From 2d815210198ce22f618e3f4b1df48aa1b7bf10d2 Mon Sep 17 00:00:00 2001 From: binbin Deng <108676127+plusbang@users.noreply.github.com> Date: Tue, 12 Sep 2023 10:36:29 +0800 Subject: [PATCH] LLM: add `optimize_model` examples for llama2 and chatglm (#8894) * add llama2 and chatglm optimize_model examples * update default usage * update command and some descriptions * move folder and remove general_int4 descriptions * change folder name --- python/llm/example/pytorch-models/README.md | 21 ++++++ .../example/pytorch-models/chatglm/README.md | 58 +++++++++++++++ .../pytorch-models/chatglm/generate.py | 61 +++++++++++++++ .../example/pytorch-models/llama2/README.md | 74 +++++++++++++++++++ .../example/pytorch-models/llama2/generate.py | 65 ++++++++++++++++ .../openai-whisper/readme.md | 0 .../openai-whisper/recognize.py | 0 7 files changed, 279 insertions(+) create mode 100644 python/llm/example/pytorch-models/README.md create mode 100644 python/llm/example/pytorch-models/chatglm/README.md create mode 100644 python/llm/example/pytorch-models/chatglm/generate.py create mode 100644 python/llm/example/pytorch-models/llama2/README.md create mode 100644 python/llm/example/pytorch-models/llama2/generate.py rename python/llm/example/{pytorch-model => pytorch-models}/openai-whisper/readme.md (100%) rename python/llm/example/{pytorch-model => pytorch-models}/openai-whisper/recognize.py (100%) diff --git a/python/llm/example/pytorch-models/README.md b/python/llm/example/pytorch-models/README.md new file mode 100644 index 00000000..ad60d414 --- /dev/null +++ b/python/llm/example/pytorch-models/README.md @@ -0,0 +1,21 @@ +# BigDL-LLM INT4 Optimization for Large Language Model +You can use `optimize_model` API to accelerate general PyTorch models on Intel servers and PCs. This directory contains example scripts to help you quickly get started using BigDL-LLM to run some popular open-source models in the community. Each model has its own dedicated folder, where you can find detailed instructions on how to install and run it. + +# Verified models +| Model | Example | +|-----------|----------------------------------------------------------| +| LLaMA 2 | [link](llama2) | +| ChatGLM | [link](chatglm) | +| Openai Whisper | [link](openai-whisper) | + +## Recommended Requirements +To run the examples, we recommend using Intel® Xeon® processors (server), or >= 12th Gen Intel® Core™ processor (client). + +For OS, BigDL-LLM supports Ubuntu 20.04 or later, CentOS 7 or later, and Windows 10/11. + +## Best Known Configuration on Linux +For better performance, it is recommended to set environment variables on Linux with the help of BigDL-Nano: +```bash +pip install bigdl-nano +source bigdl-nano-init +``` diff --git a/python/llm/example/pytorch-models/chatglm/README.md b/python/llm/example/pytorch-models/chatglm/README.md new file mode 100644 index 00000000..c71bd436 --- /dev/null +++ b/python/llm/example/pytorch-models/chatglm/README.md @@ -0,0 +1,58 @@ +# ChatGLM +In this directory, you will find examples on how you could use BigDL-LLM `optimize_model` API to accelerate ChatGLM models. For illustration purposes, we utilize the [THUDM/chatglm-6b](https://huggingface.co/THUDM/chatglm-6b) as a reference ChatGLM model. + +## Requirements +To run these examples with BigDL-LLM, 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 ChatGLM model to predict the next N tokens using `generate()` API, with BigDL-LLM INT4 optimizations. +### 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 + +pip install --pre --upgrade bigdl-llm[all] # install the latest bigdl-llm nightly build with 'all' option +``` + +### 2. Run +After setting up the Python environment, you could run the example by following steps. + +#### 2.1 Client +On client Windows machines, it is recommended to run directly with full utilization of all cores: +```powershell +python ./generate.py --prompt 'AI是什么？' +``` +More information