---
license: other
tags:
  - robotics
  - fastwam
  - flashwam
  - place-cube
---

# pcbnew100_flashwam_scratch — FlashWAM (M1 FusedKV, RoPE-fixed), from scratch, place_cube_new 100 traj

FlashWAM (M1 FusedKV, RoPE-fixed) action-video model trained **from scratch** (`resume: null` = no LIBERO
init) on the **100-trajectory** "pick and place new" dataset
(`place_cube_new_lerobot_v21`, 100 episodes; task *"place the cube in the bowl"*).

- **From scratch:** resume = `None` (scratch init: True).
- **Model:** FlashWAM (M1 FusedKV, RoPE-fixed); `kv_source_mode=fused_kv`, `fixed_rope=true`, `decoupled=true`.
- **Recipe:** num_epochs=30, save_every=2445, global batch 32, lr 1e-4, bf16.
- **Status:** Training **completed**.

## Checkpoints (`checkpoints/weights/`)

| file | size |
|------|------|
| `step_002445.pt` | 10.13 GB |
| `step_004890.pt` | 10.13 GB |
| `step_007335.pt` | 10.13 GB |
| `step_009780.pt` | 10.13 GB |
| `step_012225.pt` | 10.13 GB |
| `step_014670.pt` | 10.13 GB |

Weights-only checkpoints. Pair each with THIS run's `dataset_stats.json` for
action/proprio (de)normalization.

## Other files
- `config.yaml` — full training/model config.
- `dataset_stats.json` — per-run normalization stats.