Publish asterix qwengr00t H8 evaluation artifacts

zero-latency/asterix/vla/starvla-qwengr00t-h1/asterix-qwengr00t-hist8-zero-s42-v1/README.md

@@ -4,5 +4,5 @@ Current RGB and eight binary decision-history windows; action horizon one. Initi
 
 - Training: 5000 steps, seed 42, global batch 128
 - W&B: https://wandb.ai/saberrr-zju/starvla_tasks/runs/42b8e7f3
-- Evaluation: 100 episodes, seeds 1000000–1000099, GPU simulator, parallel 32, capacity 1, 60/15 Hz, 7200 raw-frame cap, zero latency
-- Mean raw return: 5714.0
+- Evaluation: 100 episodes, seeds 1000000–1000099, GPU simulator, parallel 32, capacity 1, 60/15 Hz, 7200 raw-frame cap, profile latency
+- Mean raw return: 182.0

zero-latency/asterix/vla/starvla-qwengr00t-h1/asterix-qwengr00t-hist8-zero-s42-v1/eval_profile.json

@@ -0,0 +1,217 @@
+{
+  "checkpoint_path": "/home/ubuntu/codex/atari-hist8-20260924/outputs/tasks/atari_hist8_v1/bundles/asterix-qwengr00t-hist8-zero-s42-v1/checkpoints/model.pt",
+  "experiment_name": "asterix-qwengr00t-hist8-zero-s42-v1-step5000-history01-profile-par32",
+  "latency": "profile_sample",
+  "latency_type": "profile_sample",
+  "lengths": [
+    1543,
+    785,
+    2069,
+    998,
+    1259,
+    841,
+    906,
+    1028,
+    1460,
+    805,
+    1006,
+    1091,
+    777,
+    1544,
+    1842,
+    708,
+    799,
+    1074,
+    1880,
+    1621,
+    1063,
+    782,
+    782,
+    1195,
+    779,
+    1287,
+    2543,
+    1403,
+    1454,
+    1060,
+    1365,
+    790,
+    799,
+    1174,
+    1267,
+    1258,
+    1532,
+    1210,
+    803,
+    1159,
+    781,
+    1316,
+    967,
+    1904,
+    1066,
+    881,
+    792,
+    1167,
+    1231,
+    790,
+    805,
+    800,
+    916,
+    782,
+    856,
+    789,
+    800,
+    1075,
+    779,
+    798,
+    1468,
+    1214,
+    1177,
+    1358,
+    1354,
+    893,
+    1551,
+    780,
+    836,
+    818,
+    933,
+    940,
+    793,
+    895,
+    1125,
+    786,
+    1587,
+    785,
+    1285,
+    1861,
+    802,
+    1941,
+    788,
+    1353,
+    1258,
+    799,
+    1680,
+    1542,
+    1344,
+    799,
+    931,
+    1467,
+    996,
+    836,
+    1032,
+    1058,
+    1831,
+    1087,
+    1238,
+    1083
+  ],
+  "mean_length": 1133.4,
+  "mean_return": 182.0,
+  "returns": [
+    300.0,
+    150.0,
+    400.0,
+    350.0,
+    200.0,
+    100.0,
+    50.0,
+    300.0,
+    150.0,
+    50.0,
+    150.0,
+    50.0,
+    100.0,
+    250.0,
+    300.0,
+    150.0,
+    100.0,
+    250.0,
+    400.0,
+    350.0,
+    150.0,
+    0.0,
+    50.0,
+    300.0,
+    50.0,
+    350.0,
+    550.0,
+    350.0,
+    150.0,
+    100.0,
+    300.0,
+    200.0,
+    100.0,
+    150.0,
+    300.0,
+    150.0,
+    300.0,
+    250.0,
+    0.0,
+    100.0,
+    50.0,
+    150.0,
+    300.0,
+    350.0,
+    50.0,
+    100.0,
+    50.0,
+    150.0,
+    200.0,
+    150.0,
+    50.0,
+    0.0,
+    150.0,
+    0.0,
+    50.0,
+    100.0,
+    50.0,
+    200.0,
+    50.0,
+    100.0,
+    250.0,
+    350.0,
+    150.0,
+    150.0,
+    200.0,
+    100.0,
+    200.0,
+    50.0,
+    200.0,
+    250.0,
+    100.0,
+    150.0,
+    100.0,
+    50.0,
+    250.0,
+    50.0,
+    400.0,
+    50.0,
+    250.0,
+    600.0,
+    50.0,
+    500.0,
+    150.0,
+    300.0,
+    200.0,
+    100.0,
+    400.0,
+    150.0,
+    200.0,
+    50.0,
+    100.0,
+    300.0,
+    200.0,
+    0.0,
+    200.0,
+    100.0,
+    300.0,
+    250.0,
+    300.0,
+    100.0
+  ],
+  "seed": 1000000,
+  "source_profile_path": "/home/ubuntu/codex/atari-hist8-20260924/outputs/tasks/atari_hist8_v1/teacher_configs/asterix-qwengr00t/profile/profile.json",
+  "std_return": 125.80143083447024,
+  "suite_name": "profile_sample",
+  "timestamp_utc": "2026-09-24T12:54:01.944742+00:00"
+}

