deftdawg/bitsandbytes-rocm
1
0
Fork 0
forked from mrq/bitsandbytes-rocm
Code Pull Requests Activity

Merge pull request #13 from chessgecko/patch-1

Browse Source 
fix param name
This commit is contained in:
Tim Dettmers 2022-09-05 16:10:47 -07:00 committed by GitHub
commit 4e4668ab09
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
1 changed files with 3 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

6
README.md
View File

@ -23,12 +23,12 @@ Resources:
1. Comment out torch.nn.Linear: ``#linear = torch.nn.Linear(...)``
2. Add bnb 8-bit linear light module: ``linear = bnb.nn.Linear8bitLt(...)`` (base arguments stay the same)
3. There are two modes:
- Mixed 8-bit training with 16-bit main weights. Pass the argument ``use_fp16_weights=True`` (default)
- Int8 inference. Pass the argument ``use_fp16_weights=False``
- Mixed 8-bit training with 16-bit main weights. Pass the argument ``has_fp16_weights=True`` (default)
- Int8 inference. Pass the argument ``has_fp16_weights=False``
4. To use the full LLM.int8() method, use the ``threshold=k`` argument. We recommend ``k=6.0``.
```python
# LLM.int8()
linear = bnb.nn.Linear8bitLt(dim1, dim2, bias=True, use_fp16_weights=False, threshold=6.0)
linear = bnb.nn.Linear8bitLt(dim1, dim2, bias=True, has_fp16_weights=False, threshold=6.0)
# inputs need to be fp16
out = linear(x.to(torch.float16))
```