Upload SAE model weights, config, and training state

README.md

@@ -6,13 +6,10 @@ tags:
 - Robotics
 datasets:
 - villekuosmanen/build_block_tower
-- villekuosmanen/dAgger_build_block_tower_1.0.0
-- villekuosmanen/dAgger_build_block_tower_1.1.0
-- villekuosmanen/dAgger_build_block_tower_1.2.0
-- villekuosmanen/dAgger_build_block_tower_1.3.0
-- villekuosmanen/dAgger_build_block_tower_1.4.0
 - villekuosmanen/fail_build_block_tower_stationary
-- villekuosmanen/fail_build_block_tower_autonomous_interaction
+- villekuosmanen/build_block_tower_val
+- villekuosmanen/dAgger_build_block_tower_1.4.0
+- villekuosmanen/dAgger_build_block_tower_dino
 library_name: physical-ai-interpretability
 ---
 
@@ -23,7 +20,7 @@ This model is a Sparse Autoencoder trained for interpretability analysis of robo
 ## Model Details
 
 - **Architecture**: Multi-modal Sparse Autoencoder
-- **Training Dataset**: `villekuosmanen/build_block_tower`, `villekuosmanen/dAgger_build_block_tower_1.0.0`, `villekuosmanen/dAgger_build_block_tower_1.1.0`, `villekuosmanen/dAgger_build_block_tower_1.2.0`, `villekuosmanen/dAgger_build_block_tower_1.3.0`, `villekuosmanen/dAgger_build_block_tower_1.4.0`, `villekuosmanen/fail_build_block_tower_stationary`, `villekuosmanen/fail_build_block_tower_autonomous_interaction`
+- **Training Dataset**: `villekuosmanen/build_block_tower`, `villekuosmanen/fail_build_block_tower_stationary`, `villekuosmanen/build_block_tower_val`, `villekuosmanen/dAgger_build_block_tower_1.4.0`, `villekuosmanen/dAgger_build_block_tower_dino`
 - **Base Policy**: LeRobot ACT policy
 - **Layer Target**: `model.encoder.layers.3.norm2`
 - **Tokens**: 77
@@ -36,7 +33,7 @@ This model is a Sparse Autoencoder trained for interpretability analysis of robo
 - **Learning Rate**: 0.0001
 - **Batch Size**: 16
 - **L1 Penalty**: 0.3
-- **Epochs**: 20
+- **Epochs**: 10
 - **Optimizer**: adam
 
 ## Usage

config.json

@@ -14,7 +14,7 @@
   "block_size": 8,
   "batch_size": 16,
   "learning_rate": 0.0001,
-  "num_epochs": 20,
+  "num_epochs": 10,
   "validation_split": 0.1,
   "l1_penalty": 0.3,
   "optimizer": "adam",
@@ -28,9 +28,9 @@
   "save_every": 1000,
   "validate_every": 500,
   "device": "cuda",
-  "repo_id": "[villekuosmanen/build_block_tower, villekuosmanen/dAgger_build_block_tower_1.0.0, villekuosmanen/dAgger_build_block_tower_1.1.0, villekuosmanen/dAgger_build_block_tower_1.2.0, villekuosmanen/dAgger_build_block_tower_1.3.0, villekuosmanen/dAgger_build_block_tower_1.4.0, villekuosmanen/fail_build_block_tower_stationary, villekuosmanen/fail_build_block_tower_autonomous_interaction]",
-  "repo_hash": "881292d6",
+  "repo_id": "[villekuosmanen/build_block_tower, villekuosmanen/fail_build_block_tower_stationary, villekuosmanen/build_block_tower_val, villekuosmanen/dAgger_build_block_tower_1.4.0, villekuosmanen/dAgger_build_block_tower_dino]",
+  "repo_hash": "b7b921b3",
   "layer_name": "model.encoder.layers.3.norm2",
   "activation_cache_path": "/home/ville/.cache/physical_ai_interpretability/sae_activations",
-  "experiment_name": "sae_fail_build_block_tow_881292d6"
+  "experiment_name": "sae_dAgger_build_block_t_b7b921b3"
 }

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0211c1fe33e69f654dd269531f5aee5e04b26a56a7552ab4e5a7f14ea2449906
+oid sha256:9f72613854f5cab813362329d2318e5b6153efa6cba7ce2c7413cf0911d9cb7b
 size 971496408

training_state.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d142098a493441e4ab7adbb34caebce779e5659f678299accf1253b46743bd21
-size 1942998303
+oid sha256:85025084e4f82e20cd2d33b4a1276fe6a1d1bd64a1848a1ebb8744daa6ce749f
+size 1942998285