Upload folder using huggingface_hub

flappy_fix_latency_2_200ep_last_8_layers_corrected/checkpoints/_initialization/StarVLA_Qwen3VL-OFT-Bridge-RT-1/checkpoints/steps_5000_pytorch_model.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:371cb744227687bb99bcad7f9ff2250cf06da75631359ad3eba4c6bc52570607
+size 9785060316

flappy_fix_latency_2_200ep_last_8_layers_corrected/checkpoints/steps_5000_state/model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:750786e2eac8b455ab6e8232416d692e8ad1d5b7a56d386b40bf53a9462ba436
+size 9138230516

flappy_fix_latency_2_200ep_last_8_layers_corrected/checkpoints/steps_5000_state/optimizer.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:faf46b8bdc9bc58183fc4ce41be09e44623fd2123c316a971fe45790ee8a1403
+size 6972357886

flappy_fix_latency_2_200ep_last_8_layers_corrected/checkpoints/steps_5000_state/random_states_0.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1534f5654cb4caf72c5fc3843a84809a4e22f60b4fe5d14f50ccfafdc9d6f3da
+size 14821

flappy_fix_latency_2_200ep_last_8_layers_corrected/config.full.yaml

@@ -0,0 +1,219 @@
+framework:
+  name: QwenOFT
+  qwenvl:
+    base_vlm: /workspace/latency-sensitive-bench/starVLA/playground/Pretrained_models/Qwen3-VL-4B-Instruct
+    attn_implementation: flash_attention_2
+    enable_gradient_checkpointing: true
+  action_model:
+    action_model_type: MLP
+    action_dim: 7
+    action_hidden_dim: 2560
+    future_action_window_size: 0
+    past_action_window_size: 0
+    loss_type: discrete_ce
+    state_dim: 7
+    action_horizon: 1
+    action_env_dim: 2
+datasets:
+  vla_data:
+    dataset_py: lerobot_datasets
+    include_state: true
+    data_root_dir: /workspace/latency-sensitive-bench/starVLA/data/flappy_fix_latency_2_200ep
+    data_mix: flappy_train__bridge
+    eval_data_mix: flappy_train__bridge__val
+    custom_mixtures_path: null
+    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
+    per_device_batch_size: 64
+    load_all_data_for_training: true
+    num_obs_frames: 1
+    image_mode: single
+    stitch_grid:
+    - 2
+    - 2
+    obs_image_size: null
+    video_backend: torchvision_av
+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
+  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: playground/Pretrained_models/Qwen3VL-OFT-Bridge-RT-1
+  checkpoint_hf_repo_id: StarVLA/Qwen3VL-OFT-Bridge-RT-1
+  checkpoint_filename: checkpoints/steps_5000_pytorch_model.pt
+trainer:
+  max_train_steps: 5000
+  num_warmup_steps: 100
+  save_interval: 500
+  eval_interval: 100
+  eval_num_batches: 100
+  per_latency_eval_num_batches: null
+  eval_action_classification: true
+  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_llm_bottom_ratio: 0.7778
+  loss_scale:
+    vla: 1.0
+    vlm: 0.1
+  max_grad_norm: 1.0
+  weight_decay: 0.0
+  logging_frequency: 1
+  gradient_clipping: 1.0
+  gradient_accumulation_steps: 2
+  distributed_backend: none
+  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: /workspace/latency-sensitive-bench/starVLA/results/Checkpoints
+seed: 42
+wandb_entity: saberrr-zju
+wandb_project: starVLA_rl_games
+auth:
+  env_file: null
+  hf_token_env: HF_TOKEN
+  wandb_api_key_env: WANDB_API_KEY
+paths:
+  run_root_dir: results/Checkpoints
+  dataset_local_dir: playground/Datasets/rl_games
+  dataset_cache_dir: null
+  base_model_dir: playground/Pretrained_models/Qwen3-VL-4B-Instruct
+  accelerate_config: starVLA/config/deepseeds/deepspeed_zero2.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
+    seed: 42
+    fixed_episode_seeds: true
+    latency_seed_stride: 0
+    task_seed_stride: 0
+    task_description: ''
+    enabled: true
+    distributed_mode: none
+    vectorized:
+      enabled: false
+      batch_size: 1
+    latency:
+      prompt_map_path: /workspace/latency-sensitive-bench/starVLA/data/flappy_fix_latency_2_200ep/flappy_train__bridge/latency_prompt_map.json
+      mode: single
+      values:
+      - 0
+    mid_train:
+      enabled: true
+      interval_steps: 250
+      latencies:
+      - 2
+      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
+  task: flappy
+  initialization_mode: bridge
+  action_carrier: bridge
+model: openvla
+env: flappy
+init: bridge
+bridge_base_model:
+  repo_id:
+    openvla: Qwen/Qwen3-VL-4B-Instruct
+    pi0: StarVLA/Qwen2.5-VL-3B-Instruct-Action
+    pi05: Qwen/Qwen3-VL-4B-Instruct
+    gr00t: Qwen/Qwen3-VL-4B-Instruct
+  local_dir:
+    openvla: playground/Pretrained_models/Qwen3-VL-4B-Instruct
+    pi0: playground/Pretrained_models/Qwen2.5-VL-3B-Instruct-Action
+    pi05: playground/Pretrained_models/Qwen3-VL-4B-Instruct
+    gr00t: playground/Pretrained_models/Qwen3-VL-4B-Instruct
+mode: single
+checkpoint:
+  load: auto
+  hf_repo_id: null
+  save_best_model: false
+  save_pt_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: flappy_fix_latency_2_200ep_last_8_layers_corrected
+output_dir: /workspace/latency-sensitive-bench/starVLA/results/Checkpoints/flappy_fix_latency_2_200ep_last_8_layers_corrected
+config_yaml: null
+is_debug: false
+version_id: '0.21'

flappy_fix_latency_2_200ep_last_8_layers_corrected/config.yaml

@@ -0,0 +1,99 @@
+checkpoint:
+  local:
+    keep_last_n: 1
+  save_best_model: false
+  save_pt_file: false
+  sync:
+    enabled: false
+    keep_last_n: 0
+    repo_id: null
+datasets:
+  vla_data:
+    data_mix: flappy_train__bridge
+    dataset_py: lerobot_datasets
+    eval_data_mix: flappy_train__bridge__val
+    eval_num_workers: 8
+    include_state: true
+    latency_curriculum:
+      enabled: false
+    obs_image_size: null
+    per_device_batch_size: 64
+    persistent_workers: true
+    pin_memory: true
+    prefetch_factor: 4
+framework:
+  action_model:
+    action_dim: 7
+    action_env_dim: 2
+    action_hidden_dim: 2560
+    action_horizon: 1
+    action_model_type: MLP
+    loss_type: discrete_ce
+    state_dim: 7
+  name: QwenOFT
+  qwenvl:
+    attn_implementation: flash_attention_2
+    base_vlm: /workspace/latency-sensitive-bench/starVLA/playground/Pretrained_models/Qwen3-VL-4B-Instruct
+    enable_gradient_checkpointing: true
+output_dir: /workspace/latency-sensitive-bench/starVLA/results/Checkpoints/flappy_fix_latency_2_200ep_last_8_layers_corrected
+rl_games:
+  env_eval:
+    distributed_mode: none
+    enabled: true
+    fixed_episode_seeds: true
+    frameskip: 1
+    image_size: 224
+    latency:
+      prompt_map_path: /workspace/latency-sensitive-bench/starVLA/data/flappy_fix_latency_2_200ep/flappy_train__bridge/latency_prompt_map.json
+    latency_seed_stride: 0
+    mid_train:
+      enabled: true
+      interval_steps: 250
+      latencies:
+      - 2
+      max_steps_per_episode: 3600
+      num_episodes: 5
+    seed: 42
+    task_description: ''
+    task_seed_stride: 0
+    vectorized:
+      enabled: false
+  model_alias: openvla
+  task: flappy
+run_id: flappy_fix_latency_2_200ep_last_8_layers_corrected
+run_root_dir: /workspace/latency-sensitive-bench/starVLA/results/Checkpoints
+seed: 42
+trainer:
+  distributed_backend: none
+  eval_action_classification: true
+  eval_action_classification_interval: null
+  eval_interval: 100
+  eval_num_batches: 100
+  freeze_llm_bottom_ratio: 0.7778
+  freeze_modules: ''
+  gradient_accumulation_steps: 2
+  gradient_clipping: 1.0
+  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
+  reload_modules: null
+  save_interval: 500
+  scheduler_specific_kwargs:
+    min_lr: 1.0e-06
+wandb_entity: saberrr-zju
+wandb_project: starVLA_rl_games

