Publish completed deadly_corridor RTX3090 mean-profile teacher

latency-aware/deadly-corridor/small-policy/sample-factory-qwenoft-rtx3090-mean-hist8-capacity1-native35-s42-v1/README.md

@@ -0,0 +1,5 @@
+# deadly_corridor Sample Factory mean-profile teacher
+
+Completed 25006080 frames with seed 42; budget 25000000. Selection: best_reward_after_full_training_budget. Fixed mean latency 73.69250777493353 ms from archived RTX3090 profile; training GPU was RTX PRO6000 Blackwell.
+
+`checkpoint.pth` is a tensor-only inference export with `model`, `train_step`, `env_steps`; every model tensor compared exactly with selected training checkpoint. Original recovery state remains on host. Training reward in source filename is not independent task evaluation.

latency-aware/deadly-corridor/small-policy/sample-factory-qwenoft-rtx3090-mean-hist8-capacity1-native35-s42-v1/checkpoint.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e631e9017634f799860c581761b5fa48601109393c11d20a35187c6f75d14b88
+size 13354997

latency-aware/deadly-corridor/small-policy/sample-factory-qwenoft-rtx3090-mean-hist8-capacity1-native35-s42-v1/config.json

@@ -0,0 +1,271 @@
+{
+  "help": false,
+  "algo": "APPO",
+  "env": "doom_deadly_corridor",
+  "experiment": "deadly_corridor_mean_3090_s42",
+  "train_dir": "/home/ubuntu/lzj/mean-profiling/deadly_corridor/teacher",
+  "restart_behavior": "resume",
+  "device": "gpu",
+  "seed": 42,
+  "num_policies": 1,
+  "async_rl": true,
+  "serial_mode": false,
+  "batched_sampling": false,
+  "num_batches_to_accumulate": 2,
+  "worker_num_splits": 2,
+  "policy_workers_per_policy": 1,
+  "max_policy_lag": 1000,
+  "num_workers": 32,
+  "num_envs_per_worker": 4,
+  "batch_size": 1024,
+  "num_batches_per_epoch": 2,
+  "num_epochs": 2,
+  "rollout": 128,
+  "recurrence": 128,
+  "shuffle_minibatches": false,
+  "gamma": 0.99,
+  "reward_scale": 1.0,
+  "reward_clip": 1000.0,
+  "value_bootstrap": false,
+  "normalize_returns": true,
+  "exploration_loss_coeff": 0.001,
+  "value_loss_coeff": 0.5,
+  "kl_loss_coeff": 0.0,
+  "exploration_loss": "symmetric_kl",
+  "gae_lambda": 0.95,
+  "ppo_clip_ratio": 0.1,
+  "ppo_clip_value": 0.2,
+  "with_vtrace": false,
+  "vtrace_rho": 1.0,
+  "vtrace_c": 1.0,
+  "optimizer": "adam",
+  "adam_eps": 1e-06,
+  "adam_beta1": 0.9,
+  "adam_beta2": 0.999,
+  "max_grad_norm": 4.0,
+  "learning_rate": 0.0001,
+  "lr_schedule": "constant",
+  "lr_schedule_kl_threshold": 0.008,
+  "lr_adaptive_min": 1e-06,
+  "lr_adaptive_max": 0.01,
+  "obs_subtract_mean": 0.0,
+  "obs_scale": 255.0,
+  "normalize_input": true,
+  "normalize_input_keys": null,
+  "decorrelate_experience_max_seconds": 0,
+  "decorrelate_envs_on_one_worker": true,
+  "actor_worker_gpus": [],
+  "set_workers_cpu_affinity": true,
+  "force_envs_single_thread": false,
+  "default_niceness": 0,
+  "log_to_file": true,
+  "experiment_summaries_interval": 1,
+  "flush_summaries_interval": 30,
+  "stats_avg": 100,
+  "summaries_use_frameskip": true,
+  "heartbeat_interval": 20,
+  "heartbeat_reporting_interval": 600,
+  "train_for_env_steps": 25000000,
+  "train_for_seconds": 10000000000,
+  "save_every_sec": 600,
+  "keep_checkpoints": 5,
+  "load_checkpoint_kind": "latest",
+  "save_milestones_sec": -1,
+  "save_best_every_sec": 5,
+  "save_best_metric": "reward",
+  "save_best_after": 100000,
+  "benchmark": false,
+  "encoder_mlp_layers": [
+    512,
+    512
+  ],
+  "encoder_conv_architecture": "convnet_simple",
+  "encoder_conv_mlp_layers": [
+    512
+  ],
+  "use_rnn": true,
+  "rnn_size": 512,
+  "rnn_type": "gru",
+  "rnn_num_layers": 1,