zero-latency/asterix/vla/starvla-qwengr00t-h1/asterix-qwengr00t-hist8-zero-s42-v1/evaluation_profile/action_audit.json

@@ -0,0 +1,153 @@
+{
+  "eval_protocol_version": "hist8-history01-sf-render-v4",
+  "history_encoding": "binary_0_1",
+  "episodes": 100,
+  "episode_seeds": [
+    1000000,
+    1000001,
+    1000002,
+    1000003,
+    1000004,
+    1000005,
+    1000006,
+    1000007,
+    1000008,
+    1000009,
+    1000010,
+    1000011,
+    1000012,
+    1000013,
+    1000014,
+    1000015,
+    1000016,
+    1000017,
+    1000018,
+    1000019,
+    1000020,
+    1000021,
+    1000022,
+    1000023,
+    1000024,
+    1000025,
+    1000026,
+    1000027,
+    1000028,
+    1000029,
+    1000030,
+    1000031,
+    1000032,
+    1000033,
+    1000034,
+    1000035,
+    1000036,
+    1000037,
+    1000038,
+    1000039,
+    1000040,
+    1000041,
+    1000042,
+    1000043,
+    1000044,
+    1000045,
+    1000046,
+    1000047,
+    1000048,
+    1000049,
+    1000050,
+    1000051,
+    1000052,
+    1000053,
+    1000054,
+    1000055,
+    1000056,
+    1000057,
+    1000058,
+    1000059,
+    1000060,
+    1000061,
+    1000062,
+    1000063,
+    1000064,
+    1000065,
+    1000066,
+    1000067,
+    1000068,
+    1000069,
+    1000070,
+    1000071,
+    1000072,
+    1000073,
+    1000074,
+    1000075,
+    1000076,
+    1000077,
+    1000078,
+    1000079,
+    1000080,
+    1000081,
+    1000082,
+    1000083,
+    1000084,
+    1000085,
+    1000086,
+    1000087,
+    1000088,
+    1000089,
+    1000090,
+    1000091,
+    1000092,
+    1000093,
+    1000094,
+    1000095,
+    1000096,
+    1000097,
+    1000098,
+    1000099
+  ],
+  "issued_actions": 7130,
+  "action_counts": {
+    "6": 2679,
+    "3": 926,
+    "1": 491,
+    "2": 178,
+    "5": 354,
+    "0": 678,
+    "7": 832,
+    "4": 113,
+    "8": 879
+  },
+  "largest_action_fraction": 0.37573632538569424,
+  "raw_score_min": [
+    -0.11250734329223633,
+    -0.08865809440612793,
+    -0.09899623692035675,
+    -0.17915654182434082,
+    -0.10699084401130676,
+    -0.10701999068260193,
+    -0.14106741547584534,
+    -0.05952806770801544,
+    -0.17616045475006104
+  ],
+  "raw_score_max": [
+    1.0584603548049927,
+    1.0680583715438843,
+    1.0592747926712036,
+    1.0604641437530518,
+    1.0169434547424316,
+    1.0631811618804932,
+    1.0627062320709229,
+    1.1027902364730835,
+    1.0733144283294678
+  ],
+  "raw_score_mean": [
+    0.08782884661199036,
+    0.06818614123342755,
+    0.020853550182093786,
+    0.12822516662689765,
+    0.020668613626249395,
+    0.05096994273499736,
+    0.3689880494977286,
+    0.12228238381834657,
+    0.12859012346830986
+  ]
+}

