Upload baselines/vitra-cotrain/H100_R000/config.json with huggingface_hub

baselines/vitra-cotrain/H100_R000/config.json

@@ -0,0 +1,92 @@
+{
+  "vla_name": "VITRA_Paligemma",
+  "task_name": "Cotrain_H100_R000_seed42",
+  "model": "vitra_paligemma2",
+  "fwd_pred_next_n": 16,
+  "seed": 42,
+  "batch_size": 8,
+  "total_batch_size": 32,
+  "output_root": "/rlwrld-unified-checkpoints/seunghoon/vitra_cotrain/runs/checkpoints",
+  "log_root": "/rlwrld-unified-checkpoints/seunghoon/vitra_cotrain/runs/logs",
+  "cache_root": "/rlwrld-unified-checkpoints/seunghoon/vitra_cotrain/runs/cache/vitra_paligemma2",
+  "pretrain_path": "/sjw_alinlab/home/seunghoon/vitra_cotrain/models/vitra-vla-3b.pt",
+  "model_load_path": null,
+  "resume": true,
+  "wandb_project": "vitra_inspire_cotrain",
+  "wandb_entity": "",
+  "save_steps": 2500,
+  "epoch_save_interval": 100000,
+  "use_bf16": true,
+  "use_fov": true,
+  "untied_cognition_token": true,
+  "use_state": "DiT",
+  "loss_type": "robot",
+  "train_setup": {
+    "freeze_option": "freeze_vision_encoder"
+  },
+  "state_encoder": {
+    "state_dim": 212
+  },
+  "action_model": {
+    "model_type": "DiT-B",
+    "token_size": 2304,
+    "action_dim": 192,
+    "hidden_size": 1024
+  },
+  "vlm": {
+    "type": "PaliGemmaForConditionalGeneration",
+    "name": "paligemma",
+    "pretrained_model_name_or_path": "/sjw_alinlab/home/seunghoon/vitra_cotrain/models/paligemma2-3b-mix-224"
+  },
+  "trainer": {
+    "sharding_strategy": "shard-grad-op",
+    "strategy": "fsdp_paligemma_with_checkpointing",
+    "lr_scheduler_type": "backbone-freeze-warmup",
+    "learning_rate": 3.5e-06,
+    "gradient_clip_val": 1.0,
+    "weight_decay": 0.1,
+    "max_epochs": 100000,
+    "max_steps": 10000,
+    "reduce_in_full_precision": true,
+    "enable_mixed_precision_training": false,
+    "enable_gradient_checkpointing": true,
+    "action_model_learning_rate": 3.5e-06,
+    "llm_freeze_step": 0,
+    "warmup_ratio": null,
+    "optimizer_betas": [
+      0.9,
+      0.95
+    ]
+  },
+  "train_dataset": {
+    "data_root_dir": "/sjw_alinlab/home/seunghoon/vitra_cotrain/subsets/H100_R000",
+    "augmentation": true,
+    "set_none_ratio": 0.0,
+    "data_mix": "inspire_cotrain",
+    "num_workers": 8,
+    "prefetch_factor": null,
+    "flip_augmentation": 1.0,
+    "action_type": "angle",
+    "use_rel": false,
+    "clip_len": null,
+    "normalization": true,
+    "state_mask_prob": 0.1,
+    "mixture": {
+      "human_dataset": "vitra_human",
+      "robot_dataset": "inspire_robot",
+      "human_ratio": 100.0,
+      "robot_ratio": 0.0,
+      "samples_per_epoch": 100000,
+      "human_root_dir": null,
+      "robot_root_dir": null,
+      "human_statistics_path": null,
+      "robot_statistics_path": null
+    }
+  },
+  "repeated_diffusion_steps": 2,
+  "config": "/tmp/tmp.LMzFpMvuD8.json",
+  "data_mix": null,
+  "debug": false,
+  "num_workers": null,
+  "prefetch_factor": null
+}