Upload LAM fine-tuned checkpoint (epoch 17, best val_loss=9.68e-5)

README.md

@@ -0,0 +1,56 @@
+# LAM Fine-Tuned on Bin-Pick-Pack
+
+Fine-tuned [DreamDojo LAM](https://arxiv.org/abs/2504.02024) (Latent Action Model, 710M params) on the [bin_pick_pack_coffee_capsules](https://huggingface.co/datasets/villekuosmanen/bin_pick_pack_coffee_capsules) manipulation dataset.
+
+## Training Details
+
+- **Base model**: LAM_400k.ckpt (pre-trained on GR1 humanoid data)
+- **Dataset**: villekuosmanen/bin_pick_pack_coffee_capsules (42846 train pairs, 4819 val pairs)
+- **Resolution**: 240x320
+- **Epochs**: 57 completed (best at epoch 17, stopped early)
+- **Batch size**: 32
+- **Learning rate**: 1e-5
+- **Weight decay**: 0.01
+- **KL beta**: 1e-6
+- **Gradient clipping**: 0.3
+- **Hardware**: NVIDIA GH200 (Isambard HPC)
+- **Training time**: ~12h
+
+## Results
+
+| Metric | Epoch 0 | Epoch 17 (best) | Epoch 56 (final) |
+|--------|---------|-----------------|------------------|
+| train_loss | 0.000154 | 0.000076 | 0.000058 |
+| val_loss | 0.000137 | 0.000097 | 0.000105 |
+| val_mse | 0.000107 | 0.000080 | 0.000092 |
+| val_kl | 29.35 | 16.57 | 12.83 |
+
+Val loss improved until epoch 17 then plateaued around 1.0e-4. Train loss continued decreasing. Mild overfitting but no divergence.
+
+## Checkpoint
+
+- **File**: `best.pt` (params only, 2.84 GB)
+- **Contents**: `model_state_dict`, `epoch`, `step`, `best_loss`
+- **SHA-256**: `72e746704080266c7c6aa265035de3bd2132b9ad2783dbfe8d9fc82670a838dc`
+
+Verify with:
+```bash
+sha256sum best.pt
+```
+
+## Usage
+
+```python
+import torch
+
+ckpt = torch.load("best.pt", map_location="cpu", weights_only=False)
+model.load_state_dict(ckpt["model_state_dict"])
+```
+
+## Config
+
+See `lam_finetune_isambard.yaml` for the full training configuration.
+
+## W&B
+
+Training curves: [wandb.ai/pravsels/lam-finetune/runs/afu3164m](https://wandb.ai/pravsels/lam-finetune/runs/afu3164m)

best.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:72e746704080266c7c6aa265035de3bd2132b9ad2783dbfe8d9fc82670a838dc
+size 2839391882

lam_finetune_isambard.yaml

@@ -0,0 +1,30 @@
+# LAM fine-tuning config for Isambard GH200.
+# Heavy artifacts live on scratch; repo code stays under /home.
+
+dataset:
+  repo_id: villekuosmanen/bin_pick_pack_coffee_capsules
+  root: /scratch/u6cr/pravsels.u6cr/rsl_rl_rwm/data/lerobot
+  hf_home: /scratch/u6cr/pravsels.u6cr/rsl_rl_rwm/huggingface
+
+training:
+  ckpt_path: /scratch/u6cr/pravsels.u6cr/rsl_rl_rwm/checkpoints/LAM_400k.ckpt
+  resolution_h: 240
+  resolution_w: 320
+  batch_size: 32
+  max_epochs: 100
+  learning_rate: 0.00001
+  weight_decay: 0.01
+  beta: 0.000001
+  grad_clip: 0.3
+  val_ratio: 0.1
+  split_seed: 0
+  num_workers: 8
+  device: cuda
+  output_dir: /scratch/u6cr/pravsels.u6cr/rsl_rl_rwm/runs/lam_finetune
+  save_every_n_epochs: 10
+  log_every_n_steps: 50
+  wandb:
+    enabled: true
+    project: lam-finetune
+    entity: pravsels
+    mode: offline