From 4de73f592e6b449eef3d1241cbae7502b8696d83 Mon Sep 17 00:00:00 2001 From: Ruonan Wang <105281011+rnwang04@users.noreply.github.com> Date: Wed, 13 Sep 2023 10:16:51 +0800 Subject: [PATCH] LLM: add gpu example of chinese-llama-2-7b (#8960) * add gpu example of chinese -llama2 * update model name and link * update name --- .../llm/example/gpu/chinese-llama2/README.md | 57 ++++++++++++ .../example/gpu/chinese-llama2/generate.py | 92 +++++++++++++++++++ 2 files changed, 149 insertions(+) create mode 100644 python/llm/example/gpu/chinese-llama2/README.md create mode 100644 python/llm/example/gpu/chinese-llama2/generate.py diff --git a/python/llm/example/gpu/chinese-llama2/README.md b/python/llm/example/gpu/chinese-llama2/README.md new file mode 100644 index 00000000..d795ad98 --- /dev/null +++ b/python/llm/example/gpu/chinese-llama2/README.md @@ -0,0 +1,57 @@ +# Chinese Llama2 +In this directory, you will find examples on how you could apply BigDL-LLM INT4 optimizations on Chinese LLaMA models on [Intel GPUs](../README.md). For illustration purposes, we utilize the [LinkSoul/Chinese-Llama-2-7b](https://huggingface.co/LinkSoul/Chinese-Llama-2-7b) as reference Chinese LLaMA models. + +## 0. Requirements +To run these examples with BigDL-LLM on Intel GPUs, we have some recommended requirements for your machine, please refer to [here](../README.md#recommended-requirements) for more information. + +## Example: Predict Tokens using `generate()` API +In the example [generate.py](./generate.py), we show a basic use case for a Llama2 model to predict the next N tokens using `generate()` API, with BigDL-LLM INT4 optimizations on Intel GPUs. +### 1. Install +We suggest using conda to manage environment: +```bash +conda create -n llm python=3.9 +conda activate llm +# below command will install intel_extension_for_pytorch==2.0.110+xpu as default +# you can install specific ipex/torch version for your need +pip install --pre --upgrade bigdl-llm[xpu] -f https://developer.intel.com/ipex-whl-stable-xpu +``` +### 2. Configures OneAPI environment variables +```bash +source /opt/intel/oneapi/setvars.sh +``` + +### 3. Run + +For optimal performance on Arc, it is recommended to set several environment variables. + +```bash +export USE_XETLA=OFF +export SYCL_PI_LEVEL_ZERO_USE_IMMEDIATE_COMMANDLISTS=1 +``` + +``` +python ./generate.py --repo-id-or-model-path REPO_ID_OR_MODEL_PATH --prompt PROMPT --n-predict N_PREDICT +``` + +Arguments info: +- `--repo-id-or-model-path REPO_ID_OR_MODEL_PATH`: argument defining the huggingface repo id for the Chinese Llama2 model (e.g. `LinkSoul/Chinese-Llama-2-7b`) to be downloaded, or the path to the huggingface checkpoint folder. It is default to be `'LinkSoul/Chinese-Llama-2-7b'`. +- `--prompt PROMPT`: argument defining the prompt to be infered (with integrated prompt format for chat). It is default to be `'What is AI?'`. +- `--n-predict N_PREDICT`: argument defining the max number of tokens to predict. It is default to be `32`. + +#### Sample Output +#### [LinkSoul/Chinese-Llama-2-7b](https://huggingface.co/LinkSoul/Chinese-Llama-2-7b) +```log +Inference time: xxxx s +-------------------- Prompt -------------------- +[INST] <> + +<> + +AI是什么？ [/INST] +-------------------- Output -------------------- +[INST] <> + +<> + +AI是什么？ [/INST] AI（人工智能）是一种计算机科学，旨在开发能够模拟人 +``` diff --git a/python/llm/example/gpu/chinese-llama2/generate.py b/python/llm/example/gpu/chinese-llama2/generate.py new file mode 100644 index 00000000..ca19ff39 --- /dev/null +++ b/python/llm/example/gpu/chinese-llama2/generate.py @@ -0,0 +1,92 @@ +# +# 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 intel_extension_for_pytorch as ipex +import time +import argparse + +from bigdl.llm.transformers import AutoModelForCausalLM +from transformers import LlamaTokenizer + +# you could tune the prompt based on your own model, +# here the prompt tuning refers to https://huggingface.co/georgesung/llama2_7b_chat_uncensored#prompt-style +DEFAULT_SYSTEM_PROMPT = """\ +""" + +def get_prompt(message: str, chat_history: list[tuple[str, str]], + system_prompt: str) -> str: + texts = [f'[INST] <>\n{system_prompt}\n<>\n\n'] + # The first user input is _not_ stripped + do_strip = False + for user_input, response in chat_history: + user_input = user_input.strip() if do_strip else user_input + do_strip = True + texts.append(f'{user_input} [/INST] {response.strip()} [INST] ') + message = message.strip() if do_strip else message + texts.append(f'{message} [/INST]') + return ''.join(texts) + +if __name__ == '__main__': + parser = argparse.ArgumentParser(description='Predict Tokens using `generate()` API for Llama2 model') + parser.add_argument('--repo-id-or-model-path', type=str, default="LinkSoul/Chinese-Llama-2-7b", + help='The huggingface repo id for the Chinese Llama2 (e.g. `LinkSoul/Chinese-Llama-2-7b`) 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 + # 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 to obtain optimal + # performance with BigDL-LLM INT4 optimizations + 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 = LlamaTokenizer.from_pretrained(model_path, trust_remote_code=True) + + # Generate predicted tokens + with torch.inference_mode(): + prompt = get_prompt(args.prompt, [], system_prompt=DEFAULT_SYSTEM_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() + output = model.generate(input_ids, + max_new_tokens=args.n_predict) + torch.xpu.synchronize() + end = time.time() + output = output.cpu() + 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)