Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Allow Unsloth Dynamic 4bit BnB quants to work #12974

Merged
merged 4 commits into from
Feb 13, 2025
+10 −2
Merged

Allow Unsloth Dynamic 4bit BnB quants to work #12974

Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
bf1a29e
Update bitsandbytes.py
danielhanchen Feb 9, 2025
3c5a5c0
Update vllm/model_executor/layers/quantization/bitsandbytes.py
danielhanchen Feb 10, 2025
4bdf8d8
Update vllm/model_executor/layers/quantization/bitsandbytes.py
danielhanchen Feb 10, 2025
87ab651
Update bitsandbytes.py
danielhanchen Feb 10, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
12 changes: 10 additions & 2 deletions vllm/model_executor/layers/quantization/bitsandbytes.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -133,8 +133,16 @@ def is_layer_skipped_bnb(prefix: str, llm_int8_skip_modules: List[str]):
components = prefix.split('.')

# Check if any of the skip modules exactly matches any component
return any(module_name in components
for module_name in llm_int8_skip_modules)
substr_check = any(module_name in components
for module_name in llm_int8_skip_modules)

# Allow certain layers to not be quantized
set_components = set(".".join(components[:i + 1])
for i in range(len(components)))
set_llm_int8_skip_modules = set(llm_int8_skip_modules)
prefix_check = len(set_llm_int8_skip_modules & set_components) != 0

return substr_check or prefix_check


class BitsAndBytesLinearMethod(LinearMethodBase):
Expand Down