about arguments can be found in [Arguments Info](#23-arguments-info) section. The expected output can be found in [Sample Output](#24-sample-output) section. + +#### 2.2 Server +For optimal performance on server, it is recommended to set several environment variables (refer to [here](../README.md#best-known-configuration-on-linux) for more information), and run the example with all the physical cores of a single socket. + +E.g. on Linux, +```bash +# set BigDL-Nano env variables +source bigdl-nano-init + +# e.g. for a server with 48 cores per socket +export OMP_NUM_THREADS=48 +numactl -C 0-47 -m 0 python ./generate.py --prompt 'AI是什么？' +``` +More information about arguments can be found in [Arguments Info](#23-arguments-info) section. The expected output can be found in [Sample Output](#24-sample-output) section. + +#### 2.3 Arguments Info +In the example, several arguments can be passed to satisfy your requirements: + +- `--repo-id-or-model-path`: str, argument defining the huggingface repo id for the ChatGLM model to be downloaded, or the path to the huggingface checkpoint folder. It is default to be `'THUDM/chatglm-6b'`. +- `--prompt`: str, argument defining the prompt to be inferred (with integrated prompt format for chat). It is default to be `'AI是什么？'`. +- `--n-predict`: int, argument defining the max number of tokens to predict. It is default to be `32`. + +#### 2.4 Sample Output +#### [THUDM/chatglm-6b](https://huggingface.co/THUDM/chatglm-6b) +```log +Inference time: xxxx s +-------------------- Output -------------------- +问:AI是什么? +答: AI是人工智能(Artificial Intelligence)的缩写,指的是一种能够模拟人类智能的技术或系统。AI包括机器学习、深度学习、自然语言处理、计算机视觉 +``` diff --git a/python/llm/example/pytorch-models/chatglm/generate.py b/python/llm/example/pytorch-models/chatglm/generate.py new file mode 100644 index 00000000..d3f4b6cd --- /dev/null +++ b/python/llm/example/pytorch-models/chatglm/generate.py @@ -0,0 +1,61 @@ +# +# 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 transformers import AutoModel, AutoTokenizer +from bigdl.llm import optimize_model + +# you could tune the prompt based on your own model, +# here the prompt tuning refers to https://huggingface.co/THUDM/chatglm-6b/blob/294cb13118a1e08ad8449ca542624a5c6aecc401/modeling_chatglm.py#L1281 +CHATGLM_V1_PROMPT_FORMAT = "问：{prompt}\n答：" + +if __name__ == '__main__': + parser = argparse.ArgumentParser(description='Predict Tokens using `generate()` API for ChatGLM model') + parser.add_argument('--repo-id-or-model-path', type=str, default="THUDM/chatglm-6b", + help='The huggingface repo id for the ChatGLM 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 = AutoModel.from_pretrained(model_path, trust_remote_code=True) + + # With only one line to enable BigDL-LLM optimization on model + model = optimize_model(model) + + # Load tokenizer + tokenizer = AutoTokenizer.from_pretrained(model_path, trust_remote_code=True) + + # Generate predicted tokens + with torch.inference_mode(): + prompt = CHATGLM_V1_PROMPT_FORMAT.format(prompt=args.prompt) + input_ids = tokenizer.encode(prompt, return_tensors="pt") + 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, 'Output', '-'*20) + print(output_str) diff --git a/python/llm/example/pytorch-models/llama2/README.md b/python/llm/example/pytorch-models/llama2/README.md new file mode 100644 index 00000000..fde21def --- /dev/null +++ b/python/llm/example/pytorch-models/llama2/README.md @@ -0,0 +1,74 @@ +# Llama2 +In this directory, you will find examples on how you could use BigDL-LLM `optimize_model` API to accelerate Llama2 models. For illustration purposes, we utilize the [meta-llama/Llama-2-7b-chat-hf](https://huggingface.co/meta-llama/Llama-2-7b-chat-hf) and [meta-llama/Llama-2-13b-chat-hf](https://huggingface.co/meta-llama/Llama-2-13b-chat-hf) as reference Llama2 models. + +## Requirements +To run these examples with BigDL-LLM, 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. +### 