flappy_fix_latency_2_200ep_last_8_layers_corrected/dataset_statistics.json

@@ -0,0 +1,127 @@
+{
+  "new_embodiment": {
+    "action": {
+      "mean": [
+        0.6028500199317932,
+        0.3971499800682068,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0
+      ],
+      "std": [
+        0.4890792667865753,
+        0.4890792667865753,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0
+      ],
+      "max": [
+        1.0,
+        1.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0
+      ],
+      "min": [
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0
+      ],
+      "q01": [
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0
+      ],
+      "q99": [
+        1.0,
+        1.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0
+      ],
+      "mask": [
+        true,
+        true,
+        true,
+        true,
+        true,
+        true,
+        true
+      ]
+    },
+    "state": {
+      "mean": [
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0
+      ],
+      "std": [
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0
+      ],
+      "max": [
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0
+      ],
+      "min": [
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0
+      ],
+      "q01": [
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0
+      ],
+      "q99": [
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0
+      ]
+    },
+    "num_transitions": 330734,
+    "num_trajectories": 180
+  }
+}

flappy_fix_latency_2_200ep_last_8_layers_corrected/dataset_statistics_eval.json

@@ -0,0 +1,127 @@
+{
+  "new_embodiment": {
+    "action": {
+      "mean": [
+        0.7959861159324646,
+        0.2040138840675354,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0
+      ],
+      "std": [
+        0.4030573070049286,
+        0.4030573070049286,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0
+      ],
+      "max": [
+        1.0,
+        1.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0
+      ],
+      "min": [
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0
+      ],
+      "q01": [
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0
+      ],
+      "q99": [
+        1.0,
+        1.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0
+      ],
+      "mask": [
+        true,
+        true,
+        true,
+        true,
+        true,
+        true,
+        true
+      ]
+    },
+    "state": {
+      "mean": [
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0
+      ],
+      "std": [
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0
+      ],
+      "max": [
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0
+      ],
+      "min": [
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0
+      ],
+      "q01": [
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0
+      ],
+      "q99": [
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0
+      ]
+    },
+    "num_transitions": 72000,
+    "num_trajectories": 20
+  }
+}

flappy_fix_latency_2_200ep_last_8_layers_corrected/eval/mid_train/step_1000.json

@@ -0,0 +1,63 @@
+{
+  "per_latency": {
+    "flappy/latency_2": {
+      "latency": 2,
+      "num_episodes": 5,
+      "mean_reward": 23.40000000000016,
+      "mean_length": 207.2,
+      "std_reward": 23.92488244485246,
+      "std_length": 192.86617121724586,
+      "episode_rewards": [
+        54.90000000000042,
+        50.40000000000037,
+        3.8999999999999986,
+        3.8999999999999986,
+        3.8999999999999986
+      ],
+      "episode_lengths": [
+        461,
+        425,
+        50,
+        50,
+        50
+      ],
+      "decoded_action_hist": {
+        "0": 837,
+        "1": 199
+      },
+      "fixed_episode_seeds": true,
+      "eval_seed": 42,
+      "episode_seeds": [
+        42,
+        43,
+        44,
+        45,
+        46
+      ],
+      "episode_indices": [
+        0,
+        1,
+        2,
+        3,
+        4
+      ]
+    }
+  },
+  "aggregate": {
+    "stage": "mid_train",
+    "step": 1000,
+    "task": "flappy",
+    "model_alias": "openvla",
+    "fixed_episode_seeds": true,
+    "eval_seed": 42,
+    "total_episodes": 5,
+    "mean_reward": 23.40000000000016,
+    "mean_length": 207.2,
+    "std_reward": 23.92488244485246,
+    "std_length": 192.86617121724586,
+    "task_count": 1,
+    "macro_mean_reward": 23.40000000000016,
+    "macro_mean_length": 207.2,
+    "distributed_eval": false
+  }
+}

flappy_fix_latency_2_200ep_last_8_layers_corrected/eval/mid_train/step_1250.json

@@ -0,0 +1,63 @@
+{
+  "per_latency": {
+    "flappy/latency_2": {
+      "latency": 2,
+      "num_episodes": 5,
+      "mean_reward": 108.73999999999823,
+      "mean_length": 896.8,
+      "std_reward": 97.07412837620372,
+      "std_length": 783.4583332890141,
+      "episode_rewards": [
+        4.999999999999995,
+        275.2999999999956,
+        124.89999999999772,
+        120.39999999999792,
+        18.100000000000012
+      ],
+      "episode_lengths": [
+        61,
+        2242,
+        1026,
+        990,
+        165
+      ],
+      "decoded_action_hist": {
+        "0": 3568,
+        "1": 916
+      },
+      "fixed_episode_seeds": true,
+      "eval_seed": 42,
+      "episode_seeds": [
+        42,
+        43,
+        44,
+        45,
+        46
+      ],
+      "episode_indices": [
+        0,
+        1,
+        2,
+        3,
+        4
+      ]
+    }
+  },
+  "aggregate": {
+    "stage": "mid_train",
+    "step": 1250,
+    "task": "flappy",
+    "model_alias": "openvla",
+    "fixed_episode_seeds": true,
+    "eval_seed": 42,
+    "total_episodes": 5,
+    "mean_reward": 108.73999999999823,
+    "mean_length": 896.8,
+    "std_reward": 97.07412837620372,
+    "std_length": 783.4583332890141,
+    "task_count": 1,
+    "macro_mean_reward": 108.73999999999823,
+    "macro_mean_length": 896.8,
+    "distributed_eval": false
+  }
+}

flappy_fix_latency_2_200ep_last_8_layers_corrected/eval/mid_train/step_1500.json

@@ -0,0 +1,63 @@
+{
+  "per_latency": {
+    "flappy/latency_2": {
+      "latency": 2,
+      "num_episodes": 5,
+      "mean_reward": 211.0200000000097,
+      "mean_length": 1717.2,
+      "std_reward": 195.658277616882,
+      "std_length": 1577.5296383903537,
+      "episode_rewards": [
+        6.799999999999992,
+        444.60000000002566,
+        134.39999999999728,
+        444.60000000002566,
+        24.700000000000077
+      ],
+      "episode_lengths": [
+        70,
+        3600,
+        1103,
+        3600,
+        213
+      ],
+      "decoded_action_hist": {
+        "0": 6875,
+        "1": 1711
+      },
+      "fixed_episode_seeds": true,
+      "eval_seed": 42,
+      "episode_seeds": [
+        42,
+        43,
+        44,
+        45,
+        46
+      ],
+      "episode_indices": [
+        0,
+        1,
+        2,
+        3,
+        4
+      ]
+    }
+  },
+  "aggregate": {
+    "stage": "mid_train",
+    "step": 1500,
+    "task": "flappy",
+    "model_alias": "openvla",
+    "fixed_episode_seeds": true,
+    "eval_seed": 42,
+    "total_episodes": 5,
+    "mean_reward": 211.0200000000097,
+    "mean_length": 1717.2,
+    "std_reward": 195.658277616882,
+    "std_length": 1577.5296383903537,
+    "task_count": 1,
+    "macro_mean_reward": 211.0200000000097,
+    "macro_mean_length": 1717.2,
+    "distributed_eval": false
+  }
+}

flappy_fix_latency_2_200ep_last_8_layers_corrected/eval/mid_train/step_1750.json

@@ -0,0 +1,63 @@
+{
+  "per_latency": {
+    "flappy/latency_2": {
+      "latency": 2,
+      "num_episodes": 5,
+      "mean_reward": 361.70000000001505,
+      "mean_length": 2934.2,
+      "std_reward": 106.76950875602273,
+      "std_length": 859.2597744570613,
+      "episode_rewards": [
+        444.60000000002566,
+        417.00000000002075,
+        444.60000000002566,
+        339.9000000000071,
+        162.39999999999603
+      ],
+      "episode_lengths": [
+        3600,
+        3380,
+        3600,
+        2762,
+        1329
+      ],
+      "decoded_action_hist": {
+        "0": 11737,
+        "1": 2934
+      },
+      "fixed_episode_seeds": true,
+      "eval_seed": 42,
+      "episode_seeds": [
+        42,
+        43,
+        44,
+        45,
+        46
+      ],
+      "episode_indices": [
+        0,
+        1,
+        2,
+        3,
+        4
+      ]
+    }
+  },
+  "aggregate": {
+    "stage": "mid_train",
+    "step": 1750,
+    "task": "flappy",
+    "model_alias": "openvla",
+    "fixed_episode_seeds": true,
+    "eval_seed": 42,
+    "total_episodes": 5,
+    "mean_reward": 361.70000000001505,
+    "mean_length": 2934.2,
+    "std_reward": 106.76950875602273,
+    "std_length": 859.2597744570613,
+    "task_count": 1,
+    "macro_mean_reward": 361.70000000001505,
+    "macro_mean_length": 2934.2,
+    "distributed_eval": false
+  }
+}

