Publish asterix qwenoft H8 evaluation artifacts

latency-aware/asterix/vla/starvla-qwenoft-h1/asterix-qwenoft-hist8-profile-s42-v1/README.md

@@ -4,4 +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/ac01280d
-- Formal profile evaluation: pending.
+- Evaluation: 100 episodes, seeds 1000000–1000099, GPU simulator, parallel 32, capacity 1, 60/15 Hz, 7200 raw-frame cap, profile latency
+- Mean raw return: 3653.5

latency-aware/asterix/vla/starvla-qwenoft-h1/asterix-qwenoft-hist8-profile-s42-v1/eval_profile.json

@@ -0,0 +1,217 @@
+{
+  "checkpoint_path": "/home/ubuntu/codex/atari-hist8-20260924/outputs/tasks/atari_hist8_v1/bundles/asterix-qwenoft-hist8-profile-s42-v1/checkpoints/model.pt",
+  "experiment_name": "asterix-qwenoft-hist8-profile-s42-v1-step5000-history01-profile-par32",
+  "latency": "profile_sample",
+  "latency_type": "profile_sample",
+  "lengths": [
+    4849,
+    3367,
+    3990,
+    5010,
+    4971,
+    882,
+    2309,
+    4592,
+    3873,
+    4147,
+    3319,
+    3205,
+    3512,
+    3912,
+    2945,
+    2934,
+    4714,
+    4058,
+    3658,
+    3970,
+    4407,
+    5094,
+    4032,
+    4334,
+    3340,
+    3351,
+    4384,
+    2722,
+    3403,
+    4687,
+    3339,
+    4050,
+    3761,
+    4302,
+    4617,
+    4172,
+    1463,
+    4609,
+    5052,
+    2382,
+    3898,
+    3369,
+    5535,
+    4609,
+    3749,
+    2703,
+    6070,
+    2850,
+    5660,
+    4476,
+    3053,
+    3918,
+    4416,
+    4130,
+    5079,
+    4196,
+    4344,
+    3797,
+    2154,
+    2241,
+    4141,
+    5410,
+    2554,
+    6125,
+    5702,
+    1895,
+    4111,
+    4181,
+    3847,
+    2098,
+    5342,
+    3845,
+    2981,
+    5259,
+    4421,
+    6471,
+    2047,
+    1121,
+    3905,
+    4163,
+    7200,
+    4345,
+    5255,
+    4456,
+    5133,
+    4228,
+    4697,
+    1794,
+    3497,
+    5691,
+    5916,
+    3914,
+    3090,
+    3540,
+    2142,
+    4511,
+    3212,
+    4183,
+    3069,
+    3798
+  ],
+  "mean_length": 3952.55,
+  "mean_return": 3653.5,
+  "returns": [
+    5200.0,
+    2900.0,
+    3000.0,
+    5700.0,
+    4700.0,
+    300.0,
+    1650.0,
+    4100.0,
+    2800.0,
+    3700.0,
+    2250.0,
+    2350.0,
+    2150.0,
+    3000.0,
+    2300.0,
+    2100.0,
+    4500.0,
+    3900.0,
+    3700.0,
+    3500.0,
+    4600.0,
+    4800.0,
+    4000.0,
+    4400.0,
+    3200.0,
+    2700.0,
+    3200.0,
+    1950.0,
+    3300.0,
+    3900.0,
+    2600.0,
+    3900.0,
+    4000.0,
+    4800.0,
+    4200.0,
+    3600.0,
+    750.0,
+    4200.0,
+    3900.0,
+    1550.0,
+    3800.0,
+    3000.0,
+    6100.0,
+    4500.0,
+    3300.0,
+    1850.0,
+    7200.0,
+    2100.0,
+    6700.0,
+    4300.0,
+    2500.0,
+    3600.0,
+    4400.0,
+    3600.0,
+    4400.0,
+    3600.0,
+    3900.0,
+    3000.0,
+    1700.0,
+    1750.0,
+    4400.0,
+    5300.0,
+    1800.0,
+    6400.0,
+    6300.0,
+    1250.0,
+    3700.0,
+    5000.0,
+    2450.0,
+    1350.0,
+    5600.0,
+    2800.0,
+    2200.0,
+    7200.0,
+    4100.0,
+    7300.0,
+    1400.0,
+    900.0,
+    3900.0,
+    2700.0,
+    10300.0,
+    4200.0,
+    4800.0,
+    4000.0,
+    5600.0,
+    4300.0,
+    4600.0,
+    1250.0,
+    3300.0,
+    5500.0,
+    5200.0,
+    4000.0,
+    1900.0,
+    3000.0,
+    1500.0,
+    4900.0,
+    2800.0,
+    3400.0,
+    2200.0,
+    3900.0
+  ],
+  "seed": 1000000,
+  "source_profile_path": "/home/ubuntu/codex/atari-hist8-20260924/outputs/tasks/atari_hist8_v1/teacher_configs/asterix-qwenoft/profile/profile.json",
+  "std_return": 1620.821628063989,
+  "suite_name": "profile_sample",
+  "timestamp_utc": "2026-09-24T17:18:28.394877+00:00"
+}

