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

disable arc ut (#9825)

Browse source
This commit is contained in:
Mingyu Wei 2024-01-03 18:10:34 +08:00 committed by GitHub
parent 20e9742fa0
commit f4eb5da42d
1 changed files with 27 additions and 27 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

54
python/llm/test/inference_gpu/test_transformers_api.py
View file

@ -186,42 +186,42 @@ class Test_Optimize_Gpu_Model:
assert all(max_diff <= lower_bound for max_diff in max_diff_tensor) assert all(max_diff <= lower_bound for max_diff in max_diff_tensor)
def test_falcon_gpu_model(self): # def test_falcon_gpu_model(self):
Model = AutoModelForCausalLM # Model = AutoModelForCausalLM
Tokenizer = AutoTokenizer # Tokenizer = AutoTokenizer
model_path = os.environ.get('FALCON_7B_ORIGIN_PATH') # model_path = os.environ.get('FALCON_7B_ORIGIN_PATH')
# currently only compare the output of the last self-attention layer. # # currently only compare the output of the last self-attention layer.
layer_norm = "transformer.h.31.input_layernorm" # layer_norm = "transformer.h.31.input_layernorm"
self_attn = "transformer.h.31.self_attention" # self_attn = "transformer.h.31.self_attention"
lower_bound = 0 # lower_bound = 0
self.run_optimize_gpu_model(Model, Tokenizer, model_path, self_attn, layer_norm, lower_bound) # self.run_optimize_gpu_model(Model, Tokenizer, model_path, self_attn, layer_norm, lower_bound)
def test_llama_gpu_model(self): # def test_llama_gpu_model(self):
Model = AutoModelForCausalLM # Model = AutoModelForCausalLM
Tokenizer = AutoTokenizer # Tokenizer = AutoTokenizer
model_path = os.environ.get('LLAMA2_7B_ORIGIN_PATH') # model_path = os.environ.get('LLAMA2_7B_ORIGIN_PATH')
# currently only compare the output of the last self-attention layer. # # currently only compare the output of the last self-attention layer.
layer_norm = "model.layers.31.input_layernorm" # layer_norm = "model.layers.31.input_layernorm"
self_attn = "model.layers.31.self_attn" # self_attn = "model.layers.31.self_attn"
lower_bound = 5e-2 # lower_bound = 5e-2
self.run_optimize_gpu_model(Model, Tokenizer, model_path, self_attn, layer_norm, lower_bound) # self.run_optimize_gpu_model(Model, Tokenizer, model_path, self_attn, layer_norm, lower_bound)
def test_chatglm2_gpu_model(self): # def test_chatglm2_gpu_model(self):
Model = AutoModel # Model = AutoModel
Tokenizer = AutoTokenizer # Tokenizer = AutoTokenizer
model_path = os.environ.get('CHATGLM2_6B_ORIGIN_PATH') # model_path = os.environ.get('CHATGLM2_6B_ORIGIN_PATH')
# currently only need to compare the output of one self-attention layer. # # currently only need to compare the output of one self-attention layer.
layer_norm = "transformer.encoder.layers.27.input_layernorm" # layer_norm = "transformer.encoder.layers.27.input_layernorm"
self_attn = "transformer.encoder.layers.27.self_attention" # self_attn = "transformer.encoder.layers.27.self_attention"
lower_bound = 1e-3 # lower_bound = 1e-3
self.run_optimize_gpu_model(Model, Tokenizer, model_path, self_attn, layer_norm, lower_bound) # self.run_optimize_gpu_model(Model, Tokenizer, model_path, self_attn, layer_norm, lower_bound)
if __name__ == '__main__': if __name__ == '__main__':