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

LLM: skip some model tests using certain api (#9163)

Browse source 
* LLM: Skip some model tests using certain api

* initialize variable named result
This commit is contained in:
WeiguangHan 2023-10-18 09:39:27 +08:00 committed by GitHub
parent 09815f7064
commit b9194c5786
1 changed files with 12 additions and 9 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

5
python/llm/dev/benchmark/all-in-one/run.py
View file

@ -40,6 +40,7 @@ results = []
def run_model(repo_id, test_api, in_out_pairs, local_model_hub=None, warm_up=1, num_trials=3, num_beams=1): def run_model(repo_id, test_api, in_out_pairs, local_model_hub=None, warm_up=1, num_trials=3, num_beams=1):
# TODO: make a parameter # TODO: make a parameter
result= {}
if test_api == 'transformer_int4': if test_api == 'transformer_int4':
result = run_transformer_int4(repo_id, local_model_hub, in_out_pairs, warm_up, num_trials, num_beams) result = run_transformer_int4(repo_id, local_model_hub, in_out_pairs, warm_up, num_trials, num_beams)
elif test_api == 'native_int4': elif test_api == 'native_int4':
@ -56,6 +57,7 @@ def run_model(repo_id, test_api, in_out_pairs, local_model_hub=None, warm_up=1,
result = run_ipex_fp16_gpu(repo_id, local_model_hub, in_out_pairs, warm_up, num_trials, num_beams) result = run_ipex_fp16_gpu(repo_id, local_model_hub, in_out_pairs, warm_up, num_trials, num_beams)
for in_out_pair in in_out_pairs: for in_out_pair in in_out_pairs:
if result:
results.append([repo_id, results.append([repo_id,
np.mean(result[in_out_pair], axis=0)[0], np.mean(result[in_out_pair], axis=0)[0],
np.mean(result[in_out_pair], axis=0)[1], np.mean(result[in_out_pair], axis=0)[1],
@ -192,7 +194,8 @@ def run_pytorch_autocast_bf16(repo_id,
st = time.perf_counter() st = time.perf_counter()
if repo_id in ['THUDM/chatglm-6b', 'THUDM/chatglm2-6b']: if repo_id in ['THUDM/chatglm-6b', 'THUDM/chatglm2-6b']:
# TODO: need verify chatglm family run bf16. # TODO: need verify chatglm family run bf16.
invalidInputError(False, "Currently pytorch do not support bfloat16 on cpu for chatglm models.") print("Currently pytorch do not support bfloat16 on cpu for chatglm models. Will skip it")
return
elif repo_id in LLAMA_IDS: elif repo_id in LLAMA_IDS:
model = AutoModelForCausalLM.from_pretrained(model_path, trust_remote_code=True, torch_dtype=torch.bfloat16, model = AutoModelForCausalLM.from_pretrained(model_path, trust_remote_code=True, torch_dtype=torch.bfloat16,
use_cache=True) use_cache=True)