flappy_fix_latency_2_200ep_last_8_layers_corrected/eval/mid_train/step_2000.json

@@ -0,0 +1,63 @@
+{
+  "per_latency": {
+    "flappy/latency_2": {
+      "latency": 2,
+      "num_episodes": 5,
+      "mean_reward": 115.8999999999974,
+      "mean_length": 954.0,
+      "std_reward": 81.98780397107578,
+      "std_length": 661.8280743516401,
+      "episode_rewards": [
+        115.89999999999812,
+        236.8999999999927,
+        167.89999999999577,
+        3.8999999999999986,
+        54.90000000000042
+      ],
+      "episode_lengths": [
+        954,
+        1930,
+        1375,
+        50,
+        461
+      ],
+      "decoded_action_hist": {
+        "0": 3834,
+        "1": 936
+      },
+      "fixed_episode_seeds": true,
+      "eval_seed": 42,
+      "episode_seeds": [
+        42,
+        43,
+        44,
+        45,
+        46
+      ],
+      "episode_indices": [
+        0,
+        1,
+        2,
+        3,
+        4
+      ]
+    }
+  },
+  "aggregate": {
+    "stage": "mid_train",
+    "step": 2000,
+    "task": "flappy",
+    "model_alias": "openvla",
+    "fixed_episode_seeds": true,
+    "eval_seed": 42,
+    "total_episodes": 5,
+    "mean_reward": 115.8999999999974,
+    "mean_length": 954.0,
+    "std_reward": 81.98780397107578,
+    "std_length": 661.8280743516401,
+    "task_count": 1,
+    "macro_mean_reward": 115.8999999999974,
+    "macro_mean_length": 954.0,
+    "distributed_eval": false
+  }
+}

flappy_fix_latency_2_200ep_last_8_layers_corrected/eval/mid_train/step_2250.json

@@ -0,0 +1,63 @@
+{
+  "per_latency": {
+    "flappy/latency_2": {
+      "latency": 2,
+      "num_episodes": 5,
+      "mean_reward": 219.8400000000019,
+      "mean_length": 1791.4,
+      "std_reward": 122.89391522773798,
+      "std_length": 989.3716389709177,
+      "episode_rewards": [
+        73.89999999999999,
+        199.89999999999435,
+        218.39999999999353,
+        444.60000000002566,
+        162.39999999999603
+      ],
+      "episode_lengths": [
+        615,
+        1632,
+        1781,
+        3600,
+        1329
+      ],
+      "decoded_action_hist": {
+        "0": 7114,
+        "1": 1843
+      },
+      "fixed_episode_seeds": true,
+      "eval_seed": 42,
+      "episode_seeds": [
+        42,
+        43,
+        44,
+        45,
+        46
+      ],
+      "episode_indices": [
+        0,
+        1,
+        2,
+        3,
+        4
+      ]
+    }
+  },
+  "aggregate": {
+    "stage": "mid_train",
+    "step": 2250,
+    "task": "flappy",
+    "model_alias": "openvla",
+    "fixed_episode_seeds": true,
+    "eval_seed": 42,
+    "total_episodes": 5,
+    "mean_reward": 219.8400000000019,
+    "mean_length": 1791.4,
+    "std_reward": 122.89391522773798,
+    "std_length": 989.3716389709177,
+    "task_count": 1,
+    "macro_mean_reward": 219.8400000000019,
+    "macro_mean_length": 1791.4,
+    "distributed_eval": false
+  }
+}

flappy_fix_latency_2_200ep_last_8_layers_corrected/eval/mid_train/step_250.json

@@ -0,0 +1,63 @@
+{
+  "per_latency": {
+    "flappy/latency_2": {
+      "latency": 2,
+      "num_episodes": 5,
+      "mean_reward": 3.2799999999999985,
+      "mean_length": 42.0,
+      "std_reward": 2.559999999999994,
+      "std_length": 22.0,
+      "episode_rewards": [
+        8.399999999999986,
+        2.0000000000000013,
+        2.0000000000000013,
+        2.0000000000000013,
+        2.0000000000000013
+      ],
+      "episode_lengths": [
+        86,
+        31,
+        31,
+        31,
+        31
+      ],
+      "decoded_action_hist": {
+        "0": 193,
+        "1": 17
+      },
+      "fixed_episode_seeds": true,
+      "eval_seed": 42,
+      "episode_seeds": [
+        42,
+        43,
+        44,
+        45,
+        46
+      ],
+      "episode_indices": [
+        0,
+        1,
+        2,
+        3,
+        4
+      ]
+    }
+  },
+  "aggregate": {
+    "stage": "mid_train",
+    "step": 250,
+    "task": "flappy",
+    "model_alias": "openvla",
+    "fixed_episode_seeds": true,
+    "eval_seed": 42,
+    "total_episodes": 5,
+    "mean_reward": 3.2799999999999985,
+    "mean_length": 42.0,
+    "std_reward": 2.559999999999994,
+    "std_length": 22.0,
+    "task_count": 1,
+    "macro_mean_reward": 3.2799999999999985,
+    "macro_mean_length": 42.0,
+    "distributed_eval": false
+  }
+}

flappy_fix_latency_2_200ep_last_8_layers_corrected/eval/mid_train/step_2500.json

@@ -0,0 +1,63 @@
+{
+  "per_latency": {
+    "flappy/latency_2": {
+      "latency": 2,
+      "num_episodes": 5,
+      "mean_reward": 194.76000000001028,
+      "mean_length": 1587.0,
+      "std_reward": 205.5255176371176,
+      "std_length": 1656.0664237886112,
+      "episode_rewards": [
+        6.799999999999992,
+        444.60000000002566,
+        444.60000000002566,
+        3.8999999999999986,
+        73.89999999999999
+      ],
+      "episode_lengths": [
+        70,
+        3600,
+        3600,
+        50,
+        615
+      ],
+      "decoded_action_hist": {
+        "0": 6337,
+        "1": 1598
+      },
+      "fixed_episode_seeds": true,
+      "eval_seed": 42,
+      "episode_seeds": [
+        42,
+        43,
+        44,
+        45,
+        46
+      ],
+      "episode_indices": [
+        0,
+        1,
+        2,
+        3,
+        4
+      ]
+    }
+  },
+  "aggregate": {
+    "stage": "mid_train",
+    "step": 2500,
+    "task": "flappy",
+    "model_alias": "openvla",
+    "fixed_episode_seeds": true,
+    "eval_seed": 42,
+    "total_episodes": 5,
+    "mean_reward": 194.76000000001028,
+    "mean_length": 1587.0,
+    "std_reward": 205.5255176371176,
+    "std_length": 1656.0664237886112,
+    "task_count": 1,
+    "macro_mean_reward": 194.76000000001028,
+    "macro_mean_length": 1587.0,
+    "distributed_eval": false
+  }
+}

flappy_fix_latency_2_200ep_last_8_layers_corrected/eval/mid_train/step_2750.json

@@ -0,0 +1,63 @@
+{
+  "per_latency": {
+    "flappy/latency_2": {
+      "latency": 2,
+      "num_episodes": 5,
+      "mean_reward": 123.74000000000481,
+      "mean_length": 1015.8,
+      "std_reward": 165.59867873870033,
+      "std_length": 1333.9035047558725,
+      "episode_rewards": [
+        115.89999999999812,
+        444.60000000002566,
+        3.8999999999999986,
+        3.8999999999999986,
+        50.40000000000037
+      ],
+      "episode_lengths": [
+        954,
+        3600,
+        50,
+        50,
+        425
+      ],
+      "decoded_action_hist": {
+        "0": 4045,
+        "1": 1034
+      },
+      "fixed_episode_seeds": true,
+      "eval_seed": 42,
+      "episode_seeds": [
+        42,
+        43,
+        44,
+        45,
+        46
+      ],
+      "episode_indices": [
+        0,
+        1,
+        2,
+        3,
+        4
+      ]
+    }
+  },
+  "aggregate": {
+    "stage": "mid_train",
+    "step": 2750,
+    "task": "flappy",
+    "model_alias": "openvla",
+    "fixed_episode_seeds": true,
+    "eval_seed": 42,
+    "total_episodes": 5,
+    "mean_reward": 123.74000000000481,
+    "mean_length": 1015.8,
+    "std_reward": 165.59867873870033,
+    "std_length": 1333.9035047558725,
+    "task_count": 1,
+    "macro_mean_reward": 123.74000000000481,
+    "macro_mean_length": 1015.8,
+    "distributed_eval": false
+  }
+}

flappy_fix_latency_2_200ep_last_8_layers_corrected/eval/mid_train/step_3000.json

@@ -0,0 +1,63 @@
+{
+  "per_latency": {
+    "flappy/latency_2": {
+      "latency": 2,
+      "num_episodes": 5,
+      "mean_reward": 300.4800000000097,
+      "mean_length": 2440.8,
+      "std_reward": 141.44521766395357,
+      "std_length": 1138.7999648753068,
+      "episode_rewards": [
+        73.89999999999999,
+        444.60000000002566,
+        218.39999999999353,
+        444.60000000002566,
+        320.9000000000037
+      ],
+      "episode_lengths": [
+        615,
+        3600,
+        1781,
+        3600,
+        2608
+      ],
+      "decoded_action_hist": {
+        "0": 9727,
+        "1": 2477
+      },
+      "fixed_episode_seeds": true,
+      "eval_seed": 42,
+      "episode_seeds": [
+        42,
+        43,
+        44,
+        45,
+        46
+      ],
+      "episode_indices": [
+        0,
+        1,
+        2,
+        3,
+        4
+      ]
+    }
+  },
+  "aggregate": {
+    "stage": "mid_train",
+    "step": 3000,
+    "task": "flappy",
+    "model_alias": "openvla",
+    "fixed_episode_seeds": true,
+    "eval_seed": 42,
+    "total_episodes": 5,
+    "mean_reward": 300.4800000000097,
+    "mean_length": 2440.8,
+    "std_reward": 141.44521766395357,
+    "std_length": 1138.7999648753068,
+    "task_count": 1,
+    "macro_mean_reward": 300.4800000000097,
+    "macro_mean_length": 2440.8,
+    "distributed_eval": false
+  }
+}

