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

fix arc rms norm UT (#10394)

Browse source
This commit is contained in:
Yishuo Wang 2024-03-13 13:09:15 +08:00 committed by GitHub
parent 0193f29411
commit ca58a69b97
1 changed files with 7 additions and 7 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

14
python/llm/test/inference_gpu/test_transformers_api_RMSNorm.py
View file

@ -109,7 +109,7 @@ class Test_Optimize_Gpu_Model:
del opt_model del opt_model
gc.collect() gc.collect()
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)
@pytest.mark.parametrize('Name, Model, Tokenizer, model_path',TEST_MODEL_LIST) @pytest.mark.parametrize('Name, Model, Tokenizer, model_path',TEST_MODEL_LIST)
def test_dynamic_functions(self, Name, Model, Tokenizer, model_path): def test_dynamic_functions(self, Name, Model, Tokenizer, model_path):
if Name == "Llama2-7B": if Name == "Llama2-7B":
@ -128,19 +128,19 @@ class Test_Optimize_Gpu_Model:
RMSNorm_layer = "model.layers.31.input_layernorm" RMSNorm_layer = "model.layers.31.input_layernorm"
lower_bound = 2e-6 lower_bound = 2e-6
self.run_optimize_gpu_model(Name, Model, Tokenizer, model_path, RMSNorm_layer, layer_before_RMSNorm, lower_bound) self.run_optimize_gpu_model(Name, Model, Tokenizer, model_path, RMSNorm_layer, layer_before_RMSNorm, lower_bound)
def Chatglm2_gpu_model(self, Name, Model, Tokenizer, model_path): def Chatglm2_gpu_model(self, Name, Model, Tokenizer, model_path):
layer_before_RMSNorm = "transformer.encoder.layers.26" layer_before_RMSNorm = "transformer.encoder.layers.26"
RMSNorm_layer = "transformer.encoder.layers.27.input_layernorm" RMSNorm_layer = "transformer.encoder.layers.27.input_layernorm"
lower_bound = 6e-6 lower_bound = 4e-6
self.run_optimize_gpu_model(Name, Model, Tokenizer, model_path, RMSNorm_layer, layer_before_RMSNorm, lower_bound) self.run_optimize_gpu_model(Name, Model, Tokenizer, model_path, RMSNorm_layer, layer_before_RMSNorm, lower_bound)
def Mistral_gpu_model(self, Name, Model, Tokenizer, model_path): def Mistral_gpu_model(self, Name, Model, Tokenizer, model_path):
layer_before_RMSNorm = "model.layers.30" layer_before_RMSNorm = "model.layers.30"
RMSNorm_layer = "model.layers.31.input_layernorm" RMSNorm_layer = "model.layers.31.input_layernorm"
lower_bound = 6e-6 lower_bound = 8e-6
self.run_optimize_gpu_model(Name, Model, Tokenizer, model_path, RMSNorm_layer, layer_before_RMSNorm, lower_bound) self.run_optimize_gpu_model(Name, Model, Tokenizer, model_path, RMSNorm_layer, layer_before_RMSNorm, lower_bound)
def Baichuan_gpu_model(self, Name, Model, Tokenizer, model_path): def Baichuan_gpu_model(self, Name, Model, Tokenizer, model_path):
layer_before_RMSNorm = "model.layers.30" layer_before_RMSNorm = "model.layers.30"
RMSNorm_layer = "model.layers.31.input_layernorm" RMSNorm_layer = "model.layers.31.input_layernorm"
@ -150,5 +150,5 @@ class Test_Optimize_Gpu_Model:
def Qwen_gpu_model(self, Name, Model, Tokenizer, model_path): def Qwen_gpu_model(self, Name, Model, Tokenizer, model_path):
layer_before_RMSNorm = "transformer.h.30" layer_before_RMSNorm = "transformer.h.30"
RMSNorm_layer = "transformer.h.31.ln_1" RMSNorm_layer = "transformer.h.31.ln_1"
lower_bound = 2e-6 lower_bound = 4e-6
self.run_optimize_gpu_model(Name, Model, Tokenizer, model_path, RMSNorm_layer, layer_before_RMSNorm, lower_bound) self.run_optimize_gpu_model(Name, Model, Tokenizer, model_path, RMSNorm_layer, layer_before_RMSNorm, lower_bound)