admin/lerobot
1
0
Fork 0
mirror of https://github.com/huggingface/lerobot.git synced 2026-07-16 14:32:03 +00:00
Code Issues Packages Projects Releases Wiki Activity

feat(pi0-fast): support automatic tokenizer fitting

Browse Source
This commit is contained in:
Pepijn
2026-07-15 15:46:04 +02:00
parent 2749cf7767
commit 696e68869c
8 changed files with 133 additions and 41 deletions
Download Patch File Download Diff File Expand all files Collapse all files
docs/source/pi0fast.mdx
+15 -9
View File
@@ -109,15 +109,21 @@ lerobot-train \
### Key Training Parameters
| Parameter | Description | Default |
| -------------------------------------- | -------------------------------------------------- | ------------------------------- |
| `--policy.gradient_checkpointing=true` | Reduces memory usage significantly during training | `false` |
| `--policy.dtype=bfloat16` | Use mixed precision training for efficiency | `float32` |
| `--policy.chunk_size` | Number of action steps to predict (action horizon) | `50` |
| `--policy.n_action_steps` | Number of action steps to execute | `50` |
| `--policy.max_action_tokens` | Maximum number of FAST tokens per action chunk | `256` |
| `--policy.action_tokenizer_name` | FAST tokenizer to use | `lerobot/fast-action-tokenizer` |
| `--policy.compile_model=true` | Enable torch.compile for faster training | `false` |
| Parameter | Description | Default |
| --------------------------------------- | -------------------------------------------------- | ------------------------------- |
| `--policy.gradient_checkpointing=true` | Reduces memory usage significantly during training | `false` |
| `--policy.dtype=bfloat16` | Use mixed precision training for efficiency | `float32` |
| `--policy.chunk_size` | Number of action steps to predict (action horizon) | `50` |
| `--policy.n_action_steps` | Number of action steps to execute | `50` |
| `--policy.max_action_tokens` | Maximum number of FAST tokens per action chunk | `256` |
| `--policy.action_tokenizer_name` | FAST tokenizer to use | `lerobot/fast-action-tokenizer` |
| `--policy.auto_fit_fast_tokenizer=true` | Fit and cache a tokenizer for the training dataset | `false` |
| `--policy.compile_model=true` | Enable torch.compile for faster training | `false` |
Set `--policy.auto_fit_fast_tokenizer=true` to sample action chunks from the
training dataset and cache a fitted tokenizer under
`~/.cache/lerobot/fast_tokenizers`. This also works when fine-tuning with
`--policy.path`; leave it disabled to retain the checkpoint's tokenizer.
## Inference