From f754ab3e60aabb9cd63f7247c82202fb057b2bb6 Mon Sep 17 00:00:00 2001 From: Jin Qiao <89779290+JinBridger@users.noreply.github.com> Date: Fri, 13 Oct 2023 13:44:31 +0800 Subject: [PATCH] LLM: add baichuan and baichuan2 to gpu pytorch model example (#9152) --- .../GPU/PyTorch-Models/Model/README.md | 2 + .../PyTorch-Models/Model/baichuan/README.md | 53 ++++++++++++++ .../PyTorch-Models/Model/baichuan/generate.py | 73 +++++++++++++++++++ .../PyTorch-Models/Model/baichuan2/README.md | 60 +++++++++++++++ .../Model/baichuan2/generate.py | 73 +++++++++++++++++++ 5 files changed, 261 insertions(+) create mode 100644 python/llm/example/GPU/PyTorch-Models/Model/baichuan/README.md create mode 100644 python/llm/example/GPU/PyTorch-Models/Model/baichuan/generate.py create mode 100644 python/llm/example/GPU/PyTorch-Models/Model/baichuan2/README.md create mode 100644 python/llm/example/GPU/PyTorch-Models/Model/baichuan2/generate.py diff --git a/python/llm/example/GPU/PyTorch-Models/Model/README.md b/python/llm/example/GPU/PyTorch-Models/Model/README.md index 6edf4077..299cc1b2 100644 --- a/python/llm/example/GPU/PyTorch-Models/Model/README.md +++ b/python/llm/example/GPU/PyTorch-Models/Model/README.md @@ -7,6 +7,8 @@ You can use `optimize_model` API to accelerate general PyTorch models on Intel G | Mistral | [link](mistral) | | LLaMA 2 | [link](llama2) | | ChatGLM2 | [link](chatglm2) | +| Baichuan | [link](baichuan) | +| Baichuan2 | [link](baichuan2) | ## Verified Hardware Platforms diff --git a/python/llm/example/GPU/PyTorch-Models/Model/baichuan/README.md b/python/llm/example/GPU/PyTorch-Models/Model/baichuan/README.md new file mode 100644 index 00000000..a6528cfc --- /dev/null +++ b/python/llm/example/GPU/PyTorch-Models/Model/baichuan/README.md @@ -0,0 +1,53 @@ +# Baichuan +In this directory, you will find examples on how you could use BigDL-LLM `optimize_model` API to accelerate Baichuan models. For illustration purposes, we utilize the [baichuan-inc/Baichuan-13B-Chat](https://huggingface.co/baichuan-inc/Baichuan-13B-Chat) as reference Baichuan models. + +## 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 Baichuan 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 the Python environment. For more information about conda installation, please refer to [here](https://docs.conda.io/en/latest/miniconda.html#). + +After installing conda, create a Python environment for BigDL-LLM: +```bash +conda create -n llm python=3.9 # recommend to use 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 +pip install transformers_stream_generator # additional package required for Baichuan-13B-Chat to conduct generation +``` + +### 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 +``` + +```bash +python ./generate.py --prompt 'AI是什么？' +``` + +In the example, several arguments can be passed to satisfy your requirements: + +- `--repo-id-or-model-path REPO_ID_OR_MODEL_PATH`: argument defining the huggingface repo id for the Baichuan model (e.g `baichuan-inc/Baichuan-13B-Chat`) to be downloaded, or the path to the huggingface checkpoint folder. It is default to be `'baichuan-inc/Baichuan-13B-Chat'`. +- `--prompt PROMPT`: argument defining the prompt to be infered (with integrated prompt format for chat). It is default to be `'AI是什么？'`. +- `--n-predict N_PREDICT`: argument defining the max number of tokens to predict. It is default to be `32`. + +#### 2.3 Sample Output +#### [baichuan-inc/Baichuan-13B-Chat](https://huggingface.co/baichuan-inc/Baichuan-13B-Chat) +```log +Inference time: xxxx s +-------------------- Output -------------------- +AI是什么？ 人工智能(Artificial Intelligence，简称AI)是指由人制造出来的系统所表现出的智能，是计算机科学的一个分支。人工智能的研究包括机器 +``` \ No newline at end of file diff --git a/python/llm/example/GPU/PyTorch-Models/Model/baichuan/generate.py b/python/llm/example/GPU/PyTorch-Models/Model/baichuan/generate.py new file mode 100644 index 00000000..d8f4642b --- /dev/null +++ b/python/llm/example/GPU/PyTorch-Models/Model/baichuan/generate.py @@ -0,0 +1,73 @@ +# +# 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 transformers import AutoModelForCausalLM, AutoTokenizer +from bigdl.llm import optimize_model + +# you could tune the prompt based on your own model +BAICHUAN_PROMPT_FORMAT = "{prompt} " + +if __name__ == '__main__': + parser = argparse.ArgumentParser(description='Predict Tokens using `generate()` API for Baichuan model') + parser.add_argument('--repo-id-or-model-path', type=str, default="baichuan-inc/Baichuan-13B-Chat", + help='The