turboderp-org/exllamav2
1
0
Fork 0
mirror of https://github.com/turboderp-org/exllamav2.git synced 2026-04-20 14:29:28 +00:00
Code Issues Packages Projects Releases Wiki Activity

Option to resume conversion job with no other args

Browse Source
This commit is contained in:
turboderp
2024-06-08 22:15:41 +02:00
parent de05ac696b
commit 6030517a6f
2 changed files with 35 additions and 18 deletions
Download Patch File Download Diff File Expand all files Collapse all files

9
doc/convert.md
View File

@@ -4,7 +4,7 @@
Here are the arguments to `convert.py`:
- **-i / --in_dir *directory***: _(required)_ The source model to convert, in HF format (FP16). The directory should
- **-i / --in_dir *directory***: _(required if not resuming)_ The source model to convert, in HF format (FP16). The directory should
contain at least a `config.json` file, a `tokenizer.model` file and one or more `.safetensors` files containing weights.
If there are multiple weights files, they will all be indexed and searched for the neccessary tensors, so sharded models are
supported.
@@ -132,6 +132,13 @@ python convert.py \
-b 4.5
```
If the working `-o` directory is not empty and you do not specify `-nr`, any existing job in that directory
will be resumed. You can resume a job with no other arguments:
```sh
python convert.py -o /mnt/temp/exl2/
```
### Notes
- If the conversion script seems to stop on the "Solving..." step, give it a moment. It's attempting to find the