flappy_fix_latency_2_200ep_last_8_layers_corrected/eval/mid_train/step_3250.json

@@ -0,0 +1,63 @@
+{
+  "per_latency": {
+    "flappy/latency_2": {
+      "latency": 2,
+      "num_episodes": 5,
+      "mean_reward": 236.180000000009,
+      "mean_length": 1922.0,
+      "std_reward": 183.11908038215006,
+      "std_length": 1474.810496301135,
+      "episode_rewards": [
+        73.89999999999999,
+        213.89999999999372,
+        3.8999999999999986,
+        444.60000000002566,
+        444.60000000002566
+      ],
+      "episode_lengths": [
+        615,
+        1745,
+        50,
+        3600,
+        3600
+      ],
+      "decoded_action_hist": {
+        "0": 7674,
+        "1": 1936
+      },
+      "fixed_episode_seeds": true,
+      "eval_seed": 42,
+      "episode_seeds": [
+        42,
+        43,
+        44,
+        45,
+        46
+      ],
+      "episode_indices": [
+        0,
+        1,
+        2,
+        3,
+        4
+      ]
+    }
+  },
+  "aggregate": {
+    "stage": "mid_train",
+    "step": 3250,
+    "task": "flappy",
+    "model_alias": "openvla",
+    "fixed_episode_seeds": true,
+    "eval_seed": 42,
+    "total_episodes": 5,
+    "mean_reward": 236.180000000009,
+    "mean_length": 1922.0,
+    "std_reward": 183.11908038215006,
+    "std_length": 1474.810496301135,
+    "task_count": 1,
+    "macro_mean_reward": 236.180000000009,
+    "macro_mean_length": 1922.0,
+    "distributed_eval": false
+  }
+}

flappy_fix_latency_2_200ep_last_8_layers_corrected/eval/mid_train/step_3500.json

@@ -0,0 +1,63 @@
+{
+  "per_latency": {
+    "flappy/latency_2": {
+      "latency": 2,
+      "num_episodes": 5,
+      "mean_reward": 161.04000000000397,
+      "mean_length": 1316.8,
+      "std_reward": 151.0126431793143,
+      "std_length": 1216.3360390944601,
+      "episode_rewards": [
+        115.89999999999812,
+        22.40000000000006,
+        171.8999999999956,
+        444.60000000002566,
+        50.40000000000037
+      ],
+      "episode_lengths": [
+        954,
+        199,
+        1406,
+        3600,
+        425
+      ],
+      "decoded_action_hist": {
+        "0": 5261,
+        "1": 1323
+      },
+      "fixed_episode_seeds": true,
+      "eval_seed": 42,
+      "episode_seeds": [
+        42,
+        43,
+        44,
+        45,
+        46
+      ],
+      "episode_indices": [
+        0,
+        1,
+        2,
+        3,
+        4
+      ]
+    }
+  },
+  "aggregate": {
+    "stage": "mid_train",
+    "step": 3500,
+    "task": "flappy",
+    "model_alias": "openvla",
+    "fixed_episode_seeds": true,
+    "eval_seed": 42,
+    "total_episodes": 5,
+    "mean_reward": 161.04000000000397,
+    "mean_length": 1316.8,
+    "std_reward": 151.0126431793143,
+    "std_length": 1216.3360390944601,
+    "task_count": 1,
+    "macro_mean_reward": 161.04000000000397,
+    "macro_mean_length": 1316.8,
+    "distributed_eval": false
+  }
+}

flappy_fix_latency_2_200ep_last_8_layers_corrected/eval/mid_train/step_3750.json

@@ -0,0 +1,63 @@
+{
+  "per_latency": {
+    "flappy/latency_2": {
+      "latency": 2,
+      "num_episodes": 5,
+      "mean_reward": 105.37999999999798,
+      "mean_length": 868.6,
+      "std_reward": 77.50794539916251,
+      "std_length": 626.177802225534,
+      "episode_rewards": [
+        50.40000000000037,
+        20.800000000000036,
+        65.40000000000036,
+        227.8999999999931,
+        162.39999999999603
+      ],
+      "episode_lengths": [
+        425,
+        183,
+        548,
+        1858,
+        1329
+      ],
+      "decoded_action_hist": {
+        "0": 3464,
+        "1": 879
+      },
+      "fixed_episode_seeds": true,
+      "eval_seed": 42,
+      "episode_seeds": [
+        42,
+        43,
+        44,
+        45,
+        46
+      ],
+      "episode_indices": [
+        0,
+        1,
+        2,
+        3,
+        4
+      ]
+    }
+  },
+  "aggregate": {
+    "stage": "mid_train",
+    "step": 3750,
+    "task": "flappy",
+    "model_alias": "openvla",
+    "fixed_episode_seeds": true,
+    "eval_seed": 42,
+    "total_episodes": 5,
+    "mean_reward": 105.37999999999798,
+    "mean_length": 868.6,
+    "std_reward": 77.50794539916251,
+    "std_length": 626.177802225534,
+    "task_count": 1,
+    "macro_mean_reward": 105.37999999999798,
+    "macro_mean_length": 868.6,
+    "distributed_eval": false
+  }
+}

flappy_fix_latency_2_200ep_last_8_layers_corrected/eval/mid_train/step_4000.json

@@ -0,0 +1,63 @@
+{
+  "per_latency": {
+    "flappy/latency_2": {
+      "latency": 2,
+      "num_episodes": 5,
+      "mean_reward": 261.3800000000085,
+      "mean_length": 2125.4,
+      "std_reward": 157.84373791824612,
+      "std_length": 1270.7383050809478,
+      "episode_rewards": [
+        115.89999999999812,
+        444.60000000002566,
+        444.60000000002566,
+        227.8999999999931,
+        73.89999999999999
+      ],
+      "episode_lengths": [
+        954,
+        3600,
+        3600,
+        1858,
+        615
+      ],
+      "decoded_action_hist": {
+        "0": 8483,
+        "1": 2144
+      },
+      "fixed_episode_seeds": true,
+      "eval_seed": 42,
+      "episode_seeds": [
+        42,
+        43,
+        44,
+        45,
+        46
+      ],
+      "episode_indices": [
+        0,
+        1,
+        2,
+        3,
+        4
+      ]
+    }
+  },
+  "aggregate": {
+    "stage": "mid_train",
+    "step": 4000,
+    "task": "flappy",
+    "model_alias": "openvla",
+    "fixed_episode_seeds": true,
+    "eval_seed": 42,
+    "total_episodes": 5,
+    "mean_reward": 261.3800000000085,
+    "mean_length": 2125.4,
+    "std_reward": 157.84373791824612,
+    "std_length": 1270.7383050809478,
+    "task_count": 1,
+    "macro_mean_reward": 261.3800000000085,
+    "macro_mean_length": 2125.4,
+    "distributed_eval": false
+  }
+}

flappy_fix_latency_2_200ep_last_8_layers_corrected/eval/mid_train/step_4250.json

@@ -0,0 +1,63 @@
+{
+  "per_latency": {
+    "flappy/latency_2": {
+      "latency": 2,
+      "num_episodes": 5,
+      "mean_reward": 321.42000000001417,
+      "mean_length": 2608.4,
+      "std_reward": 159.47787808973385,
+      "std_length": 1284.1907334971702,
+      "episode_rewards": [
+        444.60000000002566,
+        444.60000000002566,
+        218.39999999999353,
+        444.60000000002566,
+        54.90000000000042
+      ],
+      "episode_lengths": [
+        3600,
+        3600,
+        1781,
+        3600,
+        461
+      ],
+      "decoded_action_hist": {
+        "0": 10405,
+        "1": 2637
+      },
+      "fixed_episode_seeds": true,
+      "eval_seed": 42,
+      "episode_seeds": [
+        42,
+        43,
+        44,
+        45,
+        46
+      ],
+      "episode_indices": [
+        0,
+        1,
+        2,
+        3,
+        4
+      ]
+    }
+  },
+  "aggregate": {
+    "stage": "mid_train",
+    "step": 4250,
+    "task": "flappy",
+    "model_alias": "openvla",
+    "fixed_episode_seeds": true,
+    "eval_seed": 42,
+    "total_episodes": 5,
+    "mean_reward": 321.42000000001417,
+    "mean_length": 2608.4,
+    "std_reward": 159.47787808973385,
+    "std_length": 1284.1907334971702,
+    "task_count": 1,
+    "macro_mean_reward": 321.42000000001417,
+    "macro_mean_length": 2608.4,
+    "distributed_eval": false
+  }
+}

