TAPA / howto /customize_paths.md
xuxw98's picture
Upload 58 files
7d52396
|
raw
history blame
876 Bytes
## Customize paths
The project is setup to use specific paths to read the original weights and save checkpoints etc.
For all scripts, you can run
```shell
python script.py -h
```
to get a list of available options. For instance, here's how you would modify the checkpoint dir:
```shell
python scripts/convert_checkpoint.py --checkpoint_dir "data/checkpoints/foo"
```
Note that this change will need to be passed along to subsequent steps, for example:
```shell
python generate.py \
--checkpoint_path "data/checkpoints/foo/7B/lit-llama.pth" \
--tokenizer_path "data/checkpoints/foo/tokenizer.model"
```
and
```shell
python quantize/gptq.py \
--checkpoint_path "data/checkpoints/foo/7B/lit-llama.pth" \
--tokenizer_path "data/checkpoints/foo/tokenizer.model"
```
To avoid this, you can use symbolic links to create shortcuts and avoid passing different paths.