| ## 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. | |