flappy_fix_latency_2_200ep_last_8_layers_corrected/eval/mid_train/step_4500.json

@@ -0,0 +1,63 @@
+{
+  "per_latency": {
+    "flappy/latency_2": {
+      "latency": 2,
+      "num_episodes": 5,
+      "mean_reward": 169.44000000000344,
+      "mean_length": 1384.6,
+      "std_reward": 156.5375558771846,
+      "std_length": 1261.0880381638706,
+      "episode_rewards": [
+        115.89999999999812,
+        444.60000000002566,
+        3.8999999999999986,
+        227.8999999999931,
+        54.90000000000042
+      ],
+      "episode_lengths": [
+        954,
+        3600,
+        50,
+        1858,
+        461
+      ],
+      "decoded_action_hist": {
+        "0": 5539,
+        "1": 1384
+      },
+      "fixed_episode_seeds": true,
+      "eval_seed": 42,
+      "episode_seeds": [
+        42,
+        43,
+        44,
+        45,
+        46
+      ],
+      "episode_indices": [
+        0,
+        1,
+        2,
+        3,
+        4
+      ]
+    }
+  },
+  "aggregate": {
+    "stage": "mid_train",
+    "step": 4500,
+    "task": "flappy",
+    "model_alias": "openvla",
+    "fixed_episode_seeds": true,
+    "eval_seed": 42,
+    "total_episodes": 5,
+    "mean_reward": 169.44000000000344,
+    "mean_length": 1384.6,
+    "std_reward": 156.5375558771846,
+    "std_length": 1261.0880381638706,
+    "task_count": 1,
+    "macro_mean_reward": 169.44000000000344,
+    "macro_mean_length": 1384.6,
+    "distributed_eval": false
+  }
+}

flappy_fix_latency_2_200ep_last_8_layers_corrected/eval/mid_train/step_4750.json

@@ -0,0 +1,63 @@
+{
+  "per_latency": {
+    "flappy/latency_2": {
+      "latency": 2,
+      "num_episodes": 5,
+      "mean_reward": 265.9800000000083,
+      "mean_length": 2162.4,
+      "std_reward": 151.80635559818955,
+      "std_length": 1222.032830982867,
+      "episode_rewards": [
+        73.89999999999999,
+        204.39999999999415,
+        444.60000000002566,
+        444.60000000002566,
+        162.39999999999603
+      ],
+      "episode_lengths": [
+        615,
+        1668,
+        3600,
+        3600,
+        1329
+      ],
+      "decoded_action_hist": {
+        "0": 8622,
+        "1": 2190
+      },
+      "fixed_episode_seeds": true,
+      "eval_seed": 42,
+      "episode_seeds": [
+        42,
+        43,
+        44,
+        45,
+        46
+      ],
+      "episode_indices": [
+        0,
+        1,
+        2,
+        3,
+        4
+      ]
+    }
+  },
+  "aggregate": {
+    "stage": "mid_train",
+    "step": 4750,
+    "task": "flappy",
+    "model_alias": "openvla",
+    "fixed_episode_seeds": true,
+    "eval_seed": 42,
+    "total_episodes": 5,
+    "mean_reward": 265.9800000000083,
+    "mean_length": 2162.4,
+    "std_reward": 151.80635559818955,
+    "std_length": 1222.032830982867,
+    "task_count": 1,
+    "macro_mean_reward": 265.9800000000083,
+    "macro_mean_length": 2162.4,
+    "distributed_eval": false
+  }
+}

flappy_fix_latency_2_200ep_last_8_layers_corrected/eval/mid_train/step_500.json

@@ -0,0 +1,63 @@
+{
+  "per_latency": {
+    "flappy/latency_2": {
+      "latency": 2,
+      "num_episodes": 5,
+      "mean_reward": 3.6599999999999975,
+      "mean_length": 45.8,
+      "std_reward": 2.48161237907937,
+      "std_length": 21.40467238712146,
+      "episode_rewards": [
+        8.399999999999986,
+        2.0000000000000013,
+        3.8999999999999986,
+        2.0000000000000013,
+        2.0000000000000013
+      ],
+      "episode_lengths": [
+        86,
+        31,
+        50,
+        31,
+        31
+      ],
+      "decoded_action_hist": {
+        "0": 203,
+        "1": 26
+      },
+      "fixed_episode_seeds": true,
+      "eval_seed": 42,
+      "episode_seeds": [
+        42,
+        43,
+        44,
+        45,
+        46
+      ],
+      "episode_indices": [
+        0,
+        1,
+        2,
+        3,
+        4
+      ]
+    }
+  },
+  "aggregate": {
+    "stage": "mid_train",
+    "step": 500,
+    "task": "flappy",
+    "model_alias": "openvla",
+    "fixed_episode_seeds": true,
+    "eval_seed": 42,
+    "total_episodes": 5,
+    "mean_reward": 3.6599999999999975,
+    "mean_length": 45.8,
+    "std_reward": 2.48161237907937,
+    "std_length": 21.40467238712146,
+    "task_count": 1,
+    "macro_mean_reward": 3.6599999999999975,
+    "macro_mean_length": 45.8,
+    "distributed_eval": false
+  }
+}

flappy_fix_latency_2_200ep_last_8_layers_corrected/eval/mid_train/step_5000.json

@@ -0,0 +1,63 @@
+{
+  "per_latency": {
+    "flappy/latency_2": {
+      "latency": 2,
+      "num_episodes": 5,
+      "mean_reward": 217.48000000001034,
+      "mean_length": 1771.0,
+      "std_reward": 185.64113121828532,
+      "std_length": 1494.9829430465086,
+      "episode_rewards": [
+        73.89999999999999,
+        444.60000000002566,
+        73.89999999999999,
+        444.60000000002566,
+        50.40000000000037
+      ],
+      "episode_lengths": [
+        615,
+        3600,
+        615,
+        3600,
+        425
+      ],
+      "decoded_action_hist": {
+        "0": 7086,
+        "1": 1769
+      },
+      "fixed_episode_seeds": true,
+      "eval_seed": 42,
+      "episode_seeds": [
+        42,
+        43,
+        44,
+        45,
+        46
+      ],
+      "episode_indices": [
+        0,
+        1,
+        2,
+        3,
+        4
+      ]
+    }
+  },
+  "aggregate": {
+    "stage": "mid_train",
+    "step": 5000,
+    "task": "flappy",
+    "model_alias": "openvla",
+    "fixed_episode_seeds": true,
+    "eval_seed": 42,
+    "total_episodes": 5,
+    "mean_reward": 217.48000000001034,
+    "mean_length": 1771.0,
+    "std_reward": 185.64113121828532,
+    "std_length": 1494.9829430465086,
+    "task_count": 1,
+    "macro_mean_reward": 217.48000000001034,
+    "macro_mean_length": 1771.0,
+    "distributed_eval": false
+  }
+}

flappy_fix_latency_2_200ep_last_8_layers_corrected/eval/mid_train/step_750.json

@@ -0,0 +1,63 @@
+{
+  "per_latency": {
+    "flappy/latency_2": {
+      "latency": 2,
+      "num_episodes": 5,
+      "mean_reward": 26.359999999999825,
+      "mean_length": 229.6,
+      "std_reward": 34.903329354088484,
+      "std_length": 280.27172529529264,
+      "episode_rewards": [
+        6.799999999999992,
+        22.40000000000006,
+        3.8999999999999986,
+        3.8999999999999986,
+        94.79999999999909
+      ],
+      "episode_lengths": [
+        70,
+        199,
+        50,
+        50,
+        779
+      ],
+      "decoded_action_hist": {
+        "0": 920,
+        "1": 228
+      },
+      "fixed_episode_seeds": true,
+      "eval_seed": 42,
+      "episode_seeds": [
+        42,
+        43,
+        44,
+        45,
+        46
+      ],
+      "episode_indices": [
+        0,
+        1,
+        2,
+        3,
+        4
+      ]
+    }
+  },
+  "aggregate": {
+    "stage": "mid_train",
+    "step": 750,
+    "task": "flappy",
+    "model_alias": "openvla",
+    "fixed_episode_seeds": true,
+    "eval_seed": 42,
+    "total_episodes": 5,
+    "mean_reward": 26.359999999999825,
+    "mean_length": 229.6,
+    "std_reward": 34.903329354088484,
+    "std_length": 280.27172529529264,
+    "task_count": 1,
+    "macro_mean_reward": 26.359999999999825,
+    "macro_mean_length": 229.6,
+    "distributed_eval": false
+  }
+}

