924070eb1 --- /dev/null +++ b/zero-latency/half-cheetah/vla/starvla-qwengr00t-h1/half_cheetah_l0_gr00t_h1_5fps_2h200_20260917/manifest.json @@ -0,0 +1,177 @@ +{ + "dataset_name": "half_cheetah_h1_5fps_l0", + "env_name": "half_cheetah_rgb_state", + "episodes": 90, + "frames": 90000, + "task_prompts": [ + "Make the Half Cheetah run forward as fast as possible. Predict six continuous torques in [-1, 1] ordered as back thigh, back shin, back foot, front thigh, front shin, and front 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/half_cheetah/zero_latency/data/raw_5fps", + "integration_name": "gymnasium", + "task_name": "half_cheetah_rgb_state", + "action_layout": "gymnasium_continuous_v1", + "action_labels": [ + "back_thigh_torque", + "back_shin_torque", + "back_foot_torque", + "front_thigh_torque", + "front_shin_torque", + "front_foot_torque" + ], + "carrier_action_labels": [ + "back_thigh_torque", + "back_shin_torque", + "back_foot_torque", + "front_thigh_torque", + "front_shin_torque", + "front_foot_torque" + ], + "action_dim": 6, + "active_action_dim": 6, + "action_carrier": "native", + "rows_unit": "decision_step", + "obs_fps": 5, + "obs_stride_raw_frames": 1, + "uses_state": true, + "state_dim": 17, + "state_labels": [ + "front_tip_height", + "front_tip_angle", + "back_thigh_angle", + "back_shin_angle", + "back_foot_angle", + "front_thigh_angle", + "front_shin_angle", + "front_foot_angle", + "front_tip_x_velocity", + "front_tip_z_velocity", + "front_tip_angular_velocity", + "back_thigh_angular_velocity", + "back_shin_angular_velocity", + "back_foot_angular_velocity", + "front_thigh_angular_velocity", + "front_shin_angular_velocity", + "front_foot_angular_velocity" + ], + "state_normalization": { + "type": "min_max", + "min": [ + -0.5893064737319946, + -1.0017330646514893, + -0.7351728081703186, + -0.8334724307060242, + -0.535628616809845, + -1.051438808441162, + -0.989691436290741, + -0.6309299468994141, + -2.3919641971588135, + -4.381032943725586, + -8.257936477661133, + -23.92654800415039, + -26.088411331176758, + -22.936174392700195, + -22.609085083007812, + -26.382095336914062, + -18.172616958618164 + ], + "max": [ + 0.7243065237998962, + 6.765694618225098, + 1.062463402748108, + 0.8386014103889465, + 0.8500239253044128, + 0.8450682759284973, + 0.9588844776153564, + 0.6611841917037964, + 9.299489974975586, + 3.903606414794922, + 8.0291166305542, + 20.77332305908203, + 29.168107986450195, + 19.582447052001953, + 24.40692138671875, + 28.59392547607422, + 22.969768524169922 + ] + }, + "robot_type": "rl_games_gymnasium", + "latency_prompt_map_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/half_cheetah/zero_latency/data/lerobot/half_cheetah_h1_5fps_l0/latency_prompt_map.json", + "custom_mixtures_path": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/half_cheetah/zero_latency/data/lerobot/_generated_mixtures/half_cheetah_h1_5fps_l0.json", + "gymnasium_task": { + "action_space": { + "dtype": "float32", + "high": [ + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0 + ], + "labels": [ + "back_thigh_torque", + "back_shin_torque", + "back_foot_torque", + "front_thigh_torque", + "front_shin_torque", + "front_foot_torque" + ], + "low": [ + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0 + ], + "type": "box" + }, + "base_prompt": "Make the Half Cheetah run forward as fast as possible. Predict six continuous torques in [-1, 1] ordered as back thigh, back shin, back foot, front thigh, front shin, and front foot.", + "env_fps": 5, + "env_id": "LatencyBench/HalfCheetahContinuous-v0", + "frame_stack": 1, + "make_kwargs": { + "base_env_id": "HalfCheetah-v4", + "base_make_kwargs": {}, + "render_mode": "rgb_array" + }, + "noop_action": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ], + "obs_fps": 5, + "registration_imports": [ + "latency_bench.envs.gymnasium_half_cheetah" + ], + "state_space": { + "labels": [ + "front_tip_height", + "front_tip_angle", + "back_thigh_angle", + "back_shin_angle", + "back_foot_angle", + "front_thigh_angle", + "front_shin_angle", + "front_foot_angle", + "front_tip_x_velocity", + "front_tip_z_velocity", + "front_tip_angular_velocity", + "back_thigh_angular_velocity", + "back_shin_angular_velocity", + "back_foot_angular_velocity", + "front_thigh_angular_velocity", + "front_shin_angular_velocity", + "front_foot_angular_velocity" + ] + }, + "task_name": "half_cheetah_rgb_state" + }, + "validation_dataset_name": "half_cheetah_h1_5fps_l0__val", + "validation_episodes": 10, + "validation_frames": 10000 +} \ No newline at end of file diff --git a/zero-latency/half-cheetah/vla/starvla-qwengr00t-h1/half_cheetah_l0_gr00t_h1_5fps_2h200_20260917/provenance.json b/zero-latency/half-cheetah/vla/starvla-qwengr00t-h1/half_cheetah_l0_gr00t_h1_5fps_2h200_20260917/provenance.json new file mode 100644 index 0000000000000000000000000000000000000000..0288175e38b38a32ef714beaa1f3d54abff52099 --- /dev/null +++ b/zero-latency/half-cheetah/vla/starvla-qwengr00t-h1/half_cheetah_l0_gr00t_h1_5fps_2h200_20260917/provenance.json @@ -0,0 +1,34 @@ +{ + "task": "half-cheetah", + "model": "qwengr00t", + "training_condition": "zero-latency", + "training_run_id": "half_cheetah_l0_gr00t_h1_5fps_2h200_20260917", + "source": { + "repo_id": "latency-sensitive-bench/Standard-Pipeline", + "revision": "2208875f92b21e3f8ec2363a30a6705dfe56cc04", + "path": "half_cheetah/zero_latency/GR00T", + "url": "https://huggingface.co/latency-sensitive-bench/Standard-Pipeline/tree/2208875f92b21e3f8ec2363a30a6705dfe56cc04/half_cheetah/zero_latency/GR00T" + }, + "hf_repo_id": "latency-sensitive-bench/benchmark-models", + "hf_path": "zero-latency/half-cheetah/vla/starvla-qwengr00t-h1/half_cheetah_l0_gr00t_h1_5fps_2h200_20260917", + "checkpoint": { + "source_file": "half_cheetah/zero_latency/GR00T/checkpoints/model.pt", + "source_sha256": "8f007f801c45ed8476546ebf7cc96e60283d2ffd05a6d30feaa529d7ea7f658f", + "source_bytes": 9976860963, + "file": "checkpoints/model.pt", + "selection_rule": "only published VLA checkpoint", + "method": "copy", + "sha256": "8f007f801c45ed8476546ebf7cc96e60283d2ffd05a6d30feaa529d7ea7f658f", + "bytes": 9976860963 + }, + "config_source": "half_cheetah/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/half-cheetah/vla/starvla-qwengr00t-h1/half_cheetah_l0_gr00t_h1_5fps_2h200_20260917/source/README.md b/zero-latency/half-cheetah/vla/starvla-qwengr00t-h1/half_cheetah_l0_gr00t_h1_5fps_2h200_20260917/source/README.md new file mode 100644 index 0000000000000000000000000000000000000000..67c5810638d3139cafa401d96675c37d31b8247e --- /dev/null +++ b/zero-latency/half-cheetah/vla/starvla-qwengr00t-h1/half_cheetah_l0_gr00t_h1_5fps_2h200_20260917/source/README.md @@ -0,0 +1,12 @@ +# HalfCheetah H1 / 5 FPS: verified stage status + +Zero-latency GR00T training completed 5000 updates. L0 simulated evaluation: **5012.472804 ± 742.293683** (population SD), 20 episodes, seeds 42–61, all 1000 raw steps. + +Measured P and B are complete. B20 return: **-262.465910 ± 74.942120**, all 1000 raw steps, zero dropped/invalid actions. B combines 16 completed original episodes whose core metrics were reconstructed with frozen `compute_episode_metrics`, plus four newly evaluated remaining seeds 58–61 after the container restart. P and the completed 16 episodes were not rerun. Some metadata for those 16 episodes, including actual environment FPS and end-drain counters, could not be recovered and remains `None`; this is not lossless full-aggregate metadata recovery. + +The actual-profile APPO teacher completed its original 10M budget (10,002,432 actual environment steps). Final20: 1843.155080 ± 29.760273; training-best20: 1847.216029 ± 27.628394. Both had **0/20 returns strictly greater than 3000**. Training-best was selected, but the scientific gate failed. E10, probe, profile data collection, profile VLA training and F were not run. The budget and gate were not relaxed. Full pipeline acceptance is not established. + +- [Pinned final B and recovery evidence](https://huggingface.co/datasets/latency-sensitive-bench/Standard-Pipeline/tree/223f2106c10b264b820963391ec546d8dce039a5/half_cheetah/profile_latency/GR00T/h1_5fps_20260917/baseline) +- [Pinned teacher gate audit](https://huggingface.co/datasets/latency-sensitive-bench/Standard-Pipeline/tree/6391d8254f2b9d38d6f4a8c5d5168cfd5f10e30a/half_cheetah/profile_latency/GR00T/h1_5fps_20260917/small_model/evaluation) + +Existing inference weights and historical validation receipts are unchanged. diff --git a/zero-latency/half-cheetah/vla/starvla-qwengr00t-h1/half_cheetah_l0_gr00t_h1_5fps_2h200_20260917/source/provenance.json b/zero-latency/half-cheetah/vla/starvla-qwengr00t-h1/half_cheetah_l0_gr00t_h1_5fps_2h200_20260917/source/provenance.json new file mode 100644 index 0000000000000000000000000000000000000000..21326293a734124afc5a33f18c42d13478f7a819 --- /dev/null +++ b/zero-latency/half-cheetah/vla/starvla-qwengr00t-h1/half_cheetah_l0_gr00t_h1_5fps_2h200_20260917/source/provenance.json @@ -0,0 +1,64 @@ +{ + "task": "half_cheetah", + "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": "half_cheetah_l0_gr00t_h1_5fps_2h200_20260917", + "condition": "zero_latency", + "source": { + "archive_sha256": "9847e8924ba7a6446aaa3ce735f4dc2849046c37e21aeb8867f3154ab163ede0", + "root_revision": "1681edb8d9e4f7d10b7dfa647670e01e0d5cce37", + "sample_factory_revision": "4b11d8754ffc989cdd15a340cfc9be0277377d5b", + "starvla_revision": "1d0d7b139d1725cab268cb9dae2007ef4c5d05d2", + "raw_data": { + "source_repo": "latency-sensitive-bench/Standard-Pipeline", + "source_revision": "571cb5801720488ed6458b9cfe3f4c595b069cea", + "source_prefix": "half_cheetah/zero_latency/shared/source_100ep_v1/demonstrations/raw", + "source_sha256": { + "metadata.json": "267da8c4d6ba9f84c1f15f70efff3869ebded9141d0ab756783b181c9ba00541", + "train.parquet": "2ceaf2f1d636455b83cc14f41607c6e59a7ce7a22c7e070e6d0958e162f56783", + "val.parquet": "34442961a49b38bf296066f2925ed1300665e2585ed7a68a81f0ecc399a4a58d" + }, + "frames": { + "train": 