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

LLM: Add decoder/layernorm unit tests (#10211)

Browse source 
* add decoder/layernorm unit tests

* update tests

* delete decoder tests

* address comments

* remove none type check

* restore nonetype checks

* delete nonetype checks; add decoder tests for Llama

* add gc

* deal with tuple output
This commit is contained in:
Ovo233 2024-03-13 19:41:47 +08:00 committed by GitHub
parent 9880ddfc17
commit 0dbce53464
3 changed files with 137 additions and 11 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

117
python/llm/test/inference_gpu/test_transformers_api_layernorm.py Normal file
View file

@ -0,0 +1,117 @@
#
# Copyright 2016 The BigDL Authors.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
#
import os
import pytest
import gc
import torch
from bigdl.llm.transformers import AutoModelForCausalLM, AutoModel
from transformers import LlamaTokenizer, AutoTokenizer
device = os.environ['DEVICE']
print(f'Running on {device}')
PROMPT = "Once upon a time, there existed a little girl who liked to have adventures. She wanted to go to places and meet new people, and have fun"
TEST_MODEL_LIST = [
("Falcon-7B", AutoModelForCausalLM, AutoTokenizer, os.environ.get('FALCON_7B_ORIGIN_PATH'))
]
class Test_Optimize_Gpu_Model:
def setup_method(self):
self.layer_outputs = []
self.pre_layer_outputs = []
def run_optimize_gpu_model(self, Name, Model, Tokenizer, model_path, LayerNorm_layer, layer_before_LayerNorm, lower_bound):
with torch.inference_mode():
def pre_forward_hook(module, input, output, layer_name):
self.pre_layer_outputs.append(output)
def forward_hook(module, input, output, layer_name):
self.layer_outputs.append(output)
tokenizer = Tokenizer.from_pretrained(model_path, trust_remote_code=True)
input_ids = tokenizer.encode(PROMPT, return_tensors="pt").to(device)
model = Model.from_pretrained(model_path,
load_in_4bit=True,
optimize_model=False,
trust_remote_code=True)
model = model.to(device)
for layer_name, layer_module in model.named_modules():
if layer_name == layer_before_LayerNorm:
layer_module.register_forward_hook(
lambda module, input, output, layer_name=layer_name: pre_forward_hook(module, input,
output, layer_name))
if layer_name == LayerNorm_layer:
layer_module.register_forward_hook(
lambda module, input, output, layer_name=layer_name: forward_hook(module, input,
output, layer_name))
logits_base_model = (model(input_ids)).logits
# the list `layer_output` has only one element.
layer_tensor = self.layer_outputs.pop()
model.to('cpu')
opt_model = Model.from_pretrained(model_path,
load_in_4bit=True,
optimize_model=True,
trust_remote_code=True)
opt_model = opt_model.to(device)
def replace_forward_hook(module, input, output, layer_name):
output = self.pre_layer_outputs[0]
return output
for layer_name, layer_module in opt_model.named_modules():
if layer_name == layer_before_LayerNorm:
layer_module.register_forward_hook(
lambda module, input, output, layer_name=layer_name: replace_forward_hook(module, input,
output, layer_name))
if layer_name == LayerNorm_layer:
layer_module.register_forward_hook(
lambda module, input, output, layer_name=layer_name: forward_hook(module, input,
output, layer_name))
logits_optimized_model = (opt_model(input_ids)).logits
# the list `layer_output` has only one element.
opt_layer_tensor = self.layer_outputs[0]
opt_model.to('cpu')
LayerNorm_output_diff = []
for i, (t1, t2) in enumerate(zip(layer_tensor, opt_layer_tensor)):
LayerNorm_output_diff.append(t1 - t2)
max_diff_tensor = [torch.max(item).item() for item in LayerNorm_output_diff]
print(max_diff_tensor)
torch.xpu.empty_cache()
del model
del opt_model
gc.collect()
assert all(max_diff <= lower_bound for max_diff in max_diff_tensor)
@pytest.mark.parametrize('Name, Model, Tokenizer, model_path',TEST_MODEL_LIST)
def test_dynamic_functions(self, Name, Model, Tokenizer, model_path):
if Name == "Falcon-7B":
self.Falcon_7B_gpu_model(Name, Model, Tokenizer, model_path)
def Falcon_7B_gpu_model(self, Name, Model, Tokenizer, model_path):
# currently only compare the output of the last LayerNorm layer.
layer_before_LayerNorm = "transformer.h.30"
LayerNorm_layer = "transformer.h.31.input_layernorm"
lower_bound = 0
self.run_optimize_gpu_model(Name, Model, Tokenizer, model_path, LayerNorm_layer, layer_before_LayerNorm, lower_bound)

