# tinker_cookbook.scripts.copy_checkpoint

[Source on GitHub](https://github.com/thinking-machines-lab/tinker-cookbook/blob/main/tinker_cookbook/scripts/copy_checkpoint.py)

Copy trainable Tinker weights into the currently authenticated account.

This copies a trainable Tinker checkpoint by loading the source weights and saving them as a new destination-owned checkpoint.

Note: this only works for trainable `weights/...` checkpoints, not sampler-only `sampler_weights/...` checkpoints.

## Usage:
```bash
python -m tinker_cookbook.scripts.copy_checkpoint \
 --source-path tinker://:train:0/weights/
```
# Prints: sampler_path: tinker://:train:0/sampler_weights/

To save into a specific destination project, pass `--destination-project-id`:
```bash
python -m tinker_cookbook.scripts.copy_checkpoint \
 --source-path tinker://:train:0/weights/ \
 --destination-project-id
```

For cross-org copies, pass `--source-access-token` or set `SRC_TINKER_ACCESS_TOKEN` to a token that can read the source checkpoint:
```bash
python -m tinker_cookbook.scripts.copy_checkpoint \
 --source-path tinker://:train:0/weights/ \
 --source-access-token "$SRC_TINKER_ACCESS_TOKEN" \
 --destination-project-id
```

By default, this script saves sampler weights; pass `--output-kind training` to save a trainable `weights/...` checkpoint instead.