228 lines
10 KiB
Python
228 lines
10 KiB
Python
#
|
|
# 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, time
|
|
import pytest
|
|
|
|
import torch
|
|
from bigdl.llm.transformers import AutoModelForCausalLM, AutoModel, AutoModelForSpeechSeq2Seq
|
|
from transformers import LlamaTokenizer, AutoTokenizer
|
|
|
|
device = os.environ['DEVICE']
|
|
print(f'Running on {device}')
|
|
if device == 'xpu':
|
|
import intel_extension_for_pytorch as ipex
|
|
|
|
@pytest.mark.parametrize('prompt, answer', [
|
|
('What is the capital of France?\n\n', 'Paris')
|
|
])
|
|
@pytest.mark.parametrize('Model, Tokenizer, model_path',[
|
|
(AutoModelForCausalLM, LlamaTokenizer, os.environ.get('LLAMA2_7B_ORIGIN_PATH')),
|
|
(AutoModel, AutoTokenizer, os.environ.get('CHATGLM2_6B_ORIGIN_PATH')),
|
|
(AutoModelForCausalLM, AutoTokenizer, os.environ.get('FALCON_7B_ORIGIN_PATH')),
|
|
(AutoModelForCausalLM, AutoTokenizer, os.environ.get('MPT_7B_ORIGIN_PATH')),
|
|
])
|
|
def test_completion(Model, Tokenizer, model_path, prompt, answer):
|
|
with torch.inference_mode():
|
|
tokenizer = Tokenizer.from_pretrained(model_path, trust_remote_code=True)
|
|
model = Model.from_pretrained(model_path,
|
|
load_in_4bit=True,
|
|
optimize_model=True,
|
|
trust_remote_code=True)
|
|
model = model.to(device)
|
|
|
|
input_ids = tokenizer.encode(prompt, return_tensors="pt").to(device)
|
|
output = model.generate(input_ids, max_new_tokens=32)
|
|
model.to('cpu') # deallocate gpu memory
|
|
output_str = tokenizer.decode(output[0], skip_special_tokens=True)
|
|
|
|
assert answer in output_str
|
|
|
|
def test_transformers_auto_model_for_speech_seq2seq_int4():
|
|
with torch.inference_mode():
|
|
from transformers import WhisperProcessor
|
|
from datasets import load_from_disk
|
|
model_path = os.environ.get('WHISPER_TINY_ORIGIN_PATH')
|
|
dataset_path = os.environ.get('SPEECH_DATASET_PATH')
|
|
processor = WhisperProcessor.from_pretrained(model_path)
|
|
ds = load_from_disk(dataset_path)
|
|
sample = ds[0]["audio"]
|
|
input_features = processor(sample["array"], sampling_rate=sample["sampling_rate"], return_tensors="pt").input_features
|
|
input_features = input_features.to(device)
|
|
model = AutoModelForSpeechSeq2Seq.from_pretrained(model_path, trust_remote_code=True, load_in_4bit=True, optimize_model=True)
|
|
model = model.to(device)
|
|
predicted_ids = model.generate(input_features)
|
|
# decode token ids to text
|
|
transcription = processor.batch_decode(predicted_ids, skip_special_tokens=False)
|
|
model.to('cpu')
|
|
print('Output:', transcription)
|
|
assert 'Mr. Quilter is the apostle of the middle classes and we are glad to welcome his gospel.' in transcription[0]
|
|
|
|
|
|
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"
|
|
|
|
@pytest.mark.parametrize('Model, Tokenizer, model_path',[
|
|
(AutoModelForCausalLM, AutoTokenizer, os.environ.get('MPT_7B_ORIGIN_PATH')),
|
|
(AutoModelForCausalLM, AutoTokenizer, os.environ.get('LLAMA2_7B_ORIGIN_PATH'))
|
|
])
|
|
def test_optimize_model(Model, Tokenizer, model_path):
|
|
with torch.inference_mode():
|
|
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)
|
|
logits_base_model = (model(input_ids)).logits
|
|
model.to('cpu') # deallocate gpu memory
|
|
|
|
model = Model.from_pretrained(model_path,
|
|
load_in_4bit=True,
|
|
optimize_model=True,
|
|
trust_remote_code=True)
|
|
model = model.to(device)
|
|