huggingface repo id for the Baichuan 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 + model = AutoModelForCausalLM.from_pretrained(model_path, + trust_remote_code=True, + torch_dtype='auto', + low_cpu_mem_usage=True) + + # With only one line to enable BigDL-LLM optimization on model + model = optimize_model(model) + + model = model.to('xpu') + + # Load tokenizer + tokenizer = AutoTokenizer.from_pretrained(model_path, trust_remote_code=True) + + # Generate predicted tokens + with torch.inference_mode(): + prompt = BAICHUAN_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() + 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, 'Output', '-'*20) + print(output_str) diff --git a/python/llm/example/GPU/PyTorch-Models/Model/baichuan2/README.md b/python/llm/example/GPU/PyTorch-Models/Model/baichuan2/README.md new file mode 100644 index 00000000..64e3bf85 --- /dev/null +++ b/python/llm/example/GPU/PyTorch-Models/Model/baichuan2/README.md @@ -0,0 +1,60 @@ +# Baichuan2 +In this directory, you will find examples on how you could use BigDL-LLM `optimize_model` API to accelerate Baichuan2 models. For illustration purposes, we utilize the [baichuan-inc/Baichuan2-7B-Chat](https://huggingface.co/baichuan-inc/Baichuan-7B-Chat) as reference Baichuan2 models. + +## 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 Baichuan2 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 the Python environment. For more information about conda installation, please refer to [here](https://docs.conda.io/en/latest/miniconda.html#). + +After installing conda, create a Python environment for BigDL-LLM: +```bash +conda create -n llm python=3.9 # recommend to use 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 +pip install transformers_stream_generator # additional package required for Baichuan2-7B-Chat to conduct generation +``` + +### 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 +``` + +```bash +python ./generate.py --prompt 'AI是什么？' +``` + +In the example, several arguments can be passed to satisfy your requirements: + +- `--repo-id-or-model-path REPO_ID_OR_MODEL_PATH`: argument defining the huggingface repo id for the Baichuan2 model (e.g `baichuan-inc/Baichuan2-7B-Chat`) to be downloaded, or the path to the huggingface checkpoint folder. It is default to be `'baichuan-inc/Baichuan2-7B-Chat'`. +- `--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`. + +#### 2.3 Sample Output +#### [baichuan-inc/Baichuan2-7B-Chat](https://huggingface.co/baichuan-inc/Baichuan2-7B-Chat) +```log +Inference time: xxxx s +-------------------- Output -------------------- +AI是什么？ +AI是人工智能（Artificial Intelligence）的缩写，它是指让计算机或机器模拟、扩展和辅助人类的智能。AI技术已经广泛应用于各个领域 +``` + +```log +Inference time: xxxx s +-------------------- Output -------------------- +What is AI? Artificial Intelligence (AI) refers to the development of computer systems that can perform tasks that would typically require human intelligence. These tasks include learning, reasoning, problem +``` \ No newline at end of file diff --git a/python/llm/example/GPU/PyTorch-Models/Model/baichuan2/generate.py b/python/llm/example/GPU/PyTorch-Models/Model/baichuan2/generate.py new file mode 100644 index 00000000..ebd052ec --- /dev/null +++ b/python/llm/example/GPU/PyTorch-Models/Model/baichuan2/generate.py @@ -0,0 +1,73 @@ +# +# 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 transformers import AutoModelForCausalLM, AutoTokenizer +from bigdl.llm import optimize_model + +# you could tune the prompt based on your own model, +BAICHUAN2_PROMPT_FORMAT = "{prompt} " + +if __name__ == '__main__': + parser = argparse.ArgumentParser(description='Predict Tokens using `generate()` API for Baichuan2 model') + parser.add_argument('--repo-id-or-model-path', type=str, default="baichuan-inc/Baichuan2-7B-Chat", + help='The huggingface repo id for the Baichuan2 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 + model = AutoModelForCausalLM.from_pretrained(model_path, + trust_remote_code=True, + torch_dtype='auto', + low_cpu_mem_usage=True) + + # With only one line to enable BigDL-LLM optimization on model + model = optimize_model(model) + + model = model.to('xpu') + + # Load tokenizer + tokenizer = AutoTokenizer.from_pretrained(model_path, trust_remote_code=True) + + # Generate predicted tokens + with torch.inference_mode(): + prompt = BAICHUAN2_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() + 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, 'Output', '-'*20) + print(output_str)