+  "decoder_mlp_layers": [],
+  "nonlinearity": "elu",
+  "policy_initialization": "orthogonal",
+  "policy_init_gain": 1.0,
+  "actor_critic_share_weights": true,
+  "adaptive_stddev": true,
+  "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": "starvla_tasks",
+  "wandb_group": "mean-profiling-3090-blackwell",
+  "wandb_job_type": "deadly_corridor_decision_history_native35_temporal",
+  "wandb_tags": [
+    "deadly_corridor_decision_history_native35_temporal_25m_seed42",
+    "deadly-corridor",
+    "frame_stack",
+    "temporal",
+    "qwenoft_rtx3090_temporal_hist8_capacity1_native35",
+    "fs4",
+    "seed42"
+  ],
+  "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,
+  "num_agents": -1,
+  "num_humans": 0,
+  "num_bots": -1,
+  "start_bot_difficulty": null,
+  "timelimit": null,
+  "res_w": 128,
+  "res_h": 72,
+  "wide_aspect_ratio": false,
+  "record_to": null,
+  "mode": "train",
+  "env_fps": 35.0,
+  "obs_fps": 8.75,
+  "latency_type": "fixed",
+  "fixed_latency_ms": 73.69250777493353,
+  "mean_latency_ms": null,
+  "std_latency_ms": null,
+  "min_latency_ms": null,
+  "max_latency_ms": null,
+  "latency_seed": 42,
+  "latency_profile_path": null,
+  "latency_profile_worker_slot": null,
+  "simulated_worker_capacity": 1,
+  "add_latency_info": false,
+  "max_pending_actions": null,
+  "hold_policy": "hold",
+  "ordering_policy": "latest_ready",
+  "eval_episodes": 100,
+  "eval_parallel_envs": 1,
+  "eval_latency_raw_frame_values": "0,1,2,3,4,5,10,15,40",
+  "eval_max_steps": 3600,
+  "eval_raw_reward": true,
+  "eval_deterministic": true,
+  "disable_reward_scaling": true,
+  "frame_stack": 4,
+  "action_history_decisions": 8,
+  "max_episode_steps": 0,
+  "simulator": "cpu",
+  "episode_metrics_path": "/home/ubuntu/lzj/mean-profiling/deadly_corridor/teacher/deadly_corridor_mean_3090_s42/episode_metrics.jsonl",
+  "vizdoom_init_max_parallel": 10,
+  "vizdoom_init_lock_timeout_sec": 5.0,
+  "vizdoom_init_lock_max_wait_sec": 60.0,
+  "vizdoom_init_retry_jitter_sec": 0.25,
+  "eval_env_frameskip": 1,
+  "fps": 35,
+  "command_line": "--mode train --algo APPO --env doom_deadly_corridor --experiment deadly_corridor_frame_stack_qwenoft_rtx3090_temporal_hist8_capacity1_native35_fs4_seed42 --train_dir results/checkpoints/deadly_corridor/decision_history_native35_temporal_25m_seed42 --restart_behavior resume --device gpu --seed 42 --episode_metrics_path results/checkpoints/deadly_corridor/decision_history_native35_temporal_25m_seed42/deadly_corridor_frame_stack_qwenoft_rtx3090_temporal_hist8_capacity1_native35_fs4_seed42/episode_metrics.jsonl --train_for_env_steps 25000000 --num_workers 32 --num_envs_per_worker 4 --num_policies 1 --batch_size 1024 --rollout 128 --recurrence 128 --num_epochs 2 --num_batches_per_epoch 2 --worker_num_splits 2 --learning_rate 0.0001 --gamma 0.99 --gae_lambda 0.95 --ppo_clip_ratio 0.1 --ppo_clip_value 0.2 --exploration_loss symmetric_kl --exploration_loss_coeff 0.001 --value_loss_coeff 0.5 --max_grad_norm 4.0 --rnn_type gru --rnn_size 512 --save_every_sec 600 --keep_checkpoints 5 --stats_avg 100 --experiment_summaries_interval 1 --async_rl True --use_rnn True --normalize_returns True --normalize_input True --latency-type profile_sample --latency-seed 42 --latency-profile-path results/profiling/profiles/openvla/1x-rtx3090/deadly_corridor/instance_a5037b165aa0cedc/profile.json --latency-profile-worker-slot 0 --simulated-worker-capacity 1 --add-latency-info False --eval-episodes 100 --eval-parallel-envs 1 --eval-max-steps 3600 --eval-deterministic True --eval-raw-reward --with_wandb