logits_optimized_model = (model(input_ids)).logits
|
|
model.to('cpu')
|
|
|
|
assert all(torch.isclose(logits_optimized_model, logits_base_model).tolist())
|
|
|
|
class Test_Optimize_Gpu_Model:
|
|
def setup(self):
|
|
|
|
self.layer_outputs = []
|
|
self.pre_layer_outputs = []
|
|
|
|
def run_optimize_gpu_model(self, Model, Tokenizer, model_path, self_attn, layer_norm, lower_bound):
|
|
with torch.inference_mode():
|
|
def forward_hook(module, input, output, layer_name):
|
|
self.layer_outputs.append(output)
|
|
|
|
def pre_forward_hook(module, input, output, layer_name):
|
|
self.pre_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_norm:
|
|
layer_module.register_forward_hook(
|
|
lambda module, input, output, layer_name=layer_name: pre_forward_hook(module, input,
|
|
output, layer_name))
|
|
if layer_name == self_attn:
|
|
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_norm:
|
|
layer_module.register_forward_hook(
|
|
lambda module, input, output, layer_name=layer_name: replace_forward_hook(module, input,
|
|
output, layer_name))
|
|
if layer_name == self_attn:
|
|
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')
|
|
|
|
attn_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):
|
|
# 'attn_output' is of type torch.Tensor.
|
|
attn_output_diff.append(t1 - t2)
|
|
else:
|
|
# 'past_key_value'is of type tuple as default.
|
|
for i, (t3, t4) in enumerate(zip(t1, t2)):
|
|
if model.config.architectures[0] == "ChatGLMModel" and \
|
|
hasattr(model.config, 'padded_vocab_size') and \
|
|
model.config.padded_vocab_size == 65024:
|
|
# chatglm2's past_key_value is expanded 16x for some speedup.
|
|
# We need to narrow it here.
|
|
t4 = t4[:, :, 15:17, :]
|
|
attn_output_diff.append(t3 - t4)
|
|
|
|
max_diff_tensor = [torch.max(item).item() for item in attn_output_diff]
|
|
print(max_diff_tensor)
|
|
assert all(max_diff <= lower_bound for max_diff in max_diff_tensor)
|
|
|
|
|
|
def test_falcon_gpu_model(self):
|
|
|
|
Model = AutoModelForCausalLM
|
|
Tokenizer = AutoTokenizer
|
|
model_path = os.environ.get('FALCON_7B_ORIGIN_PATH')
|
|
# currently only compare the output of the last self-attention layer.
|
|
layer_norm = "transformer.h.31.input_layernorm"
|
|
self_attn = "transformer.h.31.self_attention"
|
|
lower_bound = 0
|
|
|
|
self.run_optimize_gpu_model(Model, Tokenizer, model_path, self_attn, layer_norm, lower_bound)
|
|
|
|
|
|
def test_llama_gpu_model(self):
|
|
|
|
Model = AutoModelForCausalLM
|
|
Tokenizer = AutoTokenizer
|
|
model_path = os.environ.get('LLAMA2_7B_ORIGIN_PATH')
|
|
# currently only compare the output of the last self-attention layer.
|
|
layer_norm = "model.layers.31.input_layernorm"
|
|
self_attn = "model.layers.31.self_attn"
|
|
lower_bound = 5e-2
|
|
|
|
self.run_optimize_gpu_model(Model, Tokenizer, model_path, self_attn, layer_norm, lower_bound)
|
|
|
|
def test_chatglm2_gpu_model(self):
|
|
|
|
Model = AutoModel
|
|
Tokenizer = AutoTokenizer
|
|
model_path = os.environ.get('CHATGLM2_6B_ORIGIN_PATH')
|
|
# currently only need to compare the output of one self-attention layer.
|
|
layer_norm = "transformer.encoder.layers.27.input_layernorm"
|
|
self_attn = "transformer.encoder.layers.27.self_attention"
|
|
lower_bound = 1e-3
|
|
|
|
self.run_optimize_gpu_model(Model, Tokenizer, model_path, self_attn, layer_norm, lower_bound)
|
|
|
|
|
|
if __name__ == '__main__':
|
|
pytest.main([__file__])
|