flappy_fix_latency_2_200ep_last_8_layers_corrected/hydra/.hydra/config.yaml

@@ -0,0 +1,217 @@
+framework:
+  qwenvl:
+    base_vlm: /workspace/latency-sensitive-bench/starVLA/playground/Pretrained_models/Qwen3-VL-4B-Instruct
+    attn_implementation: flash_attention_2
+    enable_gradient_checkpointing: true
+  action_model:
+    state_dim: 7
+    loss_type: discrete_ce
+    action_horizon: 1
+    future_action_window_size: 0
+    past_action_window_size: 0
+    action_dim: 7
+    action_env_dim: 2
+  name: QwenOFT
+datasets:
+  vla_data:
+    dataset_py: lerobot_datasets
+    include_state: true
+    data_root_dir: /workspace/latency-sensitive-bench/starVLA/data/flappy_fix_latency_2_200ep
+    data_mix: flappy_train__bridge
+    eval_data_mix: flappy_train__bridge__val
+    custom_mixtures_path: null
+    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
+    per_device_batch_size: 64
+    load_all_data_for_training: true
+    num_obs_frames: 1
+    image_mode: single
+    stitch_grid:
+    - 2
+    - 2
+    obs_image_size: null
+    video_backend: torchvision_av
+dataset:
+  source_hf: ${dataset.single_source_hf}
+  config_name: null
+  source_subdir: null
+  converted_name: ${dataset.single_converted_name}
+  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
+  verify_rows: 200
+  max_episodes: null
+  episodes_per_latency: null
+  latency_filter: ${dataset.single_latency_filter}
+  debug_subset:
+    enabled: false
+    max_episodes: 5
+    suffix: debug
+base_model:
+  repo_id: ${bridge_base_model.repo_id.${model}}
+initialization:
+  checkpoint_local_dir: playground/Pretrained_models/Qwen3VL-OFT-Bridge-RT-1
+  checkpoint_hf_repo_id: StarVLA/Qwen3VL-OFT-Bridge-RT-1
+  checkpoint_filename: checkpoints/steps_5000_pytorch_model.pt
+trainer:
+  max_train_steps: 5000
+  num_warmup_steps: 100
+  save_interval: 500
+  eval_interval: 100
+  eval_num_batches: 100
+  per_latency_eval_num_batches: null
+  eval_action_classification: true
+  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_llm_bottom_ratio: 0.7778
+  loss_scale:
+    vla: 1.0
+    vlm: 0.1
+  max_grad_norm: 1.0
+  weight_decay: 0.0
+  logging_frequency: 1
+  gradient_clipping: 1.0
+  gradient_accumulation_steps: 2
+  distributed_backend: none
+  is_resume: false
+  pretrained_checkpoint: /workspace/latency-sensitive-bench/starVLA/results/Checkpoints/flappy_fix_latency_2_200ep_last_8_layers_corrected/checkpoints/_initialization/StarVLA_Qwen3VL-OFT-Bridge-RT-1/checkpoints/steps_5000_pytorch_model.pt
+  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: /workspace/latency-sensitive-bench/starVLA/results/Checkpoints
+seed: 42
+wandb_entity: saberrr-zju
+wandb_project: starVLA_rl_games
+auth:
+  env_file: null
+  hf_token_env: HF_TOKEN
+  wandb_api_key_env: WANDB_API_KEY
+paths:
+  run_root_dir: results/Checkpoints
+  dataset_local_dir: playground/Datasets/rl_games
+  dataset_cache_dir: null
+  base_model_dir: ${bridge_base_model.local_dir.${model}}
+  accelerate_config: starVLA/config/deepseeds/deepspeed_zero2.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
+    seed: 42
+    fixed_episode_seeds: true
+    latency_seed_stride: 0
+    task_seed_stride: 0
+    task_description: ''
+    enabled: true
+    distributed_mode: none
+    vectorized:
+      enabled: false
+      batch_size: 1
+    latency:
+      prompt_map_path: /workspace/latency-sensitive-bench/starVLA/data/flappy_fix_latency_2_200ep/flappy_train__bridge/latency_prompt_map.json
+      mode: single
+      values:
+      - 0
+    mid_train:
+      enabled: true
+      interval_steps: 250
+      latencies:
+      - 2
+      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
+  task: flappy
+  initialization_mode: bridge
+  action_carrier: bridge
+model: openvla
+env: flappy
+init: bridge
+bridge_base_model:
+  repo_id:
+    openvla: Qwen/Qwen3-VL-4B-Instruct
+    pi0: StarVLA/Qwen2.5-VL-3B-Instruct-Action
+    pi05: Qwen/Qwen3-VL-4B-Instruct
+    gr00t: Qwen/Qwen3-VL-4B-Instruct
+  local_dir:
+    openvla: playground/Pretrained_models/Qwen3-VL-4B-Instruct
+    pi0: playground/Pretrained_models/Qwen2.5-VL-3B-Instruct-Action
+    pi05: playground/Pretrained_models/Qwen3-VL-4B-Instruct
+    gr00t: playground/Pretrained_models/Qwen3-VL-4B-Instruct
+mode: single
+checkpoint:
+  load: auto
+  hf_repo_id: null
+  save_best_model: false
+  save_pt_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: flappy_fix_latency_2_200ep_last_8_layers_corrected
+output_dir: null
+config_yaml: null
+is_debug: false
+version_id: '0.21'