90000, + "val": 10000 + }, + "episodes": { + "train": 90, + "val": 10 + }, + "returns": { + "train": { + "min": 3259.0414638389484, + "mean": 5647.154794339976 + }, + "val": { + "min": 5400.108074769378, + "mean": 5715.760363031365 + } + }, + "image_size": [ + 224, + 224 + ], + "state_dim": 17, + "action_dim": 6, + "action_horizon": 1, + "fps": 5, + "zero_latency": true, + "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/half_cheetah/zero_latency/train.yaml", + "smoke_config": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/half_cheetah/zero_latency/smoke.yaml", + "dataset": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/half_cheetah/zero_latency/data/lerobot/half_cheetah_h1_5fps_l0" + } + }, + "training_config_sha256": "1132a794b9e2a27c0f87a9d89a46db1b6d824707af5c8d24b4d5ca2d758fe6a9", + "dataset_manifest_sha256": "1fa7da4fe46e3e7c71374e7a75fb12ce42b9c2d4dc1cdc8c62825eb2b1185de5" +} diff --git a/zero-latency/half-cheetah/vla/starvla-qwengr00t-h1/half_cheetah_l0_gr00t_h1_5fps_2h200_20260917/task_contract.json b/zero-latency/half-cheetah/vla/starvla-qwengr00t-h1/half_cheetah_l0_gr00t_h1_5fps_2h200_20260917/task_contract.json new file mode 100644 index 0000000000000000000000000000000000000000..7ad1ce208f16919ea674c92c9a46a6709adc3f68 --- /dev/null +++ b/zero-latency/half-cheetah/vla/starvla-qwengr00t-h1/half_cheetah_l0_gr00t_h1_5fps_2h200_20260917/task_contract.json @@ -0,0 +1,73 @@ +{ + "action_space": { + "dtype": "float32", + "high": [ + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0 + ], + "labels": [ + "back_thigh_torque", + "back_shin_torque", + "back_foot_torque", + "front_thigh_torque", + "front_shin_torque", + "front_foot_torque" + ], + "low": [ + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0 + ], + "type": "box" + }, + "base_prompt": "Make the Half Cheetah run forward as fast as possible. Predict six continuous torques in [-1, 1] ordered as back thigh, back shin, back foot, front thigh, front shin, and front foot.", + "env_fps": 5, + "env_id": "LatencyBench/HalfCheetahContinuous-v0", + "frame_stack": 1, + "make_kwargs": { + "base_env_id": "HalfCheetah-v4", + "base_make_kwargs": {}, + "render_mode": "rgb_array" + }, + "noop_action": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ], + "obs_fps": 5, + "registration_imports": [ + "latency_bench.envs.gymnasium_half_cheetah" + ], + "state_space": { + "labels": [ + "front_tip_height", + "front_tip_angle", + "back_thigh_angle", + "back_shin_angle", + "back_foot_angle", + "front_thigh_angle", + "front_shin_angle", + "front_foot_angle", + "front_tip_x_velocity", + "front_tip_z_velocity", + "front_tip_angular_velocity", + "back_thigh_angular_velocity", + "back_shin_angular_velocity", + "back_foot_angular_velocity", + "front_thigh_angular_velocity", + "front_shin_angular_velocity", + "front_foot_angular_velocity" + ] + }, + "task_name": "half_cheetah_rgb_state" +} diff --git a/zero-latency/half-cheetah/vla/starvla-qwengr00t-h1/half_cheetah_l0_gr00t_h1_5fps_2h200_20260917/validation/first_completed_reload_episode.json b/zero-latency/half-cheetah/vla/starvla-qwengr00t-h1/half_cheetah_l0_gr00t_h1_5fps_2h200_20260917/validation/first_completed_reload_episode.json new file mode 100644 index 0000000000000000000000000000000000000000..1e0121300dbdf006207d843dd2a674d01d40d2b2 --- /dev/null +++ b/zero-latency/half-cheetah/vla/starvla-qwengr00t-h1/half_cheetah_l0_gr00t_h1_5fps_2h200_20260917/validation/first_completed_reload_episode.json @@ -0,0 +1,10 @@ +{ + "evidence": "existingL0sim20log", + "completed_episodes": 2, + "checkpoint": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/half_cheetah/zero_latency/bundle/checkpoints/model.pt", + "completed_episode_log_lines": [ + "[bench] sweep 1/1 episode 1/20 done", + "[bench] sweep 1/1 episode 2/20 done" + ], + "evaluation": "running; bufferedtracefilesnotyetflushed; nofinalscores" +} diff --git a/zero-latency/half-cheetah/vla/starvla-qwengr00t-h1/half_cheetah_l0_gr00t_h1_5fps_2h200_20260917/validation/formal.status.json b/zero-latency/half-cheetah/vla/starvla-qwengr00t-h1/half_cheetah_l0_gr00t_h1_5fps_2h200_20260917/validation/formal.status.json new file mode 100644 index 0000000000000000000000000000000000000000..be350f6c85db9d6d0ec540276797833a0359915a --- /dev/null +++ b/zero-latency/half-cheetah/vla/starvla-qwengr00t-h1/half_cheetah_l0_gr00t_h1_5fps_2h200_20260917/validation/formal.status.json @@ -0,0 +1,27 @@ +{ + "task": "half_cheetah", + "condition": "zero_latency", + "stage": "formal", + "state": "COMPLETED", + "started_at": "2026-09-17T12:32:42.303839+00:00", + "command": [ + "taskset", + "-c", + "0-27,112-139", + "/home/ubuntu/miniconda3/envs/starvla_rl_games_gr00t/bin/python", + "-m", + "accelerate.commands.launch", + "--config_file", + "/mnt/local/lzj/latency-sensitive-bench/code/h1-5fps-20260917-v2/third_party/starVLA/starVLA/config/deepseeds/deepspeed_zero2.yaml", + "--num_processes", + "2", + "--main_process_port", + "29782", + "starVLA/training/train_starvla.py", + "--config_yaml", + "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/half_cheetah/zero_latency/train.yaml" + ], + "supervisor_pid": 54662, + "exit_code": 0, + "ended_at": "2026-09-17T14:21:17.765655+00:00" +} diff --git a/zero-latency/half-cheetah/vla/starvla-qwengr00t-h1/half_cheetah_l0_gr00t_h1_5fps_2h200_20260917/validation/l0_sim20.yaml b/zero-latency/half-cheetah/vla/starvla-qwengr00t-h1/half_cheetah_l0_gr00t_h1_5fps_2h200_20260917/validation/l0_sim20.yaml new file mode 100644 index 0000000000000000000000000000000000000000..f3d6a9668a87be5e48504d555675d70573544d2b --- /dev/null +++ b/zero-latency/half-cheetah/vla/starvla-qwengr00t-h1/half_cheetah_l0_gr00t_h1_5fps_2h200_20260917/validation/l0_sim20.yaml @@ -0,0 +1,169 @@ +experiment: + name: gr00t_half_cheetah_zero + seed: 42 +backend: + type: sample_factory + algo: APPO + device: cuda + train_dir: /mnt/checkpoints/latency-sensitive-bench/small_models/half_cheetah + restart_behavior: overwrite + run_mode: eval +executor: + mode: simulated + inference_devices: + - cuda:0 + inference_batch_size: 1 + simulated_inference_pool: true + simulated_worker_capacity: 1 +env: + name: gymnasium + task_name: half_cheetah + env_id: LatencyBench/HalfCheetahContinuous-v0 + registration_imports: + - latency_bench.envs.gymnasium_half_cheetah + make_kwargs: + base_env_id: HalfCheetah-v4 + base_make_kwargs: {} + render_mode: rgb_array + env_fps: 5 + obs_fps: 5 + frame_stack: 1 + state_space: + labels: + - front_tip_height + - front_tip_angle + - back_thigh_angle + - back_shin_angle + - back_foot_angle + - front_thigh_angle + - front_shin_angle + - front_foot_angle + - front_tip_x_velocity + - front_tip_z_velocity + - front_tip_angular_velocity + - back_thigh_angular_velocity + - back_shin_angular_velocity + - back_foot_angular_velocity + - front_thigh_angular_velocity + - front_shin_angular_velocity + - front_foot_angular_velocity + action_space: + dtype: float32 + high: + - 1.0 + - 1.0 + - 1.0 + - 1.0 + - 1.0 + - 1.0 + labels: + - back_thigh_torque + - back_shin_torque + - back_foot_torque + - front_thigh_torque + - front_shin_torque + - front_foot_torque + low: + - -1.0 + - -1.0 + - -1.0 + - -1.0 + - -1.0 + - -1.0 + type: box + noop_action: + - 0.0 + - 0.0 + - 0.0 + - 0.0 + - 0.0 + - 0.0 + base_prompt: Make the Half Cheetah run forward as fast as possible. Predict six + continuous torques in [-1, 1] ordered as back thigh, back shin, back foot, front + thigh, front shin, and front foot. + obs_resize: + - 224 + - 224 +latency: + method: zero + sync_cuda: false +scheduler: + hold_policy: hold + ordering_policy: issue_order_fifo +policy: + type: starvla + checkpoint_path: /mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/half_cheetah/zero_latency/bundle/checkpoints/model.pt + model_config_path: /mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/half_cheetah/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/half_cheetah/zero_latency/bundle/manifest.json + latency_prompt_map_path: /mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/half_cheetah/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 + 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/half_cheetah/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/half-cheetah/vla/starvla-qwengr00t-h1/half_cheetah_l0_gr00t_h1_5fps_2h200_20260917/validation/loader_check.json b/zero-latency/half-cheetah/vla/starvla-qwengr00t-h1/half_cheetah_l0_gr00t_h1_5fps_2h200_20260917/validation/loader_check.json new file mode 100644 index 0000000000000000000000000000000000000000..11afd341aee88393d4f2400da63dca064f33889d --- /dev/null +++ b/zero-latency/half-cheetah/vla/starvla-qwengr00t-h1/half_cheetah_l0_gr00t_h1_5fps_2h200_20260917/validation/loader_check.json @@ -0,0 +1,24 @@ +{ + "train": { + "loader_frames": 90000, + "checked_indices": [ + 0, + 45000, + 89999 + ], + "max_action_float16_roundtrip_error": 0.0002390146255493164, + "state_exact_train_minmax": true, + "action_transform": "tensorize_only; no percentile normalization; decoder preserves native torque" + }, + "val": { + "loader_frames": 10000, + "checked_indices": [ + 0, + 5000, + 9999 + ], + "max_action_float16_roundtrip_error": 0.00020807981491088867, + "state_exact_train_minmax": true, + "action_transform": "tensorize_only; no percentile normalization; decoder preserves native torque" + } +} diff --git a/zero-latency/half-cheetah/vla/starvla-qwengr00t-h1/half_cheetah_l0_gr00t_h1_5fps_2h200_20260917/validation/preparation.json b/zero-latency/half-cheetah/vla/starvla-qwengr00t-h1/half_cheetah_l0_gr00t_h1_5fps_2h200_20260917/validation/preparation.json new file mode 100644 index 0000000000000000000000000000000000000000..b0af5038bc906741b2a0a3961a9caf513549bc73 --- /dev/null +++ b/zero-latency/half-cheetah/vla/starvla-qwengr00t-h1/half_cheetah_l0_gr00t_h1_5fps_2h200_20260917/validation/preparation.json @@ -0,0 +1,43 @@ +{ + "source_repo": "latency-sensitive-bench/Standard-Pipeline", + "source_revision": "571cb5801720488ed6458b9cfe3f4c595b069cea", + "source_prefix": "half_cheetah/zero_latency/shared/source_100ep_v1/demonstrations/raw", + "source_sha256": { + "metadata.json": "267da8c4d6ba9f84c1f15f70efff3869ebded9141d0ab756783b181c9ba00541", + "train.parquet": "2ceaf2f1d636455b83cc14f41607c6e59a7ce7a22c7e070e6d0958e162f56783", + "val.parquet": "34442961a49b38bf296066f2925ed1300665e2585ed7a68a81f0ecc399a4a58d" + }, + "frames": { + "train": 90000, + "val": 10000 + }, + "episodes": { + "train": 90, + "val": 10 + }, + "returns": { + "train": { + "min": 3259.0414638389484, + "mean": 5647.154794339976 + }, + "val": { + "min": 5400.108074769378, + "mean": 5715.760363031365 + } + }, + "image_size": [ + 224, + 224 + ], + "state_dim": 17, + "action_dim": 6, + "action_horizon": 1, + "fps": 5, + "zero_latency": true, + "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/half_cheetah/zero_latency/train.yaml", + "smoke_config": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/half_cheetah/zero_latency/smoke.yaml", + "dataset": "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/half_cheetah/zero_latency/data/lerobot/half_cheetah_h1_5fps_l0" +} diff --git a/zero-latency/half-cheetah/vla/starvla-qwengr00t-h1/half_cheetah_l0_gr00t_h1_5fps_2h200_20260917/validation/publication_state.json b/zero-latency/half-cheetah/vla/starvla-qwengr00t-h1/half_cheetah_l0_gr00t_h1_5fps_2h200_20260917/validation/publication_state.json new file mode 100644 index 0000000000000000000000000000000000000000..bdf57b385e33c3278f87216130cdb840b7a396ed --- /dev/null +++ b/zero-latency/half-cheetah/vla/starvla-qwengr00t-h1/half_cheetah_l0_gr00t_h1_5fps_2h200_20260917/validation/publication_state.json @@ -0,0 +1,34 @@ +{ + "task": "half_cheetah", + "verified_at": "2026-09-17T14:25:23.389086+00:00", + "training": "COMPLETED5000exit0", + "l0_eval_status_snapshot": { + "task": "half_cheetah", + "condition": "zero_latency", + "stage": "l0_sim20", + "state": "RUNNING", + "started_at": "2026-09-17T14:21:29.384520+00:00", + "command": [ + "/home/ubuntu/miniconda3/envs/starvla_rl_games_gr00t/bin/python", + "-m", + "latency_bench.run", + "--config", + "/mnt/local/lzj/latency-sensitive-bench/serial_h1_5fps_20260917/half_cheetah/zero_latency/l0_sim20.yaml" + ], + "supervisor_pid": 54662 + }, + "l0_episodes_completed_at_snapshot": 2, + "l0_final_scores": "notreportedwhile20episodeevaluationpending", + "reload_forward_evidence": "firstcompletedepisodefromexistingfinalbundleL0evaluation;noextraevaluation", + "downstream": "P/B/Fpending;parentmanaged3090queue", + "bundle_files": 8, + "bundle_sha256": { + "checkpoints/model.pt": "8f007f801c45ed8476546ebf7cc96e60283d2ffd05a6d30feaa529d7ea7f658f", + "config.full.yaml": "8f82dc4d2736cac8e19408ea47a257a465767cfe8ddb1095b4f3239799747bc0", + "config.yaml": "6f2442da8b109d628223aab3bb2ed99ed029db2b14586baeb9fc667aa6001bb1", + "dataset_statistics.json": "0e3668a95debf5f1a337b57542912ad3bf44801390d60864a13b6587d9a1bc2b", + "latency_prompt_map.json": "86b9ca38a6d19f2da16ac017aa2af280025bce688e5d02915ebe8b6a02fbf605", + "manifest.json": "1fa7da4fe46e3e7c71374e7a75fb12ce42b9c2d4dc1cdc8c62825eb2b1185de5", + "provenance.json": "f7f703c880d16353f2610c7587a7d93ff633936dc7950aa5d31f02a5066dbd9f" + } +} diff --git a/zero-latency/half-cheetah/vla/starvla-qwenoft-h1/half_cheetah_rgb_state_l0_return_gt3000_100ep_openvla_native_sft_5k/README.md b/zero-latency/half-cheetah/vla/starvla-qwenoft-h1/half_cheetah_rgb_state_l0_return_gt3000_100ep_openvla_native_sft_5k/README.md new file mode 100644 index 0000000000000000000000000000000000000000..6e683e22c94f9e98312005d227dd84a7418a71da --- /dev/null +++ b/zero-latency/half-cheetah/vla/starvla-qwenoft-h1/half_cheetah_rgb_state_l0_return_gt3000_100ep_openvla_native_sft_5k/README.md @@ -0,0 +1,30 @@ +# half-cheetah / qwenoft + +Training condition: `zero-latency`. Run: `half_cheetah_rgb_state_l0_return_gt3000_100ep_openvla_native_sft_5k`. + +Copied from [Standard-Pipeline](https://huggingface.co/latency-sensitive-bench/Standard-Pipeline/tree/2208875f92b21e3f8ec2363a30a6705dfe56cc04/half_cheetah/zero_latency/OpenVLA) at `2208875f92b21e3f8ec2363a30a6705dfe56cc04`. + +- Checkpoint: `checkpoints/model.pt` +- Selection: only published VLA checkpoint +- Checkpoint SHA256: `09d9d65f810c9c5e335e4c1f9cf555a9538efbfa60c1424b7894dbb974fda7a3` +- 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/half-cheetah/vla/starvla-qwenoft-h1/half_cheetah_rgb_state_l0_return_gt3000_100ep_openvla_native_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/half-cheetah/vla/starvla-qwenoft-h1/half_cheetah_rgb_state_l0_return_gt3000_100ep_openvla_native_sft_5k/checkpoints/model.pt b/zero-latency/half-cheetah/vla/starvla-qwenoft-h1/half_cheetah_rgb_state_l0_return_gt3000_100ep_openvla_native_sft_5k/checkpoints/model.pt new file mode 100644 index 0000000000000000000000000000000000000000..9c58e2a9e33ad25fcb5a6878edd5b50df17d0d1e --- /dev/null +++ b/zero-latency/half-cheetah/vla/starvla-qwenoft-h1/half_cheetah_rgb_state_l0_return_gt3000_100ep_openvla_native_sft_5k/checkpoints/model.pt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:09d9d65f810c9c5e335e4c1f9cf555a9538efbfa60c1424b7894dbb974fda7a3 +size 9785049835 diff --git a/zero-latency/half-cheetah/vla/starvla-qwenoft-h1/half_cheetah_rgb_state_l0_return_gt3000_100ep_openvla_native_sft_5k/config.full.yaml b/zero-latency/half-cheetah/vla/starvla-qwenoft-h1/half_cheetah_rgb_state_l0_return_gt3000_100ep_openvla_native_sft_5k/config.full.yaml new file mode 100644 index 0000000000000000000000000000000000000000..2d275b464e235ddd3428463b8bb9c149acb4ccfb --- /dev/null +++ b/zero-latency/half-cheetah/vla/starvla-qwenoft-h1/half_cheetah_rgb_state_l0_return_gt3000_100ep_openvla_native_sft_5k/config.full.yaml @@ -0,0 +1,380 @@ +framework: + name: QwenOFT + qwenvl: + base_vlm: /mnt/checkpoints/latency-sensitive-bench/openvla/Qwen3-VL-4B-Instruct + attn_implementation: flash_attention_2 + enable_gradient_checkpointing: true + action_model: + action_model_type: MLP + action_dim: 6 + action_hidden_dim: 2560 + future_action_window_size: 0 + past_action_window_size: 0 + loss_type: l1 + state_dim: 17 + action_horizon: 1 + action_env_dim: 6 + kv_memory: + enabled: false + window: 4 + rollout_len: 8 + packed_train: false + rebased_sink: true +datasets: + vla_data: + dataset_py: lerobot_datasets + include_state: true + data_root_dir: /mnt/data/latency-sensitive-bench/vla_datasets/rl_games + data_mix: half_cheetah_rgb_state_l0_return_gt3000_100ep + eval_data_mix: half_cheetah_rgb_state_l0_return_gt3000_100ep__val + custom_mixtures_path: /mnt/data/latency-sensitive-bench/vla_datasets/rl_games/_generated_mixtures/half_cheetah_rgb_state_l0_return_gt3000_100ep.json + action_type: discrete + sequential_step_sampling: false + eval_sequential_step_sampling: null + num_workers: 8 + eval_num_workers: 8 + prefetch_factor: 4 + persistent_workers: true + pin_memory: true + shuffle: true + action_balance: + enabled: false + strategy: balanced_epoch + action_key: action_id + target_flap_fraction: 0.3 + noop_id: 0 + flap_id: 1 + latency_curriculum: + enabled: false + strategy: exclusive + latencies: null + phase_steps: null + phase_distributions: null + new_latency_passes: 1.0 + replay_passes: 0.25 + target_total_passes: 2.0 + final_equalization: true + step_budget_mode: auto + eval_at_phase_end: false + save_at_phase_end: false + computed_plan: null + per_device_batch_size: 16 + load_all_data_for_training: true + num_obs_frames: 1 + image_mode: single + prompt_mode: raw + stitch_grid: + - 2 + - 2 + obs_image_size: null + video_backend: torchvision_av + gymnasium_task_contract: + action_space: + dtype: float32 + high: + - 1.0 + - 1.0 + - 1.0 + - 1.0 + - 1.0 + - 1.0 + labels: + - back_thigh_torque + - back_shin_torque + - back_foot_torque + - front_thigh_torque + - front_shin_torque + - front_foot_torque + low: + - -1.0 + - -1.0 + - -1.0 + - -1.0 + - -1.0 + - -1.0 + type: box + base_prompt: Make the Half Cheetah run forward as fast as possible. Predict + six continuous torques in [-1, 1] ordered as back thigh, back shin, back foot, + front thigh, front shin, and front foot. + env_fps: 20.0 + env_id: LatencyBench/HalfCheetahContinuous-v0 + frame_stack: 1 + make_kwargs: + base_env_id: HalfCheetah-v4 + base_make_kwargs: {} + render_mode: rgb_array + noop_action: + - 0.0 + - 0.0 + - 0.0 + - 0.0 + - 0.0 + - 0.0 + obs_fps: 20.0 + registration_imports: + - latency_bench.envs.gymnasium_half_cheetah + state_space: + labels: + - front_tip_height + - front_tip_angle + - back_thigh_angle + - back_shin_angle + - back_foot_angle + - front_thigh_angle + - front_shin_angle + - front_foot_angle + - front_tip_x_velocity + - front_tip_z_velocity + - front_tip_angular_velocity + - back_thigh_angular_velocity + - back_shin_angular_velocity + - back_foot_angular_velocity + - front_thigh_angular_velocity + - front_shin_angular_velocity + - front_foot_angular_velocity + task_name: half_cheetah_rgb_state + active_action_dim: 6 +dataset: + source_hf: '' + config_name: null + source_subdir: null + converted_name: half_cheetah_rgb_state_l0_return_gt3000_100ep + single_source_hf: '' + mixed_source_hf: '' + single_converted_name: half_cheetah_rgb_state_l0_return_gt3000_100ep + mixed_converted_name: half_cheetah_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: StarVLA/Qwen3-VL-4B-Instruct-Action +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: WORKSPACE_DIR +run_root_dir: /mnt/checkpoints/latency-sensitive-bench/openvla_runs +seed: 42 +wandb_entity: dongqianyu99-zhejiang-university +wandb_project: latency-sensitive-bench +auth: + env_file: null + hf_token_env: HF_TOKEN + wandb_api_key_env: WANDB_API_KEY +paths: + run_root_dir: /mnt/checkpoints/latency-sensitive-bench/openvla_runs + dataset_local_dir: /mnt/data/latency-sensitive-bench/vla_datasets/rl_games + dataset_cache_dir: null + base_model_dir: /mnt/checkpoints/latency-sensitive-bench/openvla/Qwen3-VL-4B-Instruct + accelerate_config: starVLA/config/deepseeds/deepspeed_zero2_cpu_offload.yaml +launch: + use_accelerate: true + gpus: null + num_processes: 1 + dry_run: false +conda: + enabled: true + env_name: null +rl_games: + model_alias: openvla + env_eval: + image_size: 224 + frameskip: 1 + image_transform: raw_rgb + prompt_mode: raw + ghost_trail: + history_frames: 5 + gamma: 1.3 + min_alpha: 35 + scroll_px_per_step: 4.0 + ground_fraction: 0.22 + seed: 42 + fixed_episode_seeds: true + latency_seed_stride: 0 + task_seed_stride: 0 + task_description: '' + eval_parallel_envs: 5 + action_chunk_execution: + enabled: false + chunk_size: null + enabled: false + gymnasium: + make_kwargs: {} + mid_train: + enabled: false + interval_steps: 100 + latencies: + - 0 + num_episodes: 5 + max_steps_per_episode: 3600 + post_train: + enabled: false + latencies: + - 0 + - 1 + - 2 + - 3 + - 4 + num_episodes: 5 + max_steps_per_episode: 3600 + eval_backend: latency_bench + distributed_mode: none + vectorized: + enabled: false + batch_size: 1 + latency: + prompt_map_path: null + mode: single + values: + - 0 + task: gymnasium + gymnasium: + task_contract: + action_space: + dtype: float32 + high: + - 1.0 + - 1.0 + - 1.0 + - 1.0 + - 1.0 + - 1.0 + labels: + - back_thigh_torque + - back_shin_torque + - back_foot_torque + - front_thigh_torque + - front_shin_torque + - front_foot_torque + low: + - -1.0 + - -1.0 + - -1.0 + - -1.0 + - -1.0 + - -1.0 + type: box + base_prompt: Make the Half Cheetah run forward as fast as possible. Predict + six continuous torques in [-1, 1] ordered as back thigh, back shin, back foot, + front thigh, front shin, and front foot. + env_fps: 20.0 + env_id: LatencyBench/HalfCheetahContinuous-v0 + frame_stack: 1 + make_kwargs: + base_env_id: HalfCheetah-v4 + render_mode: rgb_array + noop_action: + - 0.0 + - 0.0 + - 0.0 + - 0.0 + - 0.0 + - 0.0 + obs_fps: 20.0 + registration_imports: + - latency_bench.envs.gymnasium_half_cheetah + state_space: + labels: + - front_tip_height + - front_tip_angle + - back_thigh_angle + - back_shin_angle + - back_foot_angle + - front_thigh_angle + - front_shin_angle + - front_foot_angle + - front_tip_x_velocity + - front_tip_z_velocity + - front_tip_angular_velocity + - back_thigh_angular_velocity + - back_shin_angular_velocity + - back_foot_angular_velocity + - front_thigh_angular_velocity + - front_shin_angular_velocity + - front_foot_angular_velocity + task_name: half_cheetah_rgb_state + task_name: half_cheetah_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: half_cheetah_rgb_state_l0_return_gt3000_100ep_openvla_native_sft_5k +output_dir: /mnt/checkpoints/latency-sensitive-bench/openvla_runs/half_cheetah_rgb_state_l0_return_gt3000_100ep_openvla_native_sft_5k +config_yaml: null +is_debug: false +version_id: '0.21' diff --git a/zero-latency/half-cheetah/vla/starvla-qwenoft-h1/half_cheetah_rgb_state_l0_return_gt3000_100ep_openvla_native_sft_5k/config.yaml b/zero-latency/half-cheetah/vla/starvla-qwenoft-h1/half_cheetah_rgb_state_l0_return_gt3000_100ep_openvla_native_sft_5k/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..2d275b464e235ddd3428463b8bb9c149acb4ccfb --- /dev/null +++ b/zero-latency/half-cheetah/vla/starvla-qwenoft-h1/half_cheetah_rgb_state_l0_return_gt3000_100ep_openvla_native_sft_5k/config.yaml @@ -0,0 +1,380 @@ +framework: + name: QwenOFT + qwenvl: + base_vlm: /mnt/checkpoints/latency-sensitive-bench/openvla/Qwen3-VL-4B-Instruct + attn_implementation: flash_attention_2 + enable_gradient_checkpointing: true + action_model: + action_model_type: MLP + action_dim: 6 + action_hidden_dim: 2560 + future_action_window_size: 0 + past_action_window_size: 0 + loss_type: l1 + state_dim: 17 + action_horizon: 1 + action_env_dim: 6 + kv_memory: + enabled: false + window: 4 + rollout_len: 8 + packed_train: false + rebased_sink: true +datasets: + vla_data: + dataset_py: lerobot_datasets + include_state: true + data_root_dir: /mnt/data/latency-sensitive-bench/vla_datasets/rl_games + data_mix: half_cheetah_rgb_state_l0_return_gt3000_100ep + eval_data_mix: half_cheetah_rgb_state_l0_return_gt3000_100ep__val + custom_mixtures_path: /mnt/data/latency-sensitive-bench/vla_datasets/rl_games/_generated_mixtures/half_cheetah_rgb_state_l0_return_gt3000_100ep.json + action_type: discrete + sequential_step_sampling: false + eval_sequential_step_sampling: null + num_workers: 8 + eval_num_workers: 8 + prefetch_factor: 4 + persistent_workers: true + pin_memory: true + shuffle: true + action_balance: + enabled: false + strategy: balanced_epoch + action_key: action_id + target_flap_fraction: 0.3 + noop_id: 0 + flap_id: 1 + latency_curriculum: + enabled: false + strategy: exclusive + latencies: null + phase_steps: null + phase_distributions: null + new_latency_passes: 1.0 + replay_passes: 0.25 + target_total_passes: 2.0 + final_equalization: true + step_budget_mode: auto + eval_at_phase_end: false + save_at_phase_end: false + computed_plan: null + per_device_batch_size: 16 + load_all_data_for_training: true + num_obs_frames: 1 + image_mode: single + prompt_mode: raw + stitch_grid: + - 2 + - 2 + obs_image_size: null + video_backend: torchvision_av + gymnasium_task_contract: + action_space: + dtype: float32 + high: + - 1.0 + - 1.0 + - 1.0 + - 1.0 + - 1.0 + - 1.0 + labels: + - back_thigh_torque + - back_shin_torque + - back_foot_torque + - front_thigh_torque + - front_shin_torque + - front_foot_torque + low: + - -1.0 + - -1.0 + - -1.0 + - -1.0 + - -1.0 + - -1.0 + type: box + base_prompt: Make the Half Cheetah run forward as fast as possible. Predict + six continuous torques in [-1, 1] ordered as back thigh, back shin, back foot, + front thigh, front shin, and front foot. + env_fps: 20.0 + env_id: LatencyBench/HalfCheetahContinuous-v0 + frame_stack: 1 + make_kwargs: + base_env_id: HalfCheetah-v4 + base_make_kwargs: {} + render_mode: rgb_array + noop_action: + - 0.0 + - 0.0 + - 0.0 + - 0.0 + - 0.0 + - 0.0 + obs_fps: 20.0 + registration_imports: + - latency_bench.envs.gymnasium_half_cheetah + state_space: + labels: + - front_tip_height + - front_tip_angle + - back_thigh_angle + - back_shin_angle + - back_foot_angle + - front_thigh_angle + - front_shin_angle + - front_foot_angle + - front_tip_x_velocity + - front_tip_z_velocity + - front_tip_angular_velocity + - back_thigh_angular_velocity + - back_shin_angular_velocity + - back_foot_angular_velocity + - front_thigh_angular_velocity + - front_shin_angular_velocity + - front_foot_angular_velocity + task_name: half_cheetah_rgb_state + active_action_dim: 6 +dataset: + source_hf: '' + config_name: null + source_subdir: null + converted_name: half_cheetah_rgb_state_l0_return_gt3000_100ep + single_source_hf: '' + mixed_source_hf: '' + single_converted_name: half_cheetah_rgb_state_l0_return_gt3000_100ep + mixed_converted_name: half_cheetah_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: StarVLA/Qwen3-VL-4B-Instruct-Action +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: WORKSPACE_DIR +run_root_dir: /mnt/checkpoints/latency-sensitive-bench/openvla_runs +seed: 42 +wandb_entity: dongqianyu99-zhejiang-university +wandb_project: latency-sensitive-bench +auth: + env_file: null + hf_token_env: HF_TOKEN + wandb_api_key_env: WANDB_API_KEY +paths: + run_root_dir: /mnt/checkpoints/latency-sensitive-bench/openvla_runs + dataset_local_dir: /mnt/data/latency-sensitive-bench/vla_datasets/rl_games + dataset_cache_dir: null + base_model_dir: /mnt/checkpoints/latency-sensitive-bench/openvla/Qwen3-VL-4B-Instruct + accelerate_config: starVLA/config/deepseeds/deepspeed_zero2_cpu_offload.yaml +launch: + use_accelerate: true + gpus: null + num_processes: 1 + dry_run: false +conda: + enabled: true + env_name: null +rl_games: + model_alias: openvla + env_eval: + image_size: 224 + frameskip: 1 + image_transform: raw_rgb + prompt_mode: raw + ghost_trail: + history_frames: 5 + gamma: 1.3 + min_alpha: 35 + scroll_px_per_step: 4.0 + ground_fraction: 0.22 + seed: 42 + fixed_episode_seeds: true + latency_seed_stride: 0 + task_seed_stride: 0 + task_description: '' + eval_parallel_envs: 5 + action_chunk_execution: + enabled: false + chunk_size: null + enabled: false + gymnasium: + make_kwargs: {} + mid_train: + enabled: false + interval_steps: 100 + latencies: + - 0 + num_episodes: 5 + max_steps_per_episode: 3600 + post_train: + enabled: false + latencies: + - 0 + - 1 + - 2 + - 3 + - 4 + num_episodes: 5 + max_steps_per_episode: 3600 + eval_backend: latency_bench + distributed_mode: none + vectorized: + enabled: false + batch_size: 1 + latency: + prompt_map_path: null + mode: single + values: + - 0 + task: gymnasium + gymnasium: + task_contract: + action_space: + dtype: float32 + high: + - 1.0 + - 1.0 + - 1.0 + - 1.0 + - 1.0 + - 1.0 + labels: + - back_thigh_torque + - back_shin_torque + - back_foot_torque + - front_thigh_torque + - front_shin_torque + - front_foot_torque + low: + - -1.0 + - -1.0 + - -1.0 + - -1.0 + - -1.0 + - -1.0 + type: box + base_prompt: Make the Half Cheetah run forward as fast as possible. Predict + six continuous torques in [-1, 1] ordered as back thigh, back shin, back foot, + front thigh, front shin, and front foot. + env_fps: 20.0 + env_id: LatencyBench/HalfCheetahContinuous-v0 + frame_stack: 1 + make_kwargs: + base_env_id: HalfCheetah-v4 + render_mode: rgb_array + noop_action: + - 0.0 + - 0.0 + - 0.0 + - 0.0 + - 0.0 + - 0.0 + obs_fps: 20.0 + registration_imports: + - latency_bench.envs.gymnasium_half_cheetah + state_space: + labels: + - front_tip_height + - front_tip_angle + - back_thigh_angle + - back_shin_angle + - back_foot_angle + - front_thigh_angle + - front_shin_angle + - front_foot_angle + - front_tip_x_velocity + - front_tip_z_velocity + - front_tip_angular_velocity + - back_thigh_angular_velocity + - back_shin_angular_velocity + - back_foot_angular_velocity + - front_thigh_angular_velocity + - front_shin_angular_velocity + - front_foot_angular_velocity + task_name: half_cheetah_rgb_state + task_name: half_cheetah_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: half_cheetah_rgb_state_l0_return_gt3000_100ep_openvla_native_sft_5k +output_dir: /mnt/checkpoints/latency-sensitive-bench/openvla_runs/half_cheetah_rgb_state_l0_return_gt3000_100ep_openvla_native_sft_5k +config_yaml: null +is_debug: false +version_id: '0.21' diff --git a/zero-latency/half-cheetah/vla/starvla-qwenoft-h1/half_cheetah_rgb_state_l0_return_gt3000_100ep_openvla_native_sft_5k/dataset_statistics.json b/zero-latency/half-cheetah/vla/starvla-qwenoft-h1/half_cheetah_rgb_state_l0_return_gt3000_100ep_openvla_native_sft_5k/dataset_statistics.json new file mode 100644 index 0000000000000000000000000000000000000000..e0a2563a2c7d2978230a2ba367603622a32b0d04 --- /dev/null +++ b/zero-latency/half-cheetah/vla/starvla-qwenoft-h1/half_cheetah_rgb_state_l0_return_gt3000_100ep_openvla_native_sft_5k/dataset_statistics.json @@ -0,0 +1,180 @@ +{ + "new_embodiment": { + "action": { + "mean": [ + 0.04852694645524025, + -0.07289591431617737, + -0.16553527116775513, + 0.3926909863948822, + 0.4024640917778015, + -0.15250922739505768 + ], + "std": [ + 0.9902900457382202, + 0.9729189872741699, + 0.92887282371521, + 0.7155725359916687, + 0.5879771113395691, + 0.8828322291374207 + ], + "max": [ + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0 + ], + "min": [ + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0 + ], + "q01": [ + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0 + ], + "q99": [ + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0 + ], + "mask": [ + true, + true, + true, + true, + true, + true + ] + }, + "state": { + "mean": [ + -0.16455742716789246, + -0.6733709573745728, + -0.07721110433340073, + -0.1043669581413269, + -0.20796853303909302, + 0.531654417514801, + 0.23800496757030487, + -0.08999839425086975, + 0.4533954858779907, + 0.03410167247056961, + -0.009492190554738045, + 0.04943292587995529, + -0.05999681353569031, + 0.11326351761817932, + -0.022488147020339966, + -0.027951883152127266, + -0.10890804976224899 + ], + "std": [ + 0.1308881938457489, + 0.11011626571416855, + 0.5556989908218384, + 0.4154678285121918, + 0.592194676399231, + 0.2803356647491455, + 0.41065067052841187, + 0.5386135578155518, + 0.2725602984428406, + 0.18254218995571136, + 0.26073697209358215, + 0.6522123217582703, + 0.29445716738700867, + 0.5733875632286072, + 0.2830783426761627, + 0.4337209165096283, + 0.5481678247451782 + ], + "max": [ + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0 + ], + "min": [ + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0 + ], + "q01": [ + -0.8618907463550568, + -0.7782192385196686, + -0.963983958363533, + -0.7420942264795304, + -0.9641105020046234, + -0.15345445454120635, + -0.5540750181674957, + -0.9644814884662628, + -0.5328765374422073, + -0.4222735470533371, + -0.5928484404087067, + -0.8578835678100586, + -0.7154878216981888, + -0.918103654384613, + -0.5789287960529328, + -0.8079770970344543, + -0.8612057662010193 + ], + "q99": [ + 0.11475664258003207, + 0.09281709313392594, + 0.7771777260303496, + 0.785496914386749, + 0.8948100674152373, + 0.9008247101306913, + 0.8394721925258635, + 0.8582286417484283, + 0.8433885538578033, + 0.42764286875724783, + 0.4806103372573851, + 0.8169456958770751, + 0.4850697159767148, + 0.9674746108055114, + 0.5811240792274471, + 0.7913007164001464, + 0.7998981952667232 + ] + }, + "num_transitions": 90000, + "num_trajectories": 90 + } +} \ No newline at end of file diff --git a/zero-latency/half-cheetah/vla/starvla-qwenoft-h1/half_cheetah_rgb_state_l0_return_gt3000_100ep_openvla_native_sft_5k/dataset_statistics_eval.json b/zero-latency/half-cheetah/vla/starvla-qwenoft-h1/half_cheetah_rgb_state_l0_return_gt3000_100ep_openvla_native_sft_5k/dataset_statistics_eval.json new file mode 100644 index 0000000000000000000000000000000000000000..89bd13b71199ed1032ddc6070d43436150df8205 --- /dev/null +++ b/zero-latency/half-cheetah/vla/starvla-qwenoft-h1/half_cheetah_rgb_state_l0_return_gt3000_100ep_openvla_native_sft_5k/dataset_statistics_eval.json @@ -0,0 +1,180 @@ +{ + "new_embodiment": { + "action": { + "mean": [ + 0.05105651915073395, + -0.07209897041320801, + -0.17561988532543182, + 0.4051264524459839, + 0.41413530707359314, + -0.16462549567222595 + ], + "std": [ + 0.9923218488693237, + 0.9729214906692505, + 0.9286880493164062, + 0.7091745138168335, + 0.5825477242469788, + 0.879936695098877 + ], + "max": [ + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0 + ], + "min": [ + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0 + ], + "q01": [ + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0 + ], + "q99": [ + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0 + ], + "mask": [ + true, + true, + true, + true, + true, + true + ] + }, + "state": { + "mean": [ + -0.15801525115966797, + -0.6853702068328857, + -0.07619248330593109, + -0.10509598255157471, + -0.21412871778011322, + 0.5437030792236328, + 0.246285542845726, + -0.09455222636461258, + 0.46554479002952576, + 0.03411850705742836, + -0.010364393703639507, + 0.04909932240843773, + -0.05942404642701149, + 0.11375650763511658, + -0.02154252491891384, + -0.02781635709106922, + -0.10924755781888962 + ], + "std": [ + 0.0925731435418129, + 0.05881912261247635, + 0.5566807389259338, + 0.4135461151599884, + 0.5901249051094055, + 0.26190266013145447, + 0.4085503816604614, + 0.5401039719581604, + 0.25604307651519775, + 0.17855967581272125, + 0.2590049207210541, + 0.6530601978302002, + 0.2929076850414276, + 0.5726403594017029, + 0.2790679931640625, + 0.4333547353744507, + 0.5537501573562622 + ], + "max": [ + 0.5848677158355713, + -0.412548303604126, + 0.8143953084945679, + 0.9106905460357666, + 0.9493037462234497, + 0.9966001510620117, + 1.0, + 0.9973659515380859, + 0.9224475622177124, + 0.8370530605316162, + 0.6996456384658813, + 0.8615819215774536, + 0.699871301651001, + 0.9964103698730469, + 0.8931897878646851, + 0.9465029239654541, + 0.9760327339172363 + ], + "min": [ + -0.46864038705825806, + -0.9890411496162415, + -1.0, + -0.8720242977142334, + -0.9908613562583923, + -0.4105069637298584, + -0.8329423069953918, + -0.9968000054359436, + -0.8470599055290222, + -0.9176910519599915, + -0.8763898611068726, + -0.9665197730064392, + -0.8229536414146423, + -0.9776057600975037, + -0.8349185585975647, + -0.8882914185523987, + -0.9202964305877686 + ], + "q01": [ + -0.31576364040374755, + -0.7773451161384582, + -0.9628803277015686, + -0.7370971035957337, + -0.9634619683027268, + 0.012047727108001713, + -0.5325007325410843, + -0.9664082968235016, + -0.43176035225391385, + -0.4093106472492218, + -0.5883447122573853, + -0.8565261036157608, + -0.7075943994522095, + -0.917148374915123, + -0.5765699028968811, + -0.804426833987236, + -0.8601254969835281 + ], + "q99": [ + 0.09756481409072877, + -0.506605694890022, + 0.7765427339076996, + 0.7838108253479005, + 0.8938083624839783, + 0.9039195466041565, + 0.8389676129817963, + 0.860252126455307, + 0.8462485337257388, + 0.41060072302818307, + 0.4733871126174928, + 0.8177655553817749, + 0.4855780243873601, + 0.9625551664829256, + 0.5755143201351167, + 0.7854277300834656, + 0.8012957441806796 + ] + }, + "num_transitions": 10000, + "num_trajectories": 10 + } +} \ No newline at end of file diff --git a/zero-latency/half-cheetah/vla/starvla-qwenoft-h1/half_cheetah_rgb_state_l0_return_gt3000_100ep_openvla_native_sft_5k/provenance.json b/zero-latency/half-cheetah/vla/starvla-qwenoft-h1/half_cheetah_rgb_state_l0_return_gt3000_100ep_openvla_native_sft_5k/provenance.json new file mode 100644 index 0000000000000000000000000000000000000000..c84a74814df14135c37389751167a294dd741533 --- /dev/null +++ b/zero-latency/half-cheetah/vla/starvla-qwenoft-h1/half_cheetah_rgb_state_l0_return_gt3000_100ep_openvla_native_sft_5k/provenance.json @@ -0,0 +1,37 @@ +{ + "task": "half-cheetah", + "model": "qwenoft", + "training_condition": "zero-latency", + "training_run_id": "half_cheetah_rgb_state_l0_return_gt3000_100ep_openvla_native_sft_5k", + "source": { + "repo_id": "latency-sensitive-bench/Standard-Pipeline", + "revision": "2208875f92b21e3f8ec2363a30a6705dfe56cc04", + "path": "half_cheetah/zero_latency/OpenVLA", + "url": "https://huggingface.co/latency-sensitive-bench/Standard-Pipeline/tree/2208875f92b21e3f8ec2363a30a6705dfe56cc04/half_cheetah/zero_latency/OpenVLA" + }, + "hf_repo_id": "latency-sensitive-bench/benchmark-models", + "hf_path": "zero-latency/half-cheetah/vla/starvla-qwenoft-h1/half_cheetah_rgb_state_l0_return_gt3000_100ep_openvla_native_sft_5k", + "checkpoint": { + "source_file": "half_cheetah/zero_latency/OpenVLA/checkpoints/steps_5000_pytorch_model.pt", + "source_sha256": "09d9d65f810c9c5e335e4c1f9cf555a9538efbfa60c1424b7894dbb974fda7a3", + "source_bytes": 9785049835, + "file": "checkpoints/model.pt", + "selection_rule": "only published VLA checkpoint", + "method": "copy", + "sha256": "09d9d65f810c9c5e335e4c1f9cf555a9538efbfa60c1424b7894dbb974fda7a3", + "bytes": 9785049835 + }, + "config_source": "half_cheetah/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/half-cheetah/vla/starvla-qwenoft-h1/half_cheetah_rgb_state_l0_return_gt3000_100ep_openvla_native_sft_5k/source/config.yaml b/zero-latency/half-cheetah/vla/starvla-qwenoft-h1/half_cheetah_rgb_state_l0_return_gt3000_100ep_openvla_native_sft_5k/source/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..ec5719021848262d3b5cceba038c4662905baaa3 --- /dev/null +++ b/zero-latency/half-cheetah/vla/starvla-qwenoft-h1/half_cheetah_rgb_state_l0_return_gt3000_100ep_openvla_native_sft_5k/source/config.yaml @@ -0,0 +1,84 @@ +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: half_cheetah_rgb_state_l0_return_gt3000_100ep + dataset_py: lerobot_datasets + eval_data_mix: half_cheetah_rgb_state_l0_return_gt3000_100ep__val + latency_curriculum: + enabled: false + per_device_batch_size: 16 +framework: + action_model: + action_dim: 6 + action_env_dim: 6 + action_hidden_dim: 2560 + action_horizon: 1 + action_model_type: MLP + loss_type: l1 + 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/half_cheetah_rgb_state_l0_return_gt3000_100ep_openvla_native_sft_5k +rl_games: + env_eval: + enabled: false + task: gymnasium +run_id: half_cheetah_rgb_state_l0_return_gt3000_100ep_openvla_native_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/half-cheetah/vla/starvla-qwenoft-h1/half_cheetah_rgb_state_l0_return_gt3000_100ep_openvla_native_sft_5k/task_contract.json b/zero-latency/half-cheetah/vla/starvla-qwenoft-h1/half_cheetah_rgb_state_l0_return_gt3000_100ep_openvla_native_sft_5k/task_contract.json new file mode 100644 index 0000000000000000000000000000000000000000..e8f8336a7c71e2e1bea35c32f0829cac90a6143e --- /dev/null +++ b/zero-latency/half-cheetah/vla/starvla-qwenoft-h1/half_cheetah_rgb_state_l0_return_gt3000_100ep_openvla_native_sft_5k/task_contract.json @@ -0,0 +1,73 @@ +{ + "action_space": { + "dtype": "float32", + "high": [ + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0 + ], + "labels": [ + "back_thigh_torque", + "back_shin_torque", + "back_foot_torque", + "front_thigh_torque", + "front_shin_torque", + "front_foot_torque" + ], + "low": [ + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0 + ], + "type": "box" + }, + "base_prompt": "Make the Half Cheetah run forward as fast as possible. Predict six continuous torques in [-1, 1] ordered as back thigh, back shin, back foot, front thigh, front shin, and front foot.", + "env_fps": 20.0, + "env_id": "LatencyBench/HalfCheetahContinuous-v0", + "frame_stack": 1, + "make_kwargs": { + "base_env_id": "HalfCheetah-v4", + "base_make_kwargs": {}, + "render_mode": "rgb_array" + }, + "noop_action": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ], + "obs_fps": 20.0, + "registration_imports": [ + "latency_bench.envs.gymnasium_half_cheetah" + ], + "state_space": { + "labels": [ + "front_tip_height", + "front_tip_angle", + "back_thigh_angle", + "back_shin_angle", + "back_foot_angle", + "front_thigh_angle", + "front_shin_angle", + "front_foot_angle", + "front_tip_x_velocity", + "front_tip_z_velocity", + "front_tip_angular_velocity", + "back_thigh_angular_velocity", + "back_shin_angular_velocity", + "back_foot_angular_velocity", + "front_thigh_angular_velocity", + "front_shin_angular_velocity", + "front_foot_angular_velocity" + ] + }, + "task_name": "half_cheetah_rgb_state" +} diff --git a/zero-latency/half-cheetah/vla/starvla-qwenpi_v3-h1/half_cheetah_pi05_l0_h1_5fps_g128_20260921/README.md b/zero-latency/half-cheetah/vla/starvla-qwenpi_v3-h1/half_cheetah_pi05_l0_h1_5fps_g128_20260921/README.md new file mode 100644 index 0000000000000000000000000000000000000000..dc6dd721aaa7d60dd17013f5c32045ba488adb48 --- /dev/null +++ b/zero-latency/half-cheetah/vla/starvla-qwenpi_v3-h1/half_cheetah_pi05_l0_h1_5fps_g128_20260921/README.md @@ -0,0 +1,31 @@ +# half-cheetah / qwenpi_v3 + +Training condition: `zero-latency`. Run: `half_cheetah_pi05_l0_h1_5fps_g128_20260921`. + +Copied from [Standard-Pipeline](https://huggingface.co/latency-sensitive-bench/Standard-Pipeline/tree/2208875f92b21e3f8ec2363a30a6705dfe56cc04/half_cheetah/zero_latency/Pi05) at `2208875f92b21e3f8ec2363a30a6705dfe56cc04`. + +- Checkpoint: `checkpoints/model.pt` +- Selection: only published VLA checkpoint +- Checkpoint SHA256: `58a8cf13cd5cae8071c7da2085410c1a039eaab255486c3814308bbaf561c2c4` +- 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/half-cheetah/vla/starvla-qwenpi_v3-h1/half_cheetah_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/half-cheetah/vla/starvla-qwenpi_v3-h1/half_cheetah_pi05_l0_h1_5fps_g128_20260921/checkpoints/model.pt b/zero-latency/half-cheetah/vla/starvla-qwenpi_v3-h1/half_cheetah_pi05_l0_h1_5fps_g128_20260921/checkpoints/model.pt new file mode 100644 index 0000000000000000000000000000000000000000..965f2c12613cbf3ab1aa926a2efb52ad337b2293 --- /dev/null +++ b/zero-latency/half-cheetah/vla/starvla-qwenpi_v3-h1/half_cheetah_pi05_l0_h1_5fps_g128_20260921/checkpoints/model.pt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:58a8cf13cd5cae8071c7da2085410c1a039eaab255486c3814308bbaf561c2c4 +size 10922653853 diff --git a/zero-latency/half-cheetah/vla/starvla-qwenpi_v3-h1/half_cheetah_pi05_l0_h1_5fps_g128_20260921/config.full.yaml b/zero-latency/half-cheetah/vla/starvla-qwenpi_v3-h1/half_cheetah_pi05_l0_h1_5fps_g128_20260921/config.full.yaml new file mode 100644 index 0000000000000000000000000000000000000000..287ba39ea412a68f012d3b9e1f945401cd562a8d --- /dev/null +++ b/zero-latency/half-cheetah/vla/starvla-qwenpi_v3-h1/half_cheetah_pi05_l0_h1_5fps_g128_20260921/config.full.yaml @@ -0,0 +1,259 @@ +framework: + name: QwenPI_v3 + qwenvl: + base_vlm: ${oc.env:PI05_BACKBONE_DIR} + attn_implementation: flash_attention_2 + vl_hidden_dim: 2560 + num_vl_layers: 36 + enable_gradient_checkpointing: false + action_model: + action_model_type: LayerwiseFM + action_dim: 6 + state_dim: 17 + state_encoding: continuous_projector + task_objective: null + action_horizon: 1 + repeated_diffusion_steps: 2 + num_inference_timesteps: 4 + add_pos_embed: true + max_seq_len: 1024 + num_target_vision_tokens: 32 + noise_beta_alpha: 1.5 + noise_beta_beta: 1.0 + noise_s: 0.999 + num_timestep_buckets: 1000 + diffusion_model_cfg: + action_dit_hidden_dim: 1024 + dropout: 0.2 + final_dropout: true + interleave_self_attention: true + norm_type: ada_norm + positional_embeddings: null + attention_head_dim: 64 + num_layers: 36 + input_embedding_dim: 1024 + cross_attention_dim: 1024 + output_dim: 1024 + num_attention_heads: 16 + action_env_dim: 6 + future_action_window_size: 0 + past_action_window_size: 0 +datasets: + vla_data: + dataset_py: lerobot_datasets + include_state: true + data_root_dir: / + data_mix: train + eval_data_mix: validation + custom_mixtures_path: ${PI05_RUN_DIR}/zero_latency/vla/mixture.json + action_type: continuous + sequential_step_sampling: true + eval_sequential_step_sampling: null + num_workers: 4 + eval_num_workers: 0 + prefetch_factor: 2 + persistent_workers: true + pin_memory: true + shuffle: true + action_balance: + enabled: false + strategy: balanced_epoch + action_key: action_id + target_flap_fraction: 0.3 + noop_id: 0 + flap_id: 1 + latency_curriculum: + enabled: false + strategy: exclusive + latencies: null + phase_steps: null + phase_distributions: null + new_latency_passes: 1.0 + replay_passes: 0.25 + target_total_passes: 2.0 + final_equalization: true + step_budget_mode: auto + eval_at_phase_end: false + save_at_phase_end: false + computed_plan: null + per_device_batch_size: 64 + load_all_data_for_training: true + num_obs_frames: 1 + image_mode: single + prompt_mode: raw + stitch_grid: + - 2 + - 2 + obs_image_size: + - 224 + - 224 + video_backend: pyav + lerobot_version: v2.0 + gymnasium_task_contract: + action_space: + dtype: float32 + high: + - 1.0 + - 1.0 + - 1.0 + - 1.0 + - 1.0 + - 1.0 + labels: + - back_thigh_torque + - back_shin_torque + - back_foot_torque + - front_thigh_torque + - front_shin_torque + - front_foot_torque + low: + - -1.0 + - -1.0 + - -1.0 + - -1.0 + - -1.0 + - -1.0 + type: box + base_prompt: Make the Half Cheetah run forward as fast as possible. Predict + six continuous torques in [-1, 1] ordered as back thigh, back shin, back foot, + front thigh, front shin, and front foot. + env_fps: 5 + env_id: LatencyBench/HalfCheetahContinuous-v0 + frame_stack: 1 + make_kwargs: + base_env_id: HalfCheetah-v4 + base_make_kwargs: {} + render_mode: rgb_array + noop_action: + - 0.0 + - 0.0 + - 0.0 + - 0.0 + - 0.0 + - 0.0 + obs_fps: 5 + registration_imports: + - latency_bench.envs.gymnasium_half_cheetah + state_space: + labels: + - front_tip_height + - front_tip_angle + - back_thigh_angle + - back_shin_angle + - back_foot_angle + - front_thigh_angle + - front_shin_angle + - front_foot_angle + - front_tip_x_velocity + - front_tip_z_velocity + - front_tip_angular_velocity + - back_thigh_angular_velocity + - back_shin_angular_velocity + - back_foot_angular_velocity + - front_thigh_angular_velocity + - front_shin_angular_velocity + - front_foot_angular_velocity + task_name: half_cheetah_rgb_state + active_action_dim: 6 +dataset: + source_hf: '' + config_name: null + source_subdir: null + converted_name: flappy_train + single_source_hf: '' + mixed_source_hf: '' + single_converted_name: flappy_train + mixed_converted_name: flappy_mixed_latency_train + single_latency_filter: null + mixed_latency_filter: null + force_download: false + setup_force: false + skip_verification: false + target_latency_unit: observation_steps + verify_rows: 200 + max_episodes: null + episodes_per_latency: null + latency_filter: null + debug_subset: + enabled: false + max_episodes: 5 + suffix: debug +base_model: + repo_id: Qwen/Qwen3-VL-4B-Instruct +initialization: + checkpoint_local_dir: null + checkpoint_hf_repo_id: null + checkpoint_filename: null +trainer: + max_train_steps: 5000 + num_warmup_steps: 100 + save_interval: 500 + eval_interval: 500 + eval_num_batches: 200 + per_latency_eval_num_batches: null + eval_action_classification: false + eval_action_classification_interval: null + cc_f1_tolerance: 1 + learning_rate: + base: 2.0e-05 + qwen_vl_interface: 1.0e-05 + action_model: 0.0001 + lr_scheduler_type: cosine_with_min_lr + scheduler_specific_kwargs: + min_lr: 1.0e-06 + freeze_modules: '' + freeze_vit: false + freeze_tied_embedding: false + freeze_llm_layers: [] + loss_scale: + vla: 1.0 + vlm: 0.1 + max_grad_norm: 1.0 + weight_decay: 0.0 + logging_frequency: 1 + profile_timing: + enabled: false + log_interval: 10 + gradient_clipping: 1.0 + gradient_accumulation_steps: 1 + distributed_backend: deepspeed + is_resume: false + pretrained_checkpoint: null + resume_step: 0 + reload_modules: null + optimizer: + name: AdamW + betas: + - 0.9 + - 0.95 + eps: 1.0e-08 + weight_decay: 1.0e-08 + fused: true + save_format: pt +run_id: half_cheetah_pi05_l0_h1_5fps_g128_20260921 +run_root_dir: ${PI05_RUN_DIR}/zero_latency/vla/training +seed: 42 +is_debug: false +version_id: '0.21' +wandb_project: latency-sensitive-bench +wandb_entity: dongqianyu99-zhejiang-university +wandb_name: half_cheetah_pi05_l0_h1_5fps_g128_20260921 +wandb_group: pi05-seven-env +wandb_tags: +- half_cheetah +- zero_latency +- Pi05 +- h1 +training_latency_condition: zero_latency +backbone: + repo_id: Qwen/Qwen3-VL-4B-Instruct + revision: ebb281ec70b05090aa6165b016eac8ec08e71b17 +checkpoint: + save_best_model: false + save_final_model: true + save_pt_file: true + save_training_state: true + local: + keep_last_n: 1 +config_yaml: ${PI05_RUN_DIR}/zero_latency/train.yaml +output_dir: ${PI05_RUN_DIR}/zero_latency/vla/training/half_cheetah_pi05_l0_h1_5fps_g128_20260921 diff --git a/zero-latency/half-cheetah/vla/starvla-qwenpi_v3-h1/half_cheetah_pi05_l0_h1_5fps_g128_20260921/config.yaml b/zero-latency/half-cheetah/vla/starvla-qwenpi_v3-h1/half_cheetah_pi05_l0_h1_5fps_g128_20260921/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..8848283bb219465082b53735a5b47e8cd220201c --- /dev/null +++ b/zero-latency/half-cheetah/vla/starvla-qwenpi_v3-h1/half_cheetah_pi05_l0_h1_5fps_g128_20260921/config.yaml @@ -0,0 +1,136 @@ +framework: + name: QwenPI_v3 + qwenvl: + base_vlm: ${oc.env:PI05_BACKBONE_DIR} + attn_implementation: flash_attention_2 + vl_hidden_dim: 2560 + num_vl_layers: 36 + enable_gradient_checkpointing: false + action_model: + action_model_type: LayerwiseFM + action_dim: 6 + state_dim: 17 + state_encoding: continuous_projector + task_objective: null + action_horizon: 1 + repeated_diffusion_steps: 2 + num_inference_timesteps: 4 + add_pos_embed: true + max_seq_len: 1024 + num_target_vision_tokens: 32 + noise_beta_alpha: 1.5 + noise_beta_beta: 1.0 + noise_s: 0.999 + num_timestep_buckets: 1000 + diffusion_model_cfg: + action_dit_hidden_dim: 1024 + dropout: 0.2 + final_dropout: true + interleave_self_attention: true + norm_type: ada_norm + positional_embeddings: null + attention_head_dim: 64 + num_layers: 36 + input_embedding_dim: 1024 + cross_attention_dim: 1024 + output_dim: 1024 + num_attention_heads: 16 + action_env_dim: 6 + future_action_window_size: 0 + past_action_window_size: 0 +version_id: '0.21' +datasets: + vla_data: + include_state: true + obs_image_size: + - 224 + - 224 + image_mode: single + stitch_grid: + - 2 + - 2 + num_obs_frames: 1 + gymnasium_task_contract: &id001 + action_space: + dtype: float32 + high: + - 1.0 + - 1.0 + - 1.0 + - 1.0 + - 1.0 + - 1.0 + labels: + - back_thigh_torque + - back_shin_torque + - back_foot_torque + - front_thigh_torque + - front_shin_torque + - front_foot_torque + low: + - -1.0 + - -1.0 + - -1.0 + - -1.0 + - -1.0 + - -1.0 + type: box + base_prompt: Make the Half Cheetah run forward as fast as possible. Predict + six continuous torques in [-1, 1] ordered as back thigh, back shin, back foot, + front thigh, front shin, and front foot. + env_fps: 5 + env_id: LatencyBench/HalfCheetahContinuous-v0 + frame_stack: 1 + make_kwargs: + base_env_id: HalfCheetah-v4 + base_make_kwargs: {} + render_mode: rgb_array + noop_action: + - 0.0 + - 0.0 + - 0.0 + - 0.0 + - 0.0 + - 0.0 + obs_fps: 5 + registration_imports: + - latency_bench.envs.gymnasium_half_cheetah + state_space: + labels: + - front_tip_height + - front_tip_angle + - back_thigh_angle + - back_shin_angle + - back_foot_angle + - front_thigh_angle + - front_shin_angle + - front_foot_angle + - front_tip_x_velocity + - front_tip_z_velocity + - front_tip_angular_velocity + - back_thigh_angular_velocity + - back_shin_angular_velocity + - back_foot_angular_velocity + - front_thigh_angular_velocity + - front_shin_angular_velocity + - front_foot_angular_velocity + task_name: half_cheetah_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/half-cheetah/vla/starvla-qwenpi_v3-h1/half_cheetah_pi05_l0_h1_5fps_g128_20260921/dataset_statistics.json b/zero-latency/half-cheetah/vla/starvla-qwenpi_v3-h1/half_cheetah_pi05_l0_h1_5fps_g128_20260921/dataset_statistics.json new file mode 100644 index 0000000000000000000000000000000000000000..e0a2563a2c7d2978230a2ba367603622a32b0d04 --- /dev/null +++ b/zero-latency/half-cheetah/vla/starvla-qwenpi_v3-h1/half_cheetah_pi05_l0_h1_5fps_g128_20260921/dataset_statistics.json @@ -0,0 +1,180 @@ +{ + "new_embodiment": { + "action": { + "mean": [ + 0.04852694645524025, + -0.07289591431617737, + -0.16553527116775513, + 0.3926909863948822, + 0.4024640917778015, + -0.15250922739505768 + ], + "std": [ + 0.9902900457382202, + 0.9729189872741699, + 0.92887282371521, + 0.7155725359916687, + 0.5879771113395691, + 0.8828322291374207 + ], + "max": [ + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0 + ], + "min": [ + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0 + ], + "q01": [ + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0 + ], + "q99": [ + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0 + ], + "mask": [ + true, + true, + true, + true, + true, + true + ] + }, + "state": { + "mean": [ + -0.16455742716789246, + -0.6733709573745728, + -0.07721110433340073, + -0.1043669581413269, + -0.20796853303909302, + 0.531654417514801, + 0.23800496757030487, + -0.08999839425086975, + 0.4533954858779907, + 0.03410167247056961, + -0.009492190554738045, + 0.04943292587995529, + -0.05999681353569031, + 0.11326351761817932, + -0.022488147020339966, + -0.027951883152127266, + -0.10890804976224899 + ], + "std": [ + 0.1308881938457489, + 0.11011626571416855, + 0.5556989908218384, + 0.4154678285121918, + 0.592194676399231, + 0.2803356647491455, + 0.41065067052841187, + 0.5386135578155518, + 0.2725602984428406, + 0.18254218995571136, + 0.26073697209358215, + 0.6522123217582703, + 0.29445716738700867, + 0.5733875632286072, + 0.2830783426761627, + 0.4337209165096283, + 0.5481678247451782 + ], + "max": [ + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0 + ], + "min": [ + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0 + ], + "q01": [ + -0.8618907463550568, + -0.7782192385196686, + -0.963983958363533, + -0.7420942264795304, + -0.9641105020046234, + -0.15345445454120635, + -0.5540750181674957, + -0.9644814884662628, + -0.5328765374422073, + -0.4222735470533371, + -0.5928484404087067, + -0.8578835678100586, + -0.7154878216981888, + -0.918103654384613, + -0.5789287960529328, + -0.8079770970344543, + -0.8612057662010193 + ], + "q99": [ + 0.11475664258003207, + 0.09281709313392594, + 0.7771777260303496, + 0.785496914386749, + 0.8948100674152373, + 0.9008247101306913, + 0.8394721925258635, + 0.8582286417484283, + 0.8433885538578033, + 0.42764286875724783, + 0.4806103372573851, + 0.8169456958770751, + 0.4850697159767148, + 0.9674746108055114, + 0.5811240792274471, + 0.7913007164001464, + 0.7998981952667232 + ] + }, + "num_transitions": 90000, + "num_trajectories": 90 + } +} \ No newline at end of file diff --git a/zero-latency/half-cheetah/vla/starvla-qwenpi_v3-h1/half_cheetah_pi05_l0_h1_5fps_g128_20260921/latency_prompt_map.json b/zero-latency/half-cheetah/vla/starvla-qwenpi_v3-h1/half_cheetah_pi05_l0_h1_5fps_g128_20260921/latency_prompt_map.json new file mode 100644 index 0000000000000000000000000000000000000000..7c2db5bd52d0a369ac248a1189b89958c14f3dd7 --- /dev/null +++ b/zero-latency/half-cheetah/vla/starvla-qwenpi_v3-h1/half_cheetah_pi05_l0_h1_5fps_g128_20260921/latency_prompt_map.json @@ -0,0 +1,7 @@ +{ + "0": { + "prompt": "Make the Half Cheetah run forward as fast as possible. Predict six continuous torques in [-1, 1] ordered as back thigh, back shin, back foot, front thigh, front shin, and front 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/half-cheetah/vla/starvla-qwenpi_v3-h1/half_cheetah_pi05_l0_h1_5fps_g128_20260921/manifest.json b/zero-latency/half-cheetah/vla/starvla-qwenpi_v3-h1/half_cheetah_pi05_l0_h1_5fps_g128_20260921/manifest.json new file mode 100644 index 0000000000000000000000000000000000000000..2c8e155f57df654227d0ba167ee641dbfc9188f7 --- /dev/null +++ b/zero-latency/half-cheetah/vla/starvla-qwenpi_v3-h1/half_cheetah_pi05_l0_h1_5fps_g128_20260921/manifest.json @@ -0,0 +1,177 @@ +{ + "dataset_name": "half_cheetah_h1_5fps_l0", + "env_name": "half_cheetah_rgb_state", + "episodes": 90, + "frames": 90000, + "task_prompts": [ + "Make the Half Cheetah run forward as fast as possible. Predict six continuous torques in [-1, 1] ordered as back thigh, back shin, back foot, front thigh, front shin, and front 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": "half_cheetah_rgb_state", + "action_layout": "gymnasium_continuous_v1", + "action_labels": [ + "back_thigh_torque", + "back_shin_torque", + "back_foot_torque", + "front_thigh_torque", + "front_shin_torque", + "front_foot_torque" + ], + "carrier_action_labels": [ + "back_thigh_torque", + "back_shin_torque", + "back_foot_torque", + "front_thigh_torque", + "front_shin_torque", + "front_foot_torque" + ], + "action_dim": 6, + "active_action_dim": 6, + "action_carrier": "native", + "rows_unit": "decision_step", + "obs_fps": 5, + "obs_stride_raw_frames": 1, + "uses_state": true, + "state_dim": 17, + "state_labels": [ + "front_tip_height", + "front_tip_angle", + "back_thigh_angle", + "back_shin_angle", + "back_foot_angle", + "front_thigh_angle", + "front_shin_angle", + "front_foot_angle", + "front_tip_x_velocity", + "front_tip_z_velocity", + "front_tip_angular_velocity", + "back_thigh_angular_velocity", + "back_shin_angular_velocity", + "back_foot_angular_velocity", + "front_thigh_angular_velocity", + "front_shin_angular_velocity", + "front_foot_angular_velocity" + ], + "state_normalization": { + "type": "min_max", + "min": [ + -0.5893064737319946, + -1.0017330646514893, + -0.7351728081703186, + -0.8334724307060242, + -0.535628616809845, + -1.051438808441162, + -0.989691436290741, + -0.6309299468994141, + -2.3919641971588135, + -4.381032943725586, + -8.257936477661133, + -23.92654800415039, + -26.088411331176758, + -22.936174392700195, + -22.609085083007812, + -26.382095336914062, + -18.172616958618164 + ], + "max": [ + 0.7243065237998962, + 6.765694618225098, + 1.062463402748108, + 0.8386014103889465, + 0.8500239253044128, + 0.8450682759284973, + 0.9588844776153564, + 0.6611841917037964, + 9.299489974975586, + 3.903606414794922, + 8.0291166305542, + 20.77332305908203, + 29.168107986450195, + 19.582447052001953, + 24.40692138671875, + 28.59392547607422, + 22.969768524169922 + ] + }, + "robot_type": "rl_games_gymnasium", + "latency_prompt_map_path": "${PI05_RUN_DIR}/zero_latency/data/lerobot/half_cheetah_h1_5fps_l0/latency_prompt_map.json", + "custom_mixtures_path": "${PI05_RUN_DIR}/zero_latency/data/lerobot/_generated_mixtures/half_cheetah_h1_5fps_l0.json", + "gymnasium_task": { + "action_space": { + "dtype": "float32", + "high": [ + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0 + ], + "labels": [ + "back_thigh_torque", + "back_shin_torque", + "back_foot_torque", + "front_thigh_torque", + "front_shin_torque", + "front_foot_torque" + ], + "low": [ + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0 + ], + "type": "box" + }, + "base_prompt": "Make the Half Cheetah run forward as fast as possible. Predict six continuous torques in [-1, 1] ordered as back thigh, back shin, back foot, front thigh, front shin, and front foot.", + "env_fps": 5, + "env_id": "LatencyBench/HalfCheetahContinuous-v0", + "frame_stack": 1, + "make_kwargs": { + "base_env_id": "HalfCheetah-v4", + "base_make_kwargs": {}, + "render_mode": "rgb_array" + }, + "noop_action": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ], + "obs_fps": 5, + "registration_imports": [ + "latency_bench.envs.gymnasium_half_cheetah" + ], + "state_space": { + "labels": [ + "front_tip_height", + "front_tip_angle", + "back_thigh_angle", + "back_shin_angle", + "back_foot_angle", + "front_thigh_angle", + "front_shin_angle", + "front_foot_angle", + "front_tip_x_velocity", + "front_tip_z_velocity", + "front_tip_angular_velocity", + "back_thigh_angular_velocity", + "back_shin_angular_velocity", + "back_foot_angular_velocity", + "front_thigh_angular_velocity", + "front_shin_angular_velocity", + "front_foot_angular_velocity" + ] + }, + "task_name": "half_cheetah_rgb_state" + }, + "validation_dataset_name": "half_cheetah_h1_5fps_l0__val", + "validation_episodes": 10, + "validation_frames": 10000 +} \ No newline at end of file diff --git a/zero-latency/half-cheetah/vla/starvla-qwenpi_v3-h1/half_cheetah_pi05_l0_h1_5fps_g128_20260921/provenance.json b/zero-latency/half-cheetah/vla/starvla-qwenpi_v3-h1/half_cheetah_pi05_l0_h1_5fps_g128_20260921/provenance.json new file mode 100644 index 0000000000000000000000000000000000000000..770447d05055156855c82a788c171959bc82b40e --- /dev/null +++ b/zero-latency/half-cheetah/vla/starvla-qwenpi_v3-h1/half_cheetah_pi05_l0_h1_5fps_g128_20260921/provenance.json @@ -0,0 +1,34 @@ +{ + "task": "half-cheetah", + "model": "qwenpi_v3", + "training_condition": "zero-latency", + "training_run_id": "half_cheetah_pi05_l0_h1_5fps_g128_20260921", + "source": { + "repo_id": "latency-sensitive-bench/Standard-Pipeline", + "revision": "2208875f92b21e3f8ec2363a30a6705dfe56cc04", + "path": "half_cheetah/zero_latency/Pi05", + "url": "https://huggingface.co/latency-sensitive-bench/Standard-Pipeline/tree/2208875f92b21e3f8ec2363a30a6705dfe56cc04/half_cheetah/zero_latency/Pi05" + }, + "hf_repo_id": "latency-sensitive-bench/benchmark-models", + "hf_path": "zero-latency/half-cheetah/vla/starvla-qwenpi_v3-h1/half_cheetah_pi05_l0_h1_5fps_g128_20260921", + "checkpoint": { + "source_file": "half_cheetah/zero_latency/Pi05/checkpoints/model.pt", + "source_sha256": "58a8cf13cd5cae8071c7da2085410c1a039eaab255486c3814308bbaf561c2c4", + "source_bytes": 10922653853, + "file": "checkpoints/model.pt", + "selection_rule": "only published VLA checkpoint", + "method": "copy", + "sha256": "58a8cf13cd5cae8071c7da2085410c1a039eaab255486c3814308bbaf561c2c4", + "bytes": 10922653853 + }, + "config_source": "half_cheetah/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/half-cheetah/vla/starvla-qwenpi_v3-h1/half_cheetah_pi05_l0_h1_5fps_g128_20260921/source/README.md b/zero-latency/half-cheetah/vla/starvla-qwenpi_v3-h1/half_cheetah_pi05_l0_h1_5fps_g128_20260921/source/README.md new file mode 100644 index 0000000000000000000000000000000000000000..528bcc326c010005e3a7ef2db0ee0604a85174b5 --- /dev/null +++ b/zero-latency/half-cheetah/vla/starvla-qwenpi_v3-h1/half_cheetah_pi05_l0_h1_5fps_g128_20260921/source/README.md @@ -0,0 +1,3 @@ +# HalfCheetah Pi0.5 H1 — zero latency + +Fresh QwenPI_v3 action head, pinned Qwen3-VL-4B-Instruct, 5000 updates, seed42, global128=micro64 x accumulation1 x2H200, GCfalse, ZeRO2. RGB224/state17/native continuous action6/H1, 5/5FPS, train-only state minmax once. Training and saved real-state forward completed; L0sim20, RTX3090 P/B/F remain separate pending stages. Inference bundle excludes optimizer and full training state. diff --git a/zero-latency/half-cheetah/vla/starvla-qwenpi_v3-h1/half_cheetah_pi05_l0_h1_5fps_g128_20260921/source/provenance.json b/zero-latency/half-cheetah/vla/starvla-qwenpi_v3-h1/half_cheetah_pi05_l0_h1_5fps_g128_20260921/source/provenance.json new file mode 100644 index 0000000000000000000000000000000000000000..f1015b7b4cce6b258b4f02fc60073728e7c10afe --- /dev/null +++ b/zero-latency/half-cheetah/vla/starvla-qwenpi_v3-h1/half_cheetah_pi05_l0_h1_5fps_g128_20260921/source/provenance.json @@ -0,0 +1,118 @@ +{ + "task": "half_cheetah", + "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": "half_cheetah_pi05_l0_h1_5fps_g128_20260921", + "condition": "zero_latency", + "source": { + "task": "half_cheetah", + "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": "half_cheetah/zero_latency/shared/source_100ep_v1/demonstrations/raw/", + "sha256": { + "metadata.json": "267da8c4d6ba9f84c1f15f70efff3869ebded9141d0ab756783b181c9ba00541", + "train.parquet": "2ceaf2f1d636455b83cc14f41607c6e59a7ce7a22c7e070e6d0958e162f56783", + "val.parquet": "34442961a49b38bf296066f2925ed1300665e2585ed7a68a81f0ecc399a4a58d" + }, + "timing_derivation": { + "source_env_fps": 20, + "source_obs_fps": 20, + "target_env_fps": 5, + "target_obs_fps": 5, + "source_revision": "b0cdd0505f6c6f2dadfca67638e4fec1b252756c", + "source_prefix": "half_cheetah/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": 90000, + "episodes": 90, + "return_min": 3259.0414638389484, + "return_mean": 5647.154794339976, + "action_min": -1.0, + "action_max": 1.0 + }, + "val": { + "rows": 10000, + "episodes": 10, + "return_min": 5400.108074769378, + "return_mean": 5715.760363031365, + "action_min": -1.0, + "action_max": 1.0 + } + } + }, + "backbone": { + "repo_id": "Qwen/Qwen3-VL-4B-Instruct", + "revision": "ebb281ec70b05090aa6165b016eac8ec08e71b17" + }, + "protocol": { + "env_fps": 5, + "obs_fps": 5, + "action_horizon": 1, + "state_dim": 17, + "action_dim": 6, + "action_units": "native continuous torques, environment clips to [-1,1]", + "native_physics_unchanged": true, + "updates": 5000, + "seed": 42, + "global_batch": 128, + "micro_batch": 64, + "accumulation": 1, + "gpus": 2, + "physical_gpu_indices": [ + 4, + 5 + ], + "gradient_checkpointing": false, + "zero_stage": 2, + "allocator": "expandable_segments:True", + "state_encoding": "continuous_projector" + } + }, + "training_config_sha256": "e4402abb97180b07f2ab4cefa272919b654f3751bc2f35b955ca32427b25f3fe", + "dataset_manifest_sha256": "20504867b431364d76fcc4b62ecf2d21c0f8fbfec280891640499c53754368b8" +} diff --git a/zero-latency/half-cheetah/vla/starvla-qwenpi_v3-h1/half_cheetah_pi05_l0_h1_5fps_g128_20260921/task_contract.json b/zero-latency/half-cheetah/vla/starvla-qwenpi_v3-h1/half_cheetah_pi05_l0_h1_5fps_g128_20260921/task_contract.json new file mode 100644 index 0000000000000000000000000000000000000000..7ad1ce208f16919ea674c92c9a46a6709adc3f68 --- /dev/null +++ b/zero-latency/half-cheetah/vla/starvla-qwenpi_v3-h1/half_cheetah_pi05_l0_h1_5fps_g128_20260921/task_contract.json @@ -0,0 +1,73 @@ +{ + "action_space": { + "dtype": "float32", + "high": [ + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0 + ], + "labels": [ + "back_thigh_torque", + "back_shin_torque", + "back_foot_torque", + "front_thigh_torque", + "front_shin_torque", + "front_foot_torque" + ], + "low": [ + -1.0, + -1.0, + -1.0, + -1.0, + -1.0, + -1.0 + ], + "type": "box" + }, + "base_prompt": "Make the Half Cheetah run forward as fast as possible. Predict six continuous torques in [-1, 1] ordered as back thigh, back shin, back foot, front thigh, front shin, and front foot.", + "env_fps": 5, + "env_id": "LatencyBench/HalfCheetahContinuous-v0", + "frame_stack": 1, + "make_kwargs": { + "base_env_id": "HalfCheetah-v4", + "base_make_kwargs": {}, + "render_mode": "rgb_array" + }, + "noop_action": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ], + "obs_fps": 5, + "registration_imports": [ + "latency_bench.envs.gymnasium_half_cheetah" + ], + "state_space": { + "labels": [ + "front_tip_height", + "front_tip_angle", + "back_thigh_angle", + "back_shin_angle", + "back_foot_angle", + "front_thigh_angle", + "front_shin_angle", + "front_foot_angle", + "front_tip_x_velocity", + "front_tip_z_velocity", + "front_tip_angular_velocity", + "back_thigh_angular_velocity", + "back_shin_angular_velocity", + "back_foot_angular_velocity", + "front_thigh_angular_velocity", + "front_shin_angular_velocity", + "front_foot_angular_velocity" + ] + }, + "task_name": "half_cheetah_rgb_state" +} diff --git a/zero-latency/half-cheetah/vla/starvla-qwenpi_v3-h1/half_cheetah_pi05_l0_h1_5fps_g128_20260921/validation.json b/zero-latency/half-cheetah/vla/starvla-qwenpi_v3-h1/half_cheetah_pi05_l0_h1_5fps_g128_20260921/validation.json new file mode 100644 index 0000000000000000000000000000000000000000..b425e619cadbed2fef383fea8dcb3d16e53701c4 --- /dev/null +++ b/zero-latency/half-cheetah/vla/starvla-qwenpi_v3-h1/half_cheetah_pi05_l0_h1_5fps_g128_20260921/validation.json @@ -0,0 +1,18 @@ +{ + "state": "SAVED_BUNDLE_RELOAD_AND_REAL_SAMPLE_FORWARD_VERIFIED", + "verified_at": "2026-09-20T23:36:19.938715+00:00", + "forward_shape": [ + 1, + 1, + 6 + ], + "forward_finite": true, + "heldout_sample": 0, + "loader_rows": 10000, + "action_horizon": 1, + "state_dim": 17, + "action_dim": 6, + "state_encoding": "continuous_projector", + "checkpoint_sha256": "58a8cf13cd5cae8071c7da2085410c1a039eaab255486c3814308bbaf561c2c4", + "reload": "Existing strict key check with documented tied Qwen lm_head equivalence only" +}