440cfe18ed
* modify aquila * modify aquila2 * add baichuan * modify baichuan2 * modify blue-lm * modify chatglm3 * modify chinese-llama2 * modiy codellama * modify distil-whisper * modify dolly-v1 * modify dolly-v2 * modify falcon * modify flan-t5 * modify gpt-j * modify internlm * modify llama2 * modify mistral * modify mixtral * modify mpt * modify phi-1_5 * modify qwen * modify qwen-vl * modify replit * modify solar * modify starcoder * modify vicuna * modify voiceassistant * modify whisper * modify yi * modify aquila2 * modify baichuan * modify baichuan2 * modify blue-lm * modify chatglm2 * modify chatglm3 * modify codellama * modify distil-whisper * modify dolly-v1 * modify dolly-v2 * modify flan-t5 * modify llama2 * modify llava * modify mistral * modify mixtral * modify phi-1_5 * modify qwen-vl * modify replit * modify solar * modify starcoder * modify yi * correct the comments * remove cpu_embedding in code for whisper and distil-whisper * remove comment * remove cpu_embedding for voice assistant * revert modify voice assistant * modify for voice assistant * add comment for voice assistant * fix comments * fix comments
84 lines
3.7 KiB
Python
84 lines
3.7 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 torch
|
||
import time
|
||
import argparse
|
||
|
||
from bigdl.llm.transformers import AutoModelForCausalLM
|
||
from transformers import AutoTokenizer
|
||
|
||
# Refer to https://huggingface.co/01-ai/Yi-6B-Chat#31-use-the-chat-model
|
||
YI_PROMPT_FORMAT = """
|
||
<|im_start|>system
|
||
You are a helpful assistant. If you don't understand what the user means, ask the user to provide more information.<|im_end|>
|
||
<|im_start|>user
|
||
{prompt}<|im_end|>
|
||
<|im_start|>assistant
|
||
"""
|
||
|
||
if __name__ == '__main__':
|
||
parser = argparse.ArgumentParser(description='Predict Tokens using `generate()` API for Yi model')
|
||
parser.add_argument('--repo-id-or-model-path', type=str, default="01-ai/Yi-6B",
|
||
help='The huggingface repo id for the Yi model to be downloaded'
|
||
', or the path to the huggingface checkpoint folder')
|
||
parser.add_argument('--prompt', type=str, default="AI是什么？",
|
||
help='Prompt to infer')
|
||
parser.add_argument('--n-predict', type=int, default=32,
|
||
help='Max tokens to predict')
|
||
|
||
args = parser.parse_args()
|
||
model_path = args.repo_id_or_model_path
|
||
|
||
# Load model in 4 bit,
|
||
# which convert the relevant layers in the model into INT4 format
|
||
# When running LLMs on Intel iGPUs for Windows users, we recommend setting `cpu_embedding=True` in the from_pretrained function.
|
||
# This will allow the memory-intensive embedding layer to utilize the CPU instead of iGPU.
|
||
model = AutoModelForCausalLM.from_pretrained(model_path,
|
||
load_in_4bit=True,
|
||
optimize_model=True,
|
||
trust_remote_code=True,
|
||
use_cache=True)
|
||
model = model.to('xpu')
|
||
|
||
# Load tokenizer
|
||
tokenizer = AutoTokenizer.from_pretrained(model_path,
|
||
trust_remote_code=True)
|
||
|
||
# Generate predicted tokens
|
||
with torch.inference_mode():
|
||
prompt = YI_PROMPT_FORMAT.format(prompt=args.prompt)
|
||
input_ids = tokenizer.encode(prompt, return_tensors="pt").to('xpu')
|
||
# ipex model needs a warmup, then inference time can be accurate
|
||
output = model.generate(input_ids,
|
||
max_new_tokens=args.n_predict)
|
||
|
||
# start inference
|
||
st = time.time()
|
||
# if your selected model is capable of utilizing previous key/value attentions
|
||
# to enhance decoding speed, but has `"use_cache": false` in its model config,
|
||
# it is important to set `use_cache=True` explicitly in the `generate` function
|
||
# to obtain optimal performance with BigDL-LLM INT4 optimizations
|
||
output = model.generate(input_ids,
|
||
max_new_tokens=args.n_predict)
|
||
torch.xpu.synchronize()
|
||
end = time.time()
|
||
output_str = tokenizer.decode(output[0], skip_special_tokens=True)
|
||
print(f'Inference time: {end-st} s')
|
||
print('-'*20, 'Prompt', '-'*20)
|
||
print(prompt)
|
||
print('-'*20, 'Output', '-'*20)
|
||
print(output_str)
|