Upload policy weights, train config and readme

README.md

@@ -5,9 +5,9 @@ license: apache-2.0
 model_name: pi05
 pipeline_tag: robotics
 tags:
+- robotics
 - pi05
 - lerobot
-- robotics
 ---
 
 # 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": "lerobot/pi05_base",
+  "base_model_name_or_path": "lerobot/pi05_libero_finetuned",
   "bias": "none",
   "corda_config": null,
   "ensure_weight_tying": false,
@@ -28,7 +28,7 @@
   "peft_type": "LORA",
   "peft_version": "0.18.1",
   "qalora_group_size": 16,
-  "r": 24,
+  "r": 2,
   "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))",

adapter_model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e8b0efeffa511fc165655e18a4bae394f04e883ee0109a40bba87620eafe205d
-size 7735976
+oid sha256:4b410c2504ed69d53a27cddfe797c1ebad84d2e5e3665498d27e0ee767b98100
+size 656312

config.json

@@ -41,10 +41,10 @@
     "private": null,
     "tags": null,
     "license": null,
-    "pretrained_path": "lerobot/pi05_base",
+    "pretrained_path": "lerobot/pi05_libero_finetuned",
     "paligemma_variant": "gemma_2b",
     "action_expert_variant": "gemma_300m",
-    "dtype": "float32",
+    "dtype": "bfloat16",
     "chunk_size": 50,
     "n_action_steps": 50,
     "max_state_dim": 32,
@@ -56,7 +56,7 @@
     "time_sampling_offset": 0.001,
     "min_period": 0.004,
     "max_period": 4.0,
-    "use_relative_actions": false,
+    "use_relative_actions": true,
     "relative_exclude_joints": [
         "gripper"
     ],
@@ -70,17 +70,20 @@
     ],
     "empty_cameras": 0,
     "tokenizer_max_length": 200,
+    "state_token_companding": "identity",
+    "state_token_companding_gain": 2.0,
+    "state_token_clip_before_discretize": false,
     "normalization_mapping": {
         "VISUAL": "IDENTITY",
         "STATE": "QUANTILES",
         "ACTION": "QUANTILES"
     },
     "gradient_checkpointing": true,
-    "compile_model": true,
+    "compile_model": false,
     "compile_mode": "max-autotune",
     "freeze_vision_encoder": false,
     "train_expert_only": false,
-    "optimizer_lr": 0.001,
+    "optimizer_lr": 1e-06,
     "optimizer_betas": [
         0.9,
         0.95

train_config.json

@@ -166,7 +166,10 @@
         "disable_env_checker": true,
         "task_ids": [
             0,
-            1
+            1,
+            2,
+            3,
+            4
         ],
         "episode_length": null,
         "obs_type": "pixels_agent_pos",
@@ -221,10 +224,10 @@
         "private": null,
         "tags": null,
         "license": null,
-        "pretrained_path": "lerobot/pi05_base",
+        "pretrained_path": "lerobot/pi05_libero_finetuned",
         "paligemma_variant": "gemma_2b",
         "action_expert_variant": "gemma_300m",
-        "dtype": "float32",
+        "dtype": "bfloat16",
         "chunk_size": 50,
         "n_action_steps": 50,
         "max_state_dim": 32,
@@ -236,7 +239,7 @@
         "time_sampling_offset": 0.001,
         "min_period": 0.004,
         "max_period": 4.0,
-        "use_relative_actions": false,
+        "use_relative_actions": true,
         "relative_exclude_joints": [
             "gripper"
         ],
@@ -250,17 +253,20 @@
         ],
         "empty_cameras": 0,
         "tokenizer_max_length": 200,
+        "state_token_companding": "identity",
+        "state_token_companding_gain": 2.0,
+        "state_token_clip_before_discretize": false,
         "normalization_mapping": {
             "VISUAL": "IDENTITY",
             "STATE": "QUANTILES",
             "ACTION": "QUANTILES"
         },
         "gradient_checkpointing": true,
-        "compile_model": true,
+        "compile_model": false,
         "compile_mode": "max-autotune",
         "freeze_vision_encoder": false,
         "train_expert_only": false,
-        "optimizer_lr": 0.001,
+        "optimizer_lr": 1e-06,
         "optimizer_betas": [
             0.9,
             0.95
@@ -278,17 +284,17 @@
     "seed": 1000,
     "cudnn_deterministic": false,
     "num_workers": 4,
-    "batch_size": 12,
-    "steps": 3,
-    "eval_freq": 20000,
-    "log_freq": 200,
+    "batch_size": 32,
+    "steps": 1,
+    "eval_freq": -1,
+    "log_freq": 1,
     "tolerance_s": 0.0001,
     "save_checkpoint": true,
-    "save_freq": 20000,
+    "save_freq": 1,
     "use_policy_training_preset": true,
     "optimizer": {
         "type": "adamw",
-        "lr": 0.001,
+        "lr": 1e-06,
         "weight_decay": 0.01,
         "grad_clip_norm": 1.0,
         "betas": [
@@ -301,12 +307,12 @@
         "type": "cosine_decay_with_warmup",
         "num_warmup_steps": 1000,
         "num_decay_steps": 30000,
-        "peak_lr": 0.001,
+        "peak_lr": 1e-06,
         "decay_lr": 0.0001
     },
     "eval": {
         "n_episodes": 10,
-        "batch_size": 1,
+        "batch_size": 2,
         "use_async_envs": false
     },
     "wandb": {
@@ -315,7 +321,7 @@
         "project": "lerobot",
         "entity": null,
         "notes": null,
-        "run_id": "cyectavr",
+        "run_id": "olje1m16",
         "mode": null,
         "add_tags": true
     },
@@ -324,7 +330,7 @@
         "full_training_modules": null,
         "method_type": "LORA",
         "init_type": null,
-        "r": 24
+        "r": 2
     },
     "use_rabc": false,
     "rabc_progress_path": null,