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

add optimize model option (#9530)

Browse source
This commit is contained in:
Chen, Zhentao 2023-11-24 17:10:49 +08:00 committed by GitHub
parent 6bec0faea5
commit 45820cf3b9
2 changed files with 5 additions and 4 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

7
python/llm/dev/benchmark/harness/bigdl_llm.py
View file

@ -46,9 +46,10 @@ class BigDLLM(BaseLM):
tokenizer=None, tokenizer=None,
batch_size=1, batch_size=1,
load_in_8bit: Optional[bool] = False, load_in_8bit: Optional[bool] = False,
trust_remote_code: Optional[bool] = False, trust_remote_code: Optional[bool] = True,
load_in_low_bit=None, load_in_low_bit=None,
dtype: Optional[Union[str, torch.dtype]] = "auto", dtype: Optional[Union[str, torch.dtype]] = "auto",
**kwargs
): ):
super().__init__() super().__init__()
@ -58,8 +59,8 @@ class BigDLLM(BaseLM):
import intel_extension_for_pytorch as ipex import intel_extension_for_pytorch as ipex
model = AutoModelForCausalLM.from_pretrained(pretrained, model = AutoModelForCausalLM.from_pretrained(pretrained,
load_in_low_bit=load_in_low_bit, load_in_low_bit=load_in_low_bit,
optimize_model=True, optimize_model=kwargs.get('optimize_model', True),
trust_remote_code=True, trust_remote_code=trust_remote_code,
use_cache=True, use_cache=True,
torch_dtype=_get_dtype(dtype)) torch_dtype=_get_dtype(dtype))
print(model) # print model to check precision print(model) # print model to check precision

2
python/llm/dev/benchmark/harness/run_llb.py
View file

@ -90,7 +90,7 @@ def main():
prec_arg = parse_precision(prec, args.model) prec_arg = parse_precision(prec, args.model)
model_args = f"pretrained={args.pretrained},{prec_arg}" model_args = f"pretrained={args.pretrained},{prec_arg}"
if len(args.model_args) > 0: if len(args.model_args) > 0:
model_args += args.model_args model_args = f"{model_args},{args.model_args}"
for task in args.tasks: for task in args.tasks:
task_names=task_map.get(task, task).split(',') task_names=task_map.get(task, task).split(',')
num_fewshot = task_to_n_few_shots.get(task, args.num_fewshot) num_fewshot = task_to_n_few_shots.get(task, args.num_fewshot)