30
python/llm/test/inference_gpu/test_transformers_api_mlp.py
View file

@ -28,7 +28,8 @@ print(f'Running on {device}')
PROMPT = "Once upon a time, there existed a little girl who liked to have adventures. She wanted to go to places and meet new people, and have fun"
TEST_MODEL_LIST = [
("Qwen-7B-Chat", AutoModelForCausalLM, AutoTokenizer, os.environ.get('QWEN_7B_ORIGIN_PATH')),
("Mistral-7B-Instruct-v0.1", AutoModelForCausalLM, AutoTokenizer, os.environ.get('MISTRAL_7B_INSTRUCT_V0_1_ORIGIN_PATH'))
("Mistral-7B-Instruct-v0.1", AutoModelForCausalLM, AutoTokenizer, os.environ.get('MISTRAL_7B_INSTRUCT_V0_1_ORIGIN_PATH')),
("Llama2-7B", AutoModelForCausalLM, LlamaTokenizer, os.environ.get('LLAMA2_7B_ORIGIN_PATH'))
]
class Test_Optimize_Gpu_Model:
@ -91,16 +92,13 @@ class Test_Optimize_Gpu_Model:
opt_layer_tensor = self.layer_outputs[0]
opt_model.to('cpu')
MLP_output_diff = []
for i, (t1, t2) in enumerate(zip(layer_tensor, opt_layer_tensor)):
if t1 is not None and t2 is not None:
if isinstance(t1, torch.Tensor) and isinstance(t2, torch.Tensor):
MLP_output_diff.append(t1 - t2)
else:
# 'past_key_value'is of type tuple as default.
for i, (t3, t4) in enumerate(zip(t1, t2)):
MLP_output_diff.append(t3 - t4)
if isinstance(t1, torch.Tensor) and isinstance(t2, torch.Tensor):
MLP_output_diff.append(t1 - t2)
else:
for i, (t3, t4) in enumerate(zip(t1, t2)):
MLP_output_diff.append(t3 - t4)
max_diff_tensor = [torch.max(item).item() for item in MLP_output_diff]
print(max_diff_tensor)
@ -116,8 +114,10 @@ class Test_Optimize_Gpu_Model:
self.Qwen_7B_gpu_model(Name, Model, Tokenizer, model_path)
elif Name == "Mistral-7B-Instruct-v0.1":
self.Mistral_7B_Instruct_gpu_model(Name, Model, Tokenizer, model_path)
elif Name == "Llama2-7B":
self.Llama2_7B_gpu_model(Name, Model, Tokenizer, model_path)
def Qwen_7B_gpu_model(self, Name, Model, Tokenizer, model_path):
# currently only compare the output of the last mlp layer.
layer_before_MLP = "transformer.h.31.ln_2"
@ -130,4 +130,12 @@ class Test_Optimize_Gpu_Model:
layer_before_MLP = "model.layers.31.post_attention_layernorm"
MLP_layer = "model.layers.31.mlp"
lower_bound = 0
self.run_optimize_gpu_model(Name, Model, Tokenizer, model_path, MLP_layer, layer_before_MLP, lower_bound)
self.run_optimize_gpu_model(Name, Model, Tokenizer, model_path, MLP_layer, layer_before_MLP, lower_bound)
def Llama2_7B_gpu_model(self, Name, Model, Tokenizer, model_path):
# The tests are actually testing the mlp layer. We can't test the mlp layer directly
# since the original Llama2 code adds residual after the mlp layer, which differs from the implementation of bigdl
layer_before_Decoder = "model.layers.30"
Decoder_layer = "model.layers.31"
lower_bound = 5e-2
self.run_optimize_gpu_model(Name, Model, Tokenizer, model_path, Decoder_layer, layer_before_Decoder, lower_bound)

1
python/llm/test/run-llm-inference-tests-gpu.sh
View file

@ -18,6 +18,7 @@ start=$(date "+%s")
# fi
# export OMP_NUM_THREADS=$THREAD_NUM
pytest ${LLM_INFERENCE_TEST_DIR}/test_transformers_api.py -v -s
pytest ${LLM_INFERENCE_TEST_DIR}/test_transformers_api_layernorm.py -v -s
export BIGDL_LLM_XMX_DISABLED=1
pytest ${LLM_INFERENCE_TEST_DIR}/test_transformers_api_final_logits.py -v -s
pytest ${LLM_INFERENCE_TEST_DIR}/test_transformers_api_attention.py -v -s -k "not Mistral"