zero-latency/asterix/vla/starvla-qwengr00t-h1/asterix-qwengr00t-hist8-zero-s42-v1/evaluation_profile/eval_config.yaml

@@ -0,0 +1,165 @@
+experiment:
+  name: asterix-qwengr00t-hist8-zero-s42-v1-step5000-history01-profile-par32
+  seed: 1000000
+backend:
+  type: sample_factory
+  algo: APPO
+  device: cuda
+  train_dir: results/sample_factory
+  restart_behavior: resume
+  run_mode: eval
+  extra_args:
+  - --encoder_conv_architecture
+  - convnet_simple
+executor:
+  mode: simulated
+  simulated_worker_capacity: 1
+  simulated_inference_pool: true
+  inference_devices:
+  - cuda:0
+  - cuda:1
+  inference_batch_size: 16
+env:
+  name: gymnasium
+  task_name: asterix
+  env_id: LatencyBench/Asterix-v0
+  registration_imports:
+  - latency_bench.envs.gymnasium_asterix
+  make_kwargs:
+    base_env_id: ALE/Asterix-v5
+    render_mode: rgb_array
+    screen_size: 84
+    noop_max: 30
+    base_make_kwargs:
+      obs_type: rgb
+      frameskip: 1
+      repeat_action_probability: 0.0
+      full_action_space: false
+      mode: 0
+      difficulty: 0
+      max_num_frames_per_episode: 108000
+  env_fps: 60
+  obs_fps: 15
+  frame_stack: 1
+  simulator: gpu
+  terminal_on_life_loss: false
+  clip_reward: false
+  noop_action: noop
+  action_map:
+    noop: 0
+    up: 1
+    right: 2
+    left: 3
+    down: 4
+    upright: 5
+    upleft: 6
+    downright: 7
+    downleft: 8
+  action_order:
+  - noop
+  - up
+  - right
+  - left
+  - down
+  - upright
+  - upleft
+  - downright
+  - downleft
+  base_prompt: 'Collect useful objects and avoid lyres. Choose exactly one action
+    from: noop, up, right, left, down, upright, upleft, downright, downleft.'
+  obs_resize:
+  - 224
+  - 224
+  gpu_env_device: auto
+  noop_max: 30
+  action_history_decisions: 8
+latency:
+  method: temporal
+  profile_path: /home/ubuntu/codex/atari-hist8-20260924/outputs/tasks/atari_hist8_v1/teacher_configs/asterix-qwengr00t/profile/profile.json
+  profile_worker_slot: 0
+  seed: 271828
+  add_latency_info: false
+scheduler:
+  hold_policy: hold
+  ordering_policy: issue_order_fifo
+policy:
+  type: starvla
+  actions:
+  - noop
+  - up
+  - right
+  - left
+  - down
+  - upright
+  - upleft
+  - downright
+  - downleft
+  checkpoint_path: /home/ubuntu/codex/atari-hist8-20260924/outputs/tasks/atari_hist8_v1/bundles/asterix-qwengr00t-hist8-zero-s42-v1/checkpoints/model.pt
