Upload policy weights, train config and readme

README.md

@@ -6,8 +6,8 @@ model_name: pi05
 pipeline_tag: robotics
 tags:
 - pi05
-- lerobot
 - robotics
+- lerobot
 ---
 
 # Model Card for pi05

adapter_config.json

@@ -6,7 +6,7 @@
     "base_model_class": "PI05Policy",
     "parent_library": "lerobot.policies.pi05.modeling_pi05"
   },
-  "base_model_name_or_path": "/data/taojiachen/Continuous_VLA/exp/sequential/spatial/seed42/task30/train/checkpoints/last/pretrained_model",
+  "base_model_name_or_path": "/data/taojiachen/Continuous_VLA/exp/er/spatial/seed42/task30/train/checkpoints/last/pretrained_model",
   "bias": "none",
   "corda_config": null,
   "ensure_weight_tying": false,
@@ -24,14 +24,16 @@
   "lora_dropout": 0.0,
   "megatron_config": null,
   "megatron_core": "megatron.core",
-  "modules_to_save": [],
+  "modules_to_save": [
+    ".*\\.vision_tower\\..*"
+  ],
   "peft_type": "LORA",
   "peft_version": "0.18.1",
   "qalora_group_size": 16,
   "r": 16,
   "rank_pattern": {},
   "revision": null,
-  "target_modules": "(.*\\.gemma_expert\\..*\\.self_attn\\.(q|v)_proj|model\\.(state_proj|action_in_proj|action_out_proj|action_time_mlp_in|action_time_mlp_out))",
+  "target_modules": ".*\\.paligemma\\.model\\.language_model\\..*\\.self_attn\\.(q|v)_proj|.*\\.gemma_expert\\..*\\.self_attn\\.(q|v)_proj|model\\.(state_proj|action_in_proj|action_out_proj|action_time_mlp_in|action_time_mlp_out)",
   "target_parameters": null,
   "task_type": null,
   "trainable_token_indices": null,

adapter_model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:720d80fd50305b29ba054075af898bb09ceb7fee63817252058099f88023c90c
-size 5161624
+oid sha256:08fba12e4837a1641214136229f4683c7abc8350bbee36cbe960d52612a91032
+size 12547808

config.json

@@ -41,7 +41,7 @@
     "private": null,
     "tags": null,
     "license": null,
-    "pretrained_path": "/data/taojiachen/Continuous_VLA/exp/sequential/spatial/seed42/task30/train/checkpoints/last/pretrained_model",
+    "pretrained_path": "/data/taojiachen/Continuous_VLA/exp/er/spatial/seed42/task30/train/checkpoints/last/pretrained_model",
     "paligemma_variant": "gemma_2b",
     "action_expert_variant": "gemma_300m",
     "dtype": "bfloat16",

train_config.json

@@ -4,22 +4,28 @@
         "root": null,
         "episodes": [
             31,
+            70,
             71,
             111,
             151,
             191,
             231,
             271,
+            310,
             311,
+            350,
             351,
             391,
             431,
             471,
             511,
             551,
+            590,
             591,
+            630,
             631,
             671,
+            710,
             711,
             751,
             791,
@@ -43,6 +49,7 @@
             1511,
             1551,
             1591,
+            1630,
             1631,
             1671
         ],
@@ -166,7 +173,7 @@
         "private": null,
         "tags": null,
         "license": null,
-        "pretrained_path": "/data/taojiachen/Continuous_VLA/exp/sequential/spatial/seed42/task30/train/checkpoints/last/pretrained_model",
+        "pretrained_path": "/data/taojiachen/Continuous_VLA/exp/er/spatial/seed42/task30/train/checkpoints/last/pretrained_model",
         "paligemma_variant": "gemma_2b",
         "action_expert_variant": "gemma_300m",
         "dtype": "bfloat16",
@@ -210,7 +217,7 @@
         "scheduler_decay_steps": 30000,
         "scheduler_decay_lr": 2.5e-06
     },
-    "output_dir": "/data/taojiachen/Continuous_VLA/exp/sequential/spatial/seed42/task31/train",
+    "output_dir": "/data/taojiachen/Continuous_VLA/exp/er/spatial/seed42/task31/train",
     "job_name": "pi05",
     "resume": false,
     "seed": 42,
@@ -257,8 +264,10 @@
         "mode": null
     },
     "peft": {
-        "target_modules": null,
-        "full_training_modules": null,
+        "target_modules": ".*\\.paligemma\\.model\\.language_model\\..*\\.self_attn\\.(q|v)_proj|.*\\.gemma_expert\\..*\\.self_attn\\.(q|v)_proj|model\\.(state_proj|action_in_proj|action_out_proj|action_time_mlp_in|action_time_mlp_out)",
+        "full_training_modules": [
+            ".*\\.vision_tower\\..*"
+        ],
         "method_type": "LORA",
         "init_type": null,
         "r": 16