flappy_fix_latency_2_200ep_last_8_layers_corrected/hydra/.hydra/hydra.yaml

@@ -0,0 +1,266 @@
+hydra:
+  run:
+    dir: ${run_root_dir}/${run_id}/hydra
+  sweep:
+    dir: multirun/${now:%Y-%m-%d}/${now:%H-%M-%S}
+    subdir: ${hydra.job.num}
+  launcher:
+    _target_: hydra._internal.core_plugins.basic_launcher.BasicLauncher
+  sweeper:
+    _target_: hydra._internal.core_plugins.basic_sweeper.BasicSweeper
+    max_batch_size: null
+    params: null
+  help:
+    app_name: ${hydra.job.name}
+    header: '${hydra.help.app_name} is powered by Hydra.
+
+      '
+    footer: 'Powered by Hydra (https://hydra.cc)
+
+      Use --hydra-help to view Hydra specific help
+
+      '
+    template: '${hydra.help.header}
+
+      == Configuration groups ==
+
+      Compose your configuration from those groups (group=option)
+
+
+      $APP_CONFIG_GROUPS
+
+
+      == Config ==
+
+      Override anything in the config (foo.bar=value)
+
+
+      $CONFIG
+
+
+      ${hydra.help.footer}
+
+      '
+  hydra_help:
+    template: 'Hydra (${hydra.runtime.version})
+
+      See https://hydra.cc for more info.
+
+
+      == Flags ==
+
+      $FLAGS_HELP
+
+
+      == Configuration groups ==
+
+      Compose your configuration from those groups (For example, append hydra/job_logging=disabled
+      to command line)
+
+
+      $HYDRA_CONFIG_GROUPS
+
+
+      Use ''--cfg hydra'' to Show the Hydra config.
+
+      '
+    hydra_help: ???
+  hydra_logging:
+    version: 1
+    formatters:
+      simple:
+        format: '[%(asctime)s][HYDRA] %(message)s'
+    handlers:
+      console:
+        class: logging.StreamHandler
+        formatter: simple
+        stream: ext://sys.stdout
+    root:
+      level: INFO
+      handlers:
+      - console
+    loggers:
+      logging_example:
+        level: DEBUG
+    disable_existing_loggers: false
+  job_logging:
+    version: 1
+    formatters:
+      simple:
+        format: '[%(asctime)s][%(name)s][%(levelname)s] - %(message)s'
+    handlers:
+      console:
+        class: logging.StreamHandler
+        formatter: simple
+        stream: ext://sys.stdout
+      file:
+        class: logging.FileHandler
+        formatter: simple
+        filename: ${hydra.runtime.output_dir}/${hydra.job.name}.log
+    root:
+      level: INFO
+      handlers:
+      - console
+      - file
+    disable_existing_loggers: false
+  env: {}
+  mode: RUN
+  searchpath: []
+  callbacks: {}
+  output_subdir: .hydra
+  overrides:
+    hydra:
+    - hydra.mode=RUN
+    task:
+    - model=openvla
+    - env=flappy
+    - init=bridge
+    - mode=single
+    - run_id=flappy_fix_latency_2_200ep_last_8_layers_corrected
+    - run_root_dir=/workspace/latency-sensitive-bench/starVLA/results/Checkpoints
+    - seed=42
+    - wandb_entity=saberrr-zju
+    - wandb_project=starVLA_rl_games
+    - rl_games.env_eval.enabled=true
+    - checkpoint.sync.enabled=false
+    - checkpoint.sync.keep_last_n=0
+    - checkpoint.local.keep_last_n=1
+    - checkpoint.save_best_model=false
+    - checkpoint.save_pt_file=false
+    - trainer.is_resume=false
+    - trainer.pretrained_checkpoint=/workspace/latency-sensitive-bench/starVLA/results/Checkpoints/flappy_fix_latency_2_200ep_last_8_layers_corrected/checkpoints/_initialization/StarVLA_Qwen3VL-OFT-Bridge-RT-1/checkpoints/steps_5000_pytorch_model.pt
+    - trainer.resume_step=0
+    - trainer.max_train_steps=5000
+    - trainer.num_warmup_steps=100
+    - trainer.save_interval=500
+    - trainer.eval_interval=100
+    - trainer.eval_num_batches=100
+    - trainer.eval_action_classification=true
+    - trainer.logging_frequency=1
+    - trainer.gradient_accumulation_steps=2
+    - trainer.distributed_backend=none
+    - trainer.learning_rate.base=2e-05
+    - trainer.learning_rate.qwen_vl_interface=1e-05
+    - trainer.learning_rate.action_model=0.0001
+    - trainer.lr_scheduler_type=cosine_with_min_lr
+    - trainer.scheduler_specific_kwargs.min_lr=1e-06
+    - trainer.freeze_llm_bottom_ratio=0.7778
+    - trainer.loss_scale.vla=1.0
+    - trainer.loss_scale.vlm=0.1
+    - trainer.max_grad_norm=1.0
+    - trainer.weight_decay=0.0
+    - trainer.gradient_clipping=1.0
+    - trainer.optimizer.name=AdamW
+    - trainer.optimizer.betas=[0.9,0.95]
+    - trainer.optimizer.eps=1e-08
+    - trainer.optimizer.weight_decay=1e-08
+    - trainer.optimizer.fused=true
+    - trainer.save_format=pt
+    - framework.name=QwenOFT
+    - framework.qwenvl.attn_implementation=flash_attention_2
+    - framework.qwenvl.enable_gradient_checkpointing=true
+    - framework.action_model.action_dim=7
+    - framework.action_model.action_env_dim=2
+    - framework.action_model.state_dim=7
+    - framework.action_model.loss_type=discrete_ce
+    - framework.action_model.action_horizon=1
+    - framework.action_model.future_action_window_size=0
+    - framework.action_model.past_action_window_size=0
+    - datasets.vla_data.include_state=true
+    - datasets.vla_data.action_type=discrete
+    - datasets.vla_data.sequential_step_sampling=false
+    - datasets.vla_data.shuffle=true
+    - datasets.vla_data.num_workers=8
+    - datasets.vla_data.eval_num_workers=8
+    - datasets.vla_data.prefetch_factor=4
+    - datasets.vla_data.persistent_workers=true
+    - datasets.vla_data.pin_memory=true
+    - datasets.vla_data.action_balance.enabled=false
+    - datasets.vla_data.action_balance.strategy=balanced_epoch
+    - datasets.vla_data.action_balance.action_key=action_id
+    - datasets.vla_data.action_balance.target_flap_fraction=0.3
+    - datasets.vla_data.action_balance.noop_id=0
+    - datasets.vla_data.action_balance.flap_id=1
+    - datasets.vla_data.latency_curriculum.enabled=false
+    - datasets.vla_data.latency_curriculum.strategy=exclusive
+    - datasets.vla_data.per_device_batch_size=64
+    - datasets.vla_data.num_workers=8
+    - datasets.vla_data.eval_num_workers=8
+    - datasets.vla_data.prefetch_factor=4
+    - datasets.vla_data.persistent_workers=true
+    - datasets.vla_data.pin_memory=true
+    - datasets.vla_data.load_all_data_for_training=true
+    - datasets.vla_data.video_backend=torchvision_av
+    - datasets.vla_data.data_root_dir=/workspace/latency-sensitive-bench/starVLA/data/flappy_fix_latency_2_200ep
+    - datasets.vla_data.data_mix=flappy_train__bridge
+    - datasets.vla_data.eval_data_mix=flappy_train__bridge__val
+    - framework.qwenvl.base_vlm=/workspace/latency-sensitive-bench/starVLA/playground/Pretrained_models/Qwen3-VL-4B-Instruct
+    - rl_games.task=flappy
+    - rl_games.model_alias=openvla
+    - rl_games.initialization_mode=bridge
+    - rl_games.action_carrier=bridge
+    - rl_games.env_eval.distributed_mode=none
+    - rl_games.env_eval.latency.mode=single
+    - rl_games.env_eval.frameskip=1
+    - rl_games.env_eval.image_size=224
+    - rl_games.env_eval.seed=42
+    - rl_games.env_eval.fixed_episode_seeds=true
+    - rl_games.env_eval.latency_seed_stride=0
+    - rl_games.env_eval.task_seed_stride=0
+    - rl_games.env_eval.latency.values=[0]
+    - rl_games.env_eval.latency.prompt_map_path=/workspace/latency-sensitive-bench/starVLA/data/flappy_fix_latency_2_200ep/flappy_train__bridge/latency_prompt_map.json
+    - rl_games.env_eval.mid_train.enabled=true
+    - rl_games.env_eval.mid_train.interval_steps=250
+    - rl_games.env_eval.mid_train.num_episodes=5
+    - rl_games.env_eval.mid_train.max_steps_per_episode=3600
+    - rl_games.env_eval.mid_train.latencies=[2]
+    - rl_games.env_eval.post_train.enabled=false
+    - rl_games.env_eval.post_train.num_episodes=5
+    - rl_games.env_eval.post_train.max_steps_per_episode=3600
+    - rl_games.env_eval.post_train.latencies=[0,1,2,3,4]
+  job:
+    name: train_starvla_hydra
+    chdir: false
+    override_dirname: checkpoint.local.keep_last_n=1,checkpoint.save_best_model=false,checkpoint.save_pt_file=false,checkpoint.sync.enabled=false,checkpoint.sync.keep_last_n=0,datasets.vla_data.action_balance.action_key=action_id,datasets.vla_data.action_balance.enabled=false,datasets.vla_data.action_balance.flap_id=1,datasets.vla_data.action_balance.noop_id=0,datasets.vla_data.action_balance.strategy=balanced_epoch,datasets.vla_data.action_balance.target_flap_fraction=0.3,datasets.vla_data.action_type=discrete,datasets.vla_data.data_mix=flappy_train__bridge,datasets.vla_data.data_root_dir=/workspace/latency-sensitive-bench/starVLA/data/flappy_fix_latency_2_200ep,datasets.vla_data.eval_data_mix=flappy_train__bridge__val,datasets.vla_data.eval_num_workers=8,datasets.vla_data.eval_num_workers=8,datasets.vla_data.include_state=true,datasets.vla_data.latency_curriculum.enabled=false,datasets.vla_data.latency_curriculum.strategy=exclusive,datasets.vla_data.load_all_data_for_training=true,datasets.vla_data.num_workers=8,datasets.vla_data.num_workers=8,datasets.vla_data.per_device_batch_size=64,datasets.vla_data.persistent_workers=true,datasets.vla_data.persistent_workers=true,datasets.vla_data.pin_memory=true,datasets.vla_data.pin_memory=true,datasets.vla_data.prefetch_factor=4,datasets.vla_data.prefetch_factor=4,datasets.vla_data.sequential_step_sampling=false,datasets.vla_data.shuffle=true,datasets.vla_data.video_backend=torchvision_av,env=flappy,framework.action_model.action_dim=7,framework.action_model.action_env_dim=2,framework.action_model.action_horizon=1,framework.action_model.future_action_window_size=0,framework.action_model.loss_type=discrete_ce,framework.action_model.past_action_window_size=0,framework.action_model.state_dim=7,framework.name=QwenOFT,framework.qwenvl.attn_implementation=flash_attention_2,framework.qwenvl.base_vlm=/workspace/latency-sensitive-bench/starVLA/playground/Pretrained_models/Qwen3-VL-4B-Instruct,framework.qwenvl.enable_gradient_checkpointing=true,init=bridge,mode=single,model=openvla,rl_games.action_carrier=bridge,rl_games.env_eval.distributed_mode=none,rl_games.env_eval.enabled=true,rl_games.env_eval.fixed_episode_seeds=true,rl_games.env_eval.frameskip=1,rl_games.env_eval.image_size=224,rl_games.env_eval.latency.mode=single,rl_games.env_eval.latency.prompt_map_path=/workspace/latency-sensitive-bench/starVLA/data/flappy_fix_latency_2_200ep/flappy_train__bridge/latency_prompt_map.json,rl_games.env_eval.latency.values=[0],rl_games.env_eval.latency_seed_stride=0,rl_games.env_eval.mid_train.enabled=true,rl_games.env_eval.mid_train.interval_steps=250,rl_games.env_eval.mid_train.latencies=[2],rl_games.env_eval.mid_train.max_steps_per_episode=3600,rl_games.env_eval.mid_train.num_episodes=5,rl_games.env_eval.post_train.enabled=false,rl_games.env_eval.post_train.latencies=[0,1,2,3,4],rl_games.env_eval.post_train.max_steps_per_episode=3600,rl_games.env_eval.post_train.num_episodes=5,rl_games.env_eval.seed=42,rl_games.env_eval.task_seed_stride=0,rl_games.initialization_mode=bridge,rl_games.model_alias=openvla,rl_games.task=flappy,run_id=flappy_fix_latency_2_200ep_last_8_layers_corrected,run_root_dir=/workspace/latency-sensitive-bench/starVLA/results/Checkpoints,seed=42,trainer.distributed_backend=none,trainer.eval_action_classification=true,trainer.eval_interval=100,trainer.eval_num_batches=100,trainer.freeze_llm_bottom_ratio=0.7778,trainer.gradient_accumulation_steps=2,trainer.gradient_clipping=1.0,trainer.is_resume=false,trainer.learning_rate.action_model=0.0001,trainer.learning_rate.base=2e-05,trainer.learning_rate.qwen_vl_interface=1e-05,trainer.logging_frequency=1,trainer.loss_scale.vla=1.0,trainer.loss_scale.vlm=0.1,trainer.lr_scheduler_type=cosine_with_min_lr,trainer.max_grad_norm=1.0,trainer.max_train_steps=5000,trainer.num_warmup_steps=100,trainer.optimizer.betas=[0.9,0.95],trainer.optimizer.eps=1e-08,trainer.optimizer.fused=true,trainer.optimizer.name=AdamW,trainer.optimizer.weight_decay=1e-08,trainer.pretrained_checkpoint=/workspace/latency-sensitive-bench/starVLA/results/Checkpoints/flappy_fix_latency_2_200ep_last_8_layers_corrected/checkpoints/_initialization/StarVLA_Qwen3VL-OFT-Bridge-RT-1/checkpoints/steps_5000_pytorch_model.pt,trainer.resume_step=0,trainer.save_format=pt,trainer.save_interval=500,trainer.scheduler_specific_kwargs.min_lr=1e-06,trainer.weight_decay=0.0,wandb_entity=saberrr-zju,wandb_project=starVLA_rl_games
+    id: ???
+    num: ???
+    config_name: train
+    env_set: {}
+    env_copy: []
+    config:
+      override_dirname:
+        kv_sep: '='
+        item_sep: ','
+        exclude_keys: []
+  runtime:
+    version: 1.3.3
+    version_base: '1.1'
+    cwd: /workspace/latency-sensitive-bench/starVLA
+    config_sources:
+    - path: hydra.conf
+      schema: pkg
+      provider: hydra
+    - path: /workspace/latency-sensitive-bench/starVLA/examples/rl_games/config
+      schema: file
+      provider: main
+    - path: ''
+      schema: structured
+      provider: schema
+    output_dir: /workspace/latency-sensitive-bench/starVLA/results/Checkpoints/flappy_fix_latency_2_200ep_last_8_layers_corrected/hydra
+    choices:
+      cross_task_setup: null
+      checkpoint: default
+      mode: single
+      init: bridge
+      env: flappy
+      model: openvla
+      hydra/env: default
+      hydra/callbacks: null
+      hydra/job_logging: default
+      hydra/hydra_logging: default
+      hydra/hydra_help: default
+      hydra/help: default
+      hydra/sweeper: basic
+      hydra/launcher: basic
+      hydra/output: default
+  verbose: false