True --wandb_project starvla_tasks --wandb_group deadly-corridor-fs4-qwenoft_rtx3090_temporal_hist8_capacity1_native35 --wandb_job_type deadly_corridor_decision_history_native35_temporal --wandb_tags deadly_corridor_decision_history_native35_temporal_25m_seed42 deadly-corridor frame_stack temporal qwenoft_rtx3090_temporal_hist8_capacity1_native35 fs4 seed42 --env-fps 35 --obs-fps 8.75 --res_w 128 --res_h 72 --wide_aspect_ratio False --frame_stack 4 --action-history-decisions 8 --simulator cpu --disable-reward-scaling True --hold-policy hold --ordering-policy latest_ready",
+  "cli_args": {
+    "algo": "APPO",
+    "env": "doom_deadly_corridor",
+    "experiment": "deadly_corridor_mean_3090_s42",
+    "train_dir": "/home/ubuntu/lzj/mean-profiling/deadly_corridor/teacher",
+    "restart_behavior": "resume",
+    "device": "gpu",
+    "seed": 42,
+    "num_policies": 1,
+    "async_rl": true,
+    "worker_num_splits": 2,
+    "num_workers": 32,
+    "num_envs_per_worker": 4,
+    "batch_size": 1024,
+    "num_batches_per_epoch": 2,
+    "num_epochs": 2,
+    "rollout": 128,
+    "recurrence": 128,
+    "gamma": 0.99,
+    "normalize_returns": true,
+    "exploration_loss_coeff": 0.001,
+    "value_loss_coeff": 0.5,
+    "exploration_loss": "symmetric_kl",
+    "gae_lambda": 0.95,
+    "ppo_clip_ratio": 0.1,
+    "ppo_clip_value": 0.2,
+    "max_grad_norm": 4.0,
+    "learning_rate": 0.0001,
+    "normalize_input": true,
+    "experiment_summaries_interval": 1,
+    "stats_avg": 100,
+    "train_for_env_steps": 25000000,
+    "save_every_sec": 600,
+    "keep_checkpoints": 5,
+    "use_rnn": true,
+    "rnn_size": 512,
+    "rnn_type": "gru",
+    "with_wandb": true,
+    "wandb_project": "starvla_tasks",
+    "wandb_group": "mean-profiling-3090-blackwell",
+    "wandb_job_type": "deadly_corridor_decision_history_native35_temporal",
+    "wandb_tags": [
+      "deadly_corridor_decision_history_native35_temporal_25m_seed42",
+      "deadly-corridor",
+      "frame_stack",
+      "temporal",
+      "qwenoft_rtx3090_temporal_hist8_capacity1_native35",
+      "fs4",
+      "seed42"
+    ],
+    "res_w": 128,
+    "res_h": 72,
+    "wide_aspect_ratio": false,
+    "mode": "train",
+    "env_fps": 35.0,
+    "obs_fps": 8.75,
+    "latency_type": "fixed",
+    "latency_seed": 42,
+    "latency_profile_path": null,
+    "latency_profile_worker_slot": null,
+    "simulated_worker_capacity": 1,
+    "add_latency_info": false,
+    "hold_policy": "hold",
+    "ordering_policy": "latest_ready",
+    "eval_episodes": 100,
+    "eval_parallel_envs": 1,
+    "eval_max_steps": 3600,
+    "eval_raw_reward": true,
+    "eval_deterministic": true,
+    "disable_reward_scaling": true,
+    "frame_stack": 4,
+    "action_history_decisions": 8,
+    "simulator": "cpu",
+    "episode_metrics_path": "/home/ubuntu/lzj/mean-profiling/deadly_corridor/teacher/deadly_corridor_mean_3090_s42/episode_metrics.jsonl",
+    "env_frameskip": 1,
+    "eval_env_frameskip": 1,
+    "output_dir": "/home/ubuntu/lzj/mean-profiling/deadly_corridor/teacher-log",
+    "wandb_unique_id": "mean3090-deadly_corridor-teacher-blackwell-20260930-nice0"
+  },
+  "git_hash": "unknown",
+  "git_repo_name": "not a git repository",
+  "initial_model_path": null,
+  "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,
+  "episode_counter": false,
+  "wandb_dir": "/home/ubuntu/lzj/mean-profiling/deadly_corridor/wandb",
+  "output_dir": "/home/ubuntu/lzj/mean-profiling/deadly_corridor/teacher-log",
+  "wandb_unique_id": "mean3090-deadly_corridor-teacher-blackwell-20260930-nice0"
+}