+  model_config_path: /home/ubuntu/codex/atari-hist8-20260924/outputs/tasks/atari_hist8_v1/bundles/asterix-qwengr00t-hist8-zero-s42-v1/config.yaml
+  task_manifest_path: /home/ubuntu/codex/atari-hist8-20260924/outputs/tasks/atari_hist8_v1/bundles/asterix-qwengr00t-hist8-zero-s42-v1/manifest.json
+  device: cuda:0
+  unnorm_key: new_embodiment
+  prompt_mode: latency_neutral
+  action_layout: discrete_9
+  backbone_path: /home/ubuntu/.cache/huggingface/hub/models--Qwen--Qwen3-VL-4B-Instruct/snapshots/ebb281ec70b05090aa6165b016eac8ec08e71b17
+  worker_python_executable: /home/ubuntu/miniconda3/envs/starvla_rl_games_openvla/bin/python3.10
+  state_source: transport
+training:
+  train_for_env_steps: 1000000
+  num_workers: 4
+  num_envs_per_worker: 16
+  worker_num_splits: 1
+  num_policies: 1
+  batch_size: 1024
+  rollout: 64
+  recurrence: 32
+  num_epochs: 2
+  num_batches_per_epoch: 8
+  max_policy_lag: 300
+  learning_rate: 0.000303
+  lr_schedule: constant
+  nonlinearity: elu
+  gamma: 0.99
+  gae_lambda: 0.95
+  ppo_clip_ratio: 0.1
+  ppo_clip_value: 0.2
+  exploration_loss: entropy
+  exploration_loss_coeff: 0.003
+  value_loss_coeff: 0.5
+  max_grad_norm: 0.0
+  adam_eps: 1.0e-06
+  obs_scale: 1.0
+  adaptive_stddev: false
+  with_vtrace: false
+  async_rl: true
+  use_rnn: false
+  normalize_input: true
+  normalize_returns: true
+  stats_avg: 100
+  experiment_summaries_interval: 1
+  save_every_sec: 600
+  keep_checkpoints: 5
+evaluation:
+  eval_interval_steps: null
+  eval_episodes: 100
+  eval_parallel_envs: 32
+  eval_max_steps: 7200
+  eval_deterministic: true
+  eval_suites:
+    fixed: []
+    normal: []
+    uniform: []
+  latency_bench_env_backend: asterix_gpu_batched
+  eval_raw_reward: true
+logging:
+  output_dir: /home/ubuntu/codex/atari-hist8-20260924/outputs/tasks/atari_hist8_v1/evals/asterix-qwengr00t-hist8-zero-s42-v1/zero_to_profile
+  video:
+    enabled: false
+    num_bins: 1
+  save_step_records: true
+  save_action_records: true
+  save_latency_records: true
+  wandb_project: null
+  wandb_group: null
+  wandb_job_type: null
+  wandb_tags: ''

zero-latency/asterix/vla/starvla-qwengr00t-h1/asterix-qwengr00t-hist8-zero-s42-v1/evaluation_profile/provenance.json