flappy_fix_latency_2_200ep_last_8_layers_corrected/hydra/.hydra/overrides.yaml

@@ -0,0 +1,106 @@
+- model=openvla
+- env=flappy
+- init=bridge
+- mode=single
+- run_id=flappy_fix_latency_2_200ep_last_8_layers_corrected
+- run_root_dir=/workspace/latency-sensitive-bench/starVLA/results/Checkpoints
+- seed=42
+- wandb_entity=saberrr-zju
+- wandb_project=starVLA_rl_games
+- rl_games.env_eval.enabled=true
+- checkpoint.sync.enabled=false
+- checkpoint.sync.keep_last_n=0
+- checkpoint.local.keep_last_n=1
+- checkpoint.save_best_model=false
+- checkpoint.save_pt_file=false
+- trainer.is_resume=false
+- trainer.pretrained_checkpoint=/workspace/latency-sensitive-bench/starVLA/results/Checkpoints/flappy_fix_latency_2_200ep_last_8_layers_corrected/checkpoints/_initialization/StarVLA_Qwen3VL-OFT-Bridge-RT-1/checkpoints/steps_5000_pytorch_model.pt
+- trainer.resume_step=0
+- trainer.max_train_steps=5000
+- trainer.num_warmup_steps=100
+- trainer.save_interval=500
+- trainer.eval_interval=100
+- trainer.eval_num_batches=100
+- trainer.eval_action_classification=true
+- trainer.logging_frequency=1
+- trainer.gradient_accumulation_steps=2
+- trainer.distributed_backend=none
+- trainer.learning_rate.base=2e-05
+- trainer.learning_rate.qwen_vl_interface=1e-05
+- trainer.learning_rate.action_model=0.0001
+- trainer.lr_scheduler_type=cosine_with_min_lr
+- trainer.scheduler_specific_kwargs.min_lr=1e-06
+- trainer.freeze_llm_bottom_ratio=0.7778
+- trainer.loss_scale.vla=1.0
+- trainer.loss_scale.vlm=0.1
+- trainer.max_grad_norm=1.0
+- trainer.weight_decay=0.0
+- trainer.gradient_clipping=1.0
+- trainer.optimizer.name=AdamW
+- trainer.optimizer.betas=[0.9,0.95]
+- trainer.optimizer.eps=1e-08
+- trainer.optimizer.weight_decay=1e-08
+- trainer.optimizer.fused=true
+- trainer.save_format=pt
+- framework.name=QwenOFT
+- framework.qwenvl.attn_implementation=flash_attention_2
+- framework.qwenvl.enable_gradient_checkpointing=true
+- framework.action_model.action_dim=7
+- framework.action_model.action_env_dim=2
+- framework.action_model.state_dim=7
+- framework.action_model.loss_type=discrete_ce
+- framework.action_model.action_horizon=1
+- framework.action_model.future_action_window_size=0
+- framework.action_model.past_action_window_size=0
+- datasets.vla_data.include_state=true
+- datasets.vla_data.action_type=discrete
+- datasets.vla_data.sequential_step_sampling=false
+- datasets.vla_data.shuffle=true
+- datasets.vla_data.num_workers=8
+- datasets.vla_data.eval_num_workers=8
+- datasets.vla_data.prefetch_factor=4
+- datasets.vla_data.persistent_workers=true
+- datasets.vla_data.pin_memory=true
+- datasets.vla_data.action_balance.enabled=false
+- datasets.vla_data.action_balance.strategy=balanced_epoch
+- datasets.vla_data.action_balance.action_key=action_id
+- datasets.vla_data.action_balance.target_flap_fraction=0.3
+- datasets.vla_data.action_balance.noop_id=0
+- datasets.vla_data.action_balance.flap_id=1
+- datasets.vla_data.latency_curriculum.enabled=false
+- datasets.vla_data.latency_curriculum.strategy=exclusive
+- datasets.vla_data.per_device_batch_size=64
+- datasets.vla_data.num_workers=8
+- datasets.vla_data.eval_num_workers=8
+- datasets.vla_data.prefetch_factor=4
+- datasets.vla_data.persistent_workers=true
+- datasets.vla_data.pin_memory=true
+- datasets.vla_data.load_all_data_for_training=true
+- datasets.vla_data.video_backend=torchvision_av
+- datasets.vla_data.data_root_dir=/workspace/latency-sensitive-bench/starVLA/data/flappy_fix_latency_2_200ep
+- datasets.vla_data.data_mix=flappy_train__bridge
+- datasets.vla_data.eval_data_mix=flappy_train__bridge__val
+- framework.qwenvl.base_vlm=/workspace/latency-sensitive-bench/starVLA/playground/Pretrained_models/Qwen3-VL-4B-Instruct
+- rl_games.task=flappy
+- rl_games.model_alias=openvla
+- rl_games.initialization_mode=bridge
+- rl_games.action_carrier=bridge
+- rl_games.env_eval.distributed_mode=none
+- rl_games.env_eval.latency.mode=single
+- rl_games.env_eval.frameskip=1
+- rl_games.env_eval.image_size=224
+- rl_games.env_eval.seed=42
+- rl_games.env_eval.fixed_episode_seeds=true
+- rl_games.env_eval.latency_seed_stride=0
+- rl_games.env_eval.task_seed_stride=0
+- rl_games.env_eval.latency.values=[0]
+- rl_games.env_eval.latency.prompt_map_path=/workspace/latency-sensitive-bench/starVLA/data/flappy_fix_latency_2_200ep/flappy_train__bridge/latency_prompt_map.json
+- rl_games.env_eval.mid_train.enabled=true
+- rl_games.env_eval.mid_train.interval_steps=250
+- rl_games.env_eval.mid_train.num_episodes=5
+- rl_games.env_eval.mid_train.max_steps_per_episode=3600
+- rl_games.env_eval.mid_train.latencies=[2]
+- rl_games.env_eval.post_train.enabled=false
+- rl_games.env_eval.post_train.num_episodes=5
+- rl_games.env_eval.post_train.max_steps_per_episode=3600
+- rl_games.env_eval.post_train.latencies=[0,1,2,3,4]

flappy_fix_latency_2_200ep_last_8_layers_corrected/summary.jsonl

@@ -0,0 +1,10 @@
+{"steps": 500}
+{"steps": 1000}
+{"steps": 1500}
+{"steps": 2000}
+{"steps": 2500}
+{"steps": 3000}
+{"steps": 3500}
+{"steps": 4000}
+{"steps": 4500}
+{"steps": 5000}