latency-aware/deadly-corridor/small-policy/sample-factory-qwenoft-rtx3090-mean-hist8-capacity1-native35-s42-v1/files.sha256.json

@@ -0,0 +1,22 @@
+{
+  "checkpoint.pth": {
+    "bytes": 13354997,
+    "sha256": "e631e9017634f799860c581761b5fa48601109393c11d20a35187c6f75d14b88"
+  },
+  "provenance.json": {
+    "bytes": 1203,
+    "sha256": "5c75e26eec6f30405f68fcfb3f7591afa6a3f90272e663bac8c7f9bb770c6fa0"
+  },
+  "README.md": {
+    "bytes": 550,
+    "sha256": "5c1cbd311b0d41c241e71c9a5dfb7c4a35287edb0ca3c6c9d43830188241bb7a"
+  },
+  "profile.source.json": {
+    "bytes": 2115,
+    "sha256": "bbfb93cef9f63750a9a159ec10a93a9fc6c25e4cb0cac3b39532663b4dc11eba"
+  },
+  "config.json": {
+    "bytes": 10277,
+    "sha256": "56d5e2ed3fd9b8dfd9c29c9c305f2b0c2267cebb32ebdfecf8bbfdb1cbbb0e53"
+  }
+}

latency-aware/deadly-corridor/small-policy/sample-factory-qwenoft-rtx3090-mean-hist8-capacity1-native35-s42-v1/profile.source.json

