feat: mimic pi05 config and episode control

configs/pi05/pi05_behavior_pretrain.toml → configs/pi05/pi05-b1k-pretrain.toml

@@ -12,6 +12,8 @@ optm_weight_decay = 1e-10
 optm_betas = [0.9, 0.95]
 optm_warmup_steps = 1000
 optm_grad_norm_clip = 1.0
+optm_decay_type = "cosine"
+optm_decay_ratio = 1.0
 epsilon = 1e-8
 compile = true
 param_dtype = "bfloat16"
@@ -73,4 +75,7 @@ model_type = "PI05"
 prompt_from_task = true
 skip_norm_stats = false
 norm_stats="/workspace/official_pi05_base_norm_stats.json"
-episodes_index = [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59, 60, 61, 62, 63, 64, 65, 66, 67, 68, 69, 70, 71, 72, 73, 74, 75, 76, 77, 78, 79, 80, 81, 82, 83, 84, 85, 86, 87, 88, 89, 90, 91, 92, 93, 94, 95, 96, 97, 98, 99, 100, 101, 102, 103, 104, 105, 106, 107, 108, 109, 110, 111, 112, 113, 114, 115, 116, 117, 118, 119, 120, 121, 122, 123, 124, 125, 126, 127, 128, 129, 130, 131, 132, 133, 134, 135, 136, 137, 138, 139, 140, 141, 142, 143, 144, 145, 146, 147, 148, 149, 150, 151, 152, 153, 154, 155, 156, 157, 158, 159, 160, 161, 162, 163, 164, 165, 166, 167, 168, 169, 170, 171, 172, 173, 174, 175, 176, 177, 178, 179, 180, 181, 182, 183, 184, 185, 186, 187, 188, 189, 190, 191, 192, 193, 194, 195, 196, 197, 198, 199]
+episodes_index = [0, 200]
+add_dataset_name = "delinqu/comet-1.5k"
+add_dataset_local_dir = "/workspace/comet-1.5k"
+add_data_multiplier = 2

configs/pi05/pi05_behavior_sft.toml → configs/pi05/pi05-b1k-sft.toml

@@ -12,6 +12,8 @@ optm_weight_decay = 1e-10
 optm_betas = [0.9, 0.95]
 optm_warmup_steps = 1000
 optm_grad_norm_clip = 1.0
+optm_decay_type = "cosine"
+optm_decay_ratio = 1.0
 epsilon = 1e-8
 compile = true
 param_dtype = "bfloat16"
@@ -73,4 +75,7 @@ model_type = "PI05"
 prompt_from_task = true
 skip_norm_stats = false
 norm_stats="/workspace/comet_weights_pytorch_2/pi05-b1kpt50-cs32/assets/behavior-1k/2025-challenge-demos/norm_stats.json" # default use json from resumed weights
-episodes_index = [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59, 60, 61, 62, 63, 64, 65, 66, 67, 68, 69, 70, 71, 72, 73, 74, 75, 76, 77, 78, 79, 80, 81, 82, 83, 84, 85, 86, 87, 88, 89, 90, 91, 92, 93, 94, 95, 96, 97, 98, 99, 100, 101, 102, 103, 104, 105, 106, 107, 108, 109, 110, 111, 112, 113, 114, 115, 116, 117, 118, 119, 120, 121, 122, 123, 124, 125, 126, 127, 128, 129, 130, 131, 132, 133, 134, 135, 136, 137, 138, 139, 140, 141, 142, 143, 144, 145, 146, 147, 148, 149, 150, 151, 152, 153, 154, 155, 156, 157, 158, 159, 160, 161, 162, 163, 164, 165, 166, 167, 168, 169, 170, 171, 172, 173, 174, 175, 176, 177, 178, 179, 180, 181, 182, 183, 184, 185, 186, 187, 188, 189, 190, 191, 192, 193, 194, 195, 196, 197, 198, 199]
+episodes_index = [0, 200]
+add_dataset_name = "delinqu/comet-1.5k"
+add_dataset_local_dir = "/workspace/comet-1.5k"
+add_data_multiplier = 2

cosmos_rl/tools/dataset/b1k_sft.py

@@ -500,6 +500,19 @@ def __call__(self, data: dict) -> dict:
         return inputs
 
 
+def parse_episodes_index(episodes: Any) -> list[int]:
+    if not isinstance(episodes, list) or not all(isinstance(i, int) for i in episodes):
+        raise ValueError("`episodes_index` must be a list of integers.")
+
+    if len(episodes) == 2:
+        start, end = episodes
+        if end < start:
+            raise ValueError(f"`episodes_index` has invalid interval: end({end}) < start({start}).")
+        return list(range(start, end))
+
+    return episodes
+
+
 class BehaviorSFTDataset(Dataset):
     """
     Thin wrapper around `BehaviorLeRobotDataset` used for SFT.
@@ -523,7 +536,7 @@ def __init__(self, config: CosmosConfig):
                 key: [t / 30.0 for t in range(config.custom["action_horizon"])]
                 for key in config.custom["action_sequence_keys"]
             },
-            episodes=config.custom["episodes_index"],
+            episodes=parse_episodes_index(config.custom["episodes_index"]),
             chunk_streaming_using_keyframe=True,
             shuffle=True,
         )