latency-aware/asterix/vla/starvla-qwenoft-h1/asterix-qwenoft-hist8-profile-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": 49405,
+  "action_counts": {
+    "2": 3308,
+    "8": 4664,
+    "1": 4419,
+    "5": 7523,
+    "0": 21887,
+    "7": 5753,
+    "6": 1626,
+    "4": 211,
+    "3": 14
+  },
+  "largest_action_fraction": 0.4430118409067908,
+  "raw_score_min": [
+    -13.119811058044434,
+    -10.853740692138672,
+    -11.006418228149414,
+    -12.597626686096191,
+    -9.706649780273438,
+    -8.856327056884766,
+    -9.544286727905273,
+    -7.748889446258545,
+    -13.909894943237305
+  ],
+  "raw_score_max": [
+    21.067642211914062,
+    21.027864456176758,
+    20.085676193237305,
+    17.18351173400879,
+    15.032642364501953,
+    21.076366424560547,
+    18.5609130859375,
+    19.459716796875,
+    21.263235092163086
+  ],
+  "raw_score_mean": [
+    6.1028937123558284,
+    0.0709031725423172,
+    0.07752298349673893,
+    -5.34277075081747,
+    -2.9248927316361186,
+    2.2591020215939106,
+    -0.04072495360855251,
+    2.744448609118694,
+    -3.116156193460523
+  ]
+}

latency-aware/asterix/vla/starvla-qwenoft-h1/asterix-qwenoft-hist8-profile-s42-v1/evaluation_profile/eval_config.yaml

@@ -0,0 +1,165 @@
+experiment:
+  name: asterix-qwenoft-hist8-profile-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-qwenoft/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-qwenoft-hist8-profile-s42-v1/checkpoints/model.pt
+  model_config_path: /home/ubuntu/codex/atari-hist8-20260924/outputs/tasks/atari_hist8_v1/bundles/asterix-qwenoft-hist8-profile-s42-v1/config.yaml
+  task_manifest_path: /home/ubuntu/codex/atari-hist8-20260924/outputs/tasks/atari_hist8_v1/bundles/asterix-qwenoft-hist8-profile-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-qwenoft-hist8-profile-s42-v1/profile_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: ''

latency-aware/asterix/vla/starvla-qwenoft-h1/asterix-qwenoft-hist8-profile-s42-v1/evaluation_profile/provenance.json

