ayo/ipex-llm
1
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions 1

LLM: fix pretraining_tp argument issue. (#9281)

Browse source
This commit is contained in:
Cengguang Zhang 2023-10-26 18:43:58 +08:00 committed by GitHub
parent 6b2a32eba2
commit 44b5fcc190
1 changed files with 4 additions and 1 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

5
python/llm/src/bigdl/llm/transformers/model.py
View file

@ -93,7 +93,10 @@ class _BaseAutoModelClass:
kwargs["torch_dtype"] = kwargs.get("torch_dtype", 'auto') kwargs["torch_dtype"] = kwargs.get("torch_dtype", 'auto')
# Avoid tensor parallel F.Linear Operations # Avoid tensor parallel F.Linear Operations
if "pretraining_tp" in config_dict: if "pretraining_tp" in config_dict:
kwargs["pretraining_tp"] = 1 if "config" in kwargs:
setattr(kwargs["config"], "pretraining_tp", 1)
else:
kwargs["pretraining_tp"] = 1
q_k = load_in_low_bit if load_in_low_bit else "sym_int4" q_k = load_in_low_bit if load_in_low_bit else "sym_int4"
model = cls.load_convert(q_k, optimize_model, *args, **kwargs) model = cls.load_convert(q_k, optimize_model, *args, **kwargs)
else: else: