245c7348bc
* add codegemma example in GPU/HF-Transformers-AutoModels/ * add README of codegemma example in GPU/HF-Transformers-AutoModels/ * add codegemma example in GPU/PyTorch-Models/ * add readme of codegemma example in GPU/PyTorch-Models/ * add codegemma example in CPU/HF-Transformers-AutoModels/ * add readme of codegemma example in CPU/HF-Transformers-AutoModels/ * add codegemma example in CPU/PyTorch-Models/ * add readme of codegemma example in CPU/PyTorch-Models/ * fix typos * fix filename typo * add codegemma in tables * add comments of lm_head * remove comments of use_cache
71 lines
3 KiB
Python
71 lines
3 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 ipex_llm.transformers import AutoModelForCausalLM
|
|
from transformers import AutoTokenizer
|
|
|
|
# The instruction-tuned models use a chat template that must be adhered to for conversational use.
|
|
# see https://huggingface.co/google/codegemma-7b-it#chat-template.
|
|
chat = [
|
|
{ "role": "user", "content": "Write a hello world program" },
|
|
]
|
|
|
|
if __name__ == '__main__':
|
|
parser = argparse.ArgumentParser(description='Predict Tokens using `generate()` API for CodeGemma model')
|
|
parser.add_argument('--repo-id-or-model-path', type=str, default="google/codegemma-7b-it",
|
|
help='The huggingface repo id for the CodeGemma to be downloaded'
|
|
', or the path to the huggingface checkpoint folder')
|
|
parser.add_argument('--prompt', type=str, default="Write a hello world program",
|
|
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
|
|
# To fix the issue that the output of codegemma-7b-it is abnormal, skip the 'lm_head' module during optimization
|
|
model = AutoModelForCausalLM.from_pretrained(model_path,
|
|
load_in_4bit=True,
|
|
trust_remote_code=True,
|
|
use_cache=True,
|
|
modules_to_not_convert=["lm_head"])
|
|
|
|
# Load tokenizer
|
|
tokenizer = AutoTokenizer.from_pretrained(model_path, trust_remote_code=True)
|
|
|
|
# Generate predicted tokens
|
|
with torch.inference_mode():
|
|
chat[0]['content'] = args.prompt
|
|
prompt = tokenizer.apply_chat_template(chat, tokenize=False, add_generation_prompt=True)
|
|
input_ids = tokenizer.encode(prompt, return_tensors="pt")
|
|
|
|
# start inference
|
|
st = time.time()
|
|
output = model.generate(input_ids,
|
|
max_new_tokens=args.n_predict)
|
|
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)
|