@@ -0,0 +1,46 @@
+{
+  "collection": {
+    "episodes": 300,
+    "filter_preset": "none",
+    "history": "decision_v1",
+    "history_decisions": 8,
+    "history_source": "teacher_transport",
+    "max_episode_raw_frames": 7200,
+    "rollout_overrides": {
+      "env_fps": 60,
+      "fixed_latency_ms": 0.0,
+      "latency_type": "fixed",
+      "obs_fps": 15,
+      "seed": 42
+    },
+    "student_image_frames": 1,
+    "teacher_checkpoint": "best_000024112_12345344_reward_5614.500.pth",
+    "teacher_path": "zero-latency/asterix/small-policy/sample-factory-hist8-cpu15m-s0-v1",
+    "teacher_repo": "latency-sensitive-bench/benchmark-models",
+    "teacher_revision": "dfaeecb84cd6b5dfcc3adc0935d0ae86d0b504d2",
+    "teacher_sha256": "0fe54ff4b3a2ccec438aca789f1a9f86fcef7cda44cda5156e3b774262c50beb",
+    "train_labels": 337557,
+    "training_latency_condition": "zero"
+  },
+  "training_run_id": "asterix-qwengr00t-hist8-zero-s42-v1",
+  "training_step": 5000,
+  "global_batch": 128,
+  "base_model": "Qwen/Qwen3-VL-4B-Instruct",
+  "wandb_run_id": "42b8e7f3",
+  "checkpoint_sha256": "7b3541140bd2d00d39eca98c8788b773e594884a623ae20d7b9e4dd7b45337f5",
+  "eval_protocol_version": "hist8-history01-sf-render-v4",
+  "history_encoding": "binary_0_1",
+  "episodes": 100,
+  "env_condition": "profile",
+  "parallel_envs": 32,
+  "max_episode_raw_frames": 7200,
+  "inference_devices": [
+    "cuda:0",
+    "cuda:1"
+  ],
+  "episode_seeds": [
+    1000000,
+    1000099
+  ],
+  "profile_sha256": "5e104064d718ac51ce56aefac5318a155fd099f5c714ff8f0bc4985509639fd1"
+}

zero-latency/asterix/vla/starvla-qwengr00t-h1/asterix-qwengr00t-hist8-zero-s42-v1/profile_publication.json

@@ -0,0 +1,29 @@
+{
+  "model": "qwengr00t",
+  "model_publication": {
+    "checkpoint_sha256": "7b3541140bd2d00d39eca98c8788b773e594884a623ae20d7b9e4dd7b45337f5",
+    "episodes": 100,
+    "mean_return": 5714.0,
+    "model": "qwengr00t",
+    "path": "zero-latency/asterix/vla/starvla-qwengr00t-h1/asterix-qwengr00t-hist8-zero-s42-v1",
+    "repo_id": "latency-sensitive-bench/benchmark-models",
+    "revision": "d445334970ab7e110c3cacb5c8b790d3df20e1ec",
+    "run_id": "asterix-qwengr00t-hist8-zero-s42-v1",
+    "task": "asterix",
+    "training_status": "complete",
+    "zero_evaluation_status": "complete",
+    "zero_mean_return": 5714.0
+  },
+  "profile": {
+    "GPU_CLASS": "1x-rtx3090",
+    "INSTANCE_ID": "instance_d907cdaf680040e0",
+    "PROFILE_PATH": "results/profiling/profiles/qwengr00t/1x-rtx3090/asterix/instance_d907cdaf680040e0/profile.json",
+    "PROFILE_REF": "qwengr00t/1x-rtx3090/asterix/instance_d907cdaf680040e0",
+    "RUN_ID": "20260924T122426657466Z"
+  },
+  "profile_path": "profiles/qwengr00t/1x-rtx3090/asterix/instance_d907cdaf680040e0/profile.json",
+  "profile_repo": "latency-sensitive-bench/profiles",
+  "profile_revision": "15001405a402e30b7706476bc92aea45a61bb193",
+  "profile_sha256": "5e104064d718ac51ce56aefac5318a155fd099f5c714ff8f0bc4985509639fd1",
+  "task": "asterix"
+}

zero-latency/asterix/vla/starvla-qwengr00t-h1/asterix-qwengr00t-hist8-zero-s42-v1/publication_status.json

@@ -1,6 +1,8 @@
 {
   "episodes": 100,
-  "mean_return": 5714.0,
+  "mean_return": 182.0,
+  "profile_evaluation_status": "complete",
+  "profile_mean_return": 182.0,
   "training_status": "complete",
   "zero_evaluation_status": "complete",
   "zero_mean_return": 5714.0