@@ -0,0 +1,46 @@
+{
+  "collection": {
+    "teacher_repo": "latency-sensitive-bench/benchmark-models",
+    "teacher_revision": "da734589fb8669108a8021ea79f823665e1b11c8",
+    "teacher_path": "latency-aware/asterix/small-policy/sample-factory-qwenoft-hist8-cpu15m-s0-v1",
+    "teacher_checkpoint": "best_000027136_13893632_reward_4562.500.pth",
+    "teacher_sha256": "93ba21325d92fa47f617c3daaffd9d7604a8c93bf8fe4bc1d4aacc4facd5430c",
+    "profile_path": "/home/ubuntu/codex/atari-hist8-20260924/results/profiling/profiles/qwenoft/1x-rtx3090/asterix/instance_d907cdaf680040e0/profile.json",
+    "profile_sha256": "4e2e3ed923c7500248ce4ce1508f124faec78282d90d6b99d070a9e29bdb6146",
+    "profile_repo": "latency-sensitive-bench/profiles",
+    "profile_revision": "fbbba76b21fa705cf4e5d05ec1ecfb82a88b12d8",
+    "rollout_overrides": {
+      "latency_profile_path": "/home/ubuntu/codex/atari-hist8-20260924/results/profiling/profiles/qwenoft/1x-rtx3090/asterix/instance_d907cdaf680040e0/profile.json",
+      "seed": 42,
+      "simulator": "gpu",
+      "probe_envs": 256,
+      "replay_envs": 256,
+      "policy_inference_batch_size": 256,
+      "max_episode_raw_frames": 7200
+    },
+    "episodes": 350,
+    "filter_preset": "none",
+    "history": "decision_v1"
+  },
+  "training_run_id": "asterix-qwenoft-hist8-profile-s42-v1",
+  "training_step": 5000,
+  "global_batch": 128,
+  "base_model": "Qwen/Qwen3-VL-4B-Instruct",
+  "wandb_run_id": "ac01280d",
+  "checkpoint_sha256": "b96bf1398d3d767b13be1cc76eefdfe70fdab349231ef5a4636495c548fe09ac",
+  "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": "4e2e3ed923c7500248ce4ce1508f124faec78282d90d6b99d070a9e29bdb6146"
+}

latency-aware/asterix/vla/starvla-qwenoft-h1/asterix-qwenoft-hist8-profile-s42-v1/profile_publication.json

@@ -0,0 +1,28 @@
+{
+  "model": "qwenoft",
+  "model_publication": {
+    "checkpoint_sha256": "9d5be1187e768a3ba6638610ad70ca58fed54a531b77b0b0dea3a33538def572",
+    "episodes": 100,
+    "mean_return": 5803.5,
+    "model": "qwenoft",
+    "path": "zero-latency/asterix/vla/starvla-qwenoft-h1/asterix-qwenoft-hist8-zero-s42-v1",
+    "repo_id": "latency-sensitive-bench/benchmark-models",
+    "revision": "15c9cd4ab12215ccbcef74a8cd4d051d8e3aae3f",
+    "run_id": "asterix-qwenoft-hist8-zero-s42-v1",
+    "task": "asterix",
+    "training_status": "complete",
+    "zero_evaluation_status": "complete"
+  },
+  "profile": {
+    "GPU_CLASS": "1x-rtx3090",
+    "INSTANCE_ID": "instance_d907cdaf680040e0",
+    "PROFILE_PATH": "results/profiling/profiles/qwenoft/1x-rtx3090/asterix/instance_d907cdaf680040e0/profile.json",
+    "PROFILE_REF": "qwenoft/1x-rtx3090/asterix/instance_d907cdaf680040e0",
+    "RUN_ID": "20260924T110210301668Z"
+  },
+  "profile_path": "profiles/qwenoft/1x-rtx3090/asterix/instance_d907cdaf680040e0/profile.json",
+  "profile_repo": "latency-sensitive-bench/profiles",
+  "profile_revision": "fbbba76b21fa705cf4e5d05ec1ecfb82a88b12d8",
+  "profile_sha256": "4e2e3ed923c7500248ce4ce1508f124faec78282d90d6b99d070a9e29bdb6146",
+  "task": "asterix"
+}

latency-aware/asterix/vla/starvla-qwenoft-h1/asterix-qwenoft-hist8-profile-s42-v1/publication_status.json

@@ -1,4 +1,7 @@
 {
-  "profile_evaluation_status": "pending",
+  "episodes": 100,
+  "mean_return": 3653.5,
+  "profile_evaluation_status": "complete",
+  "profile_mean_return": 3653.5,
   "training_status": "complete"
 }