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

force lm_head optimization in any model if set environment variable (#11830)

Browse source
This commit is contained in:
Yishuo Wang 2024-08-16 16:48:45 +08:00 committed by GitHub
parent 3b630fb9df
commit e966e85df8
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
1 changed files with 8 additions and 8 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

16
python/llm/src/ipex_llm/transformers/convert.py
View file

@ -400,14 +400,14 @@ def _replace_with_low_bit_linear(model, qtype, modules_to_not_convert=None,
if is_linear and not isinstance(module, LowBitLinear):
in_features, out_features, mp_group = linear_args
optimize_lm_head = False
if is_lm_head(name, model_config, out_features):
model_type = getattr(model_config, "model_type", None)
if model_type in ["gptj", "llama", "qwen2"]:
if os.environ.get("IPEX_LLM_LAST_LM_HEAD", None) is not None:
optimize_lm_head = os.environ.get("IPEX_LLM_LAST_LM_HEAD", None) == "1"
elif os.environ.get("IPEX_LLM_LOW_MEM", None) is not None:
optimize_lm_head = os.environ.get("IPEX_LLM_LOW_MEM", None) == "1"
optimize_lm_head = (
is_lm_head(name, model_config, out_features)
and (
os.environ.get("IPEX_LLM_LAST_LM_HEAD", "0") == "1"
or os.environ.get("IPEX_LLM_LOW_MEM", "0") == "1"
and getattr(model_config, "model_type", "") in ["gptj", "llama", "qwen2"]
)
)
with init_empty_weights():
new_linear = None
is_gptq = is_gptq_linear(module)