@@ -0,0 +1,66 @@
+{
+  "burst_model_path": "latency_burst_model.json",
+  "distribution_path": "latency_distribution.json",
+  "env_fps": 60,
+  "frame_ms": 16.666666666666668,
+  "gpu_class": "1x-rtx3090",
+  "instance_id": "instance_a5037b165aa0cedc",
+  "latency_kind": "observation_to_action_latency",
+  "latency_method": "temporal",
+  "model_id": "openvla",
+  "n_admitted_observations": 506,
+  "n_capacity_drops": 493,
+  "n_observation_attempts": 999,
+  "per_slot_summary": {
+    "0": {
+      "admitted_count": 506,
+      "mean_observation_to_action_latency_ms": 73.69250777493353,
+      "mean_worker_service_time_ms": 73.00619814047229,
+      "p95_observation_to_action_latency_ms": 78.0192643404007,
+      "p95_worker_service_time_ms": 76.60398542881012,
+      "p99_worker_service_time_ms": 93.66708476543427
+    }
+  },
+  "provenance": {
+    "base_config": "configs/experiments/deadly_corridor/starvla_profile.yaml",
+    "checkpoint_kind": "best",
+    "model_artifact": {
+      "checkpoint": "checkpoints/steps_500_state/model.safetensors",
+      "model_config": "config.full.yaml",
+      "path_in_repo": ".",
+      "repo_id": "talha15032/openvla_bridge_deadly_corridor_single_latency_clean_data_bce_exp2",
+      "source": "local"
+    },
+    "session_ids": [
+      0,
+      1,
+      2,
+      3,
+      4
+    ]
+  },
+  "sample_model_type": "hidden_regime",
+  "source_run_id": "20260914T171446047509Z",
+  "summary": {
+    "frame_ms": 16.666666666666668,
+    "max_ms": 101.71253681182861,
+    "mean_effective_frames": 4.4215504664960115,
+    "mean_ms": 73.69250777493353,
+    "min_ms": 68.33836698532104,
+    "n_samples": 506,
+    "p50_frames": 4.3794349193572994,
+    "p50_ms": 72.99058198928833,
+    "p90_frames": 4.606898102760314,
+    "p90_ms": 76.78163504600525,
+    "p95_frames": 4.681155860424042,
+    "p95_ms": 78.0192643404007,
+    "p99_frames": 5.741582728385924,
+    "p99_ms": 95.69304547309875,
+    "prob_latency_gt_1_frame": 1.0,
+    "prob_latency_gt_2_frames": 1.0,
+    "prob_latency_gt_3_frames": 1.0,
+    "std_ms": 4.766097168829327
+  },
+  "visualization_path": "latency_profile.png",
+  "workload_id": "deadly_corridor"
+}

latency-aware/deadly-corridor/small-policy/sample-factory-qwenoft-rtx3090-mean-hist8-capacity1-native35-s42-v1/provenance.json

@@ -0,0 +1,26 @@
+{
+  "task": "deadly_corridor",
+  "algorithm": "APPO",
+  "selection_rule": "best_reward_after_full_training_budget",
+  "budget_env_steps": 25000000,
+  "completed_env_steps": 25006080,
+  "selected_train_step": 43436,
+  "selected_env_steps": 22239232,
+  "selected_source": "/home/ubuntu/lzj/mean-profiling/deadly_corridor/teacher/deadly_corridor_mean_3090_s42/checkpoint_p0/best_000043436_22239232_reward_2141.427.pth",
+  "selected_source_sha256": "7c2d1106627daed048648060b5c694c88366e35adb2baaacd3a820d7dc3b6590",
+  "checkpoint_sha256": "e631e9017634f799860c581761b5fa48601109393c11d20a35187c6f75d14b88",
+  "tensor_equality_verified": true,
+  "weights_only_export": true,
+  "seed": 42,
+  "env_fps": 35.0,
+  "obs_fps": 8.75,
+  "fixed_latency_ms": 73.69250777493353,
+  "profile_source": {
+    "mean_ms": 73.69250777493353,
+    "profile": "/home/ubuntu/lzj/profiles/published/profiles/openvla/1x-rtx3090/deadly_corridor/instance_a5037b165aa0cedc/profile.json",
+    "sha256": "bbfb93cef9f63750a9a159ec10a93a9fc6c25e4cb0cac3b39532663b4dc11eba"
+  },
+  "training_hardware": "RTX PRO6000 Blackwell physicalGPU0",
+  "training_status": "complete",
+  "closed_loop_evaluation": "not performed by this publication"
+}