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 + +pip install --pre --upgrade bigdl-llm[all] # install the latest bigdl-llm nightly build with 'all' option +``` + +### 2. Run +After setting up the Python environment, you could run the example by following steps. + +#### 2.1 Client +On client Windows machines, it is recommended to run directly with full utilization of all cores: +```powershell +python ./generate.py --prompt 'What is AI?' +``` +More information about arguments can be found in [Arguments Info](#23-arguments-info) section. The expected output can be found in [Sample Output](#24-sample-output) section. + +#### 2.2 Server +For optimal performance on server, it is recommended to set several environment variables (refer to [here](../README.md#best-known-configuration-on-linux) for more information), and run the example with all the physical cores of a single socket. + +E.g. on Linux, +```bash +# set BigDL-Nano env variables +source bigdl-nano-init + +# e.g. for a server with 48 cores per socket +export OMP_NUM_THREADS=48 +numactl -C 0-47 -m 0 python ./generate.py --prompt 'What is AI?' +``` +More information about arguments can be found in [Arguments Info](#23-arguments-info) section. The expected output can be found in [Sample Output](#24-sample-output) section. + +#### 2.3 Arguments Info +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 Llama2 model (e.g. `meta-llama/Llama-2-7b-chat-hf` and `meta-llama/Llama-2-13b-chat-hf`) to be downloaded, or the path to the huggingface checkpoint folder. It is default to be `'meta-llama/Llama-2-7b-chat-hf'`. +- `--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 +#### [meta-llama/Llama-2-7b-chat-hf](https://huggingface.co/meta-llama/Llama-2-7b-chat-hf) +```log +Inference time: xxxx s +-------------------- Output -------------------- +### HUMAN: +What is AI? + +### RESPONSE: + +AI is a branch of computer science that focuses on creating intelligent machines that can perform tasks that typically require human intelligence, such as understanding natural language, +``` + +#### [meta-llama/Llama-2-13b-chat-hf](https://huggingface.co/meta-llama/Llama-2-13b-chat-hf) +```log +Inference time: xxxx s +-------------------- Output -------------------- +### HUMAN: +What is AI? + +### RESPONSE: + +AI, or artificial intelligence, refers to the ability of machines to perform tasks that would typically require human intelligence, such as learning, problem-solving, +``` diff --git a/python/llm/example/pytorch-models/llama2/generate.py b/python/llm/example/pytorch-models/llama2/generate.py new file mode 100644 index 00000000..b2c5ca70 --- /dev/null +++ b/python/llm/example/pytorch-models/llama2/generate.py @@ -0,0 +1,65 @@ +# +# 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 transformers import AutoModelForCausalLM, LlamaTokenizer +from bigdl.llm import optimize_model + +# 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 +LLAMA2_PROMPT_FORMAT = """### HUMAN: +{prompt} + +### RESPONSE: +""" + +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="meta-llama/Llama-2-7b-chat-hf", + help='The huggingface repo id for the Llama2 (e.g. `meta-llama/Llama-2-7b-chat-hf` and `meta-llama/Llama-2-13b-chat-hf`) to be downloaded' + ', or the path to the huggingface checkpoint folder') + parser.add_argument('--prompt', type=str, default="What is AI?", + 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) + + # With only one line to enable BigDL-LLM optimization on model + model = optimize_model(model) + + # Load tokenizer + tokenizer = LlamaTokenizer.from_pretrained(model_path, trust_remote_code=True) + + # Generate predicted tokens + with torch.inference_mode(): + prompt = LLAMA2_PROMPT_FORMAT.format(prompt=args.prompt) + input_ids = tokenizer.encode(prompt, return_tensors="pt") + 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, 'Output', '-'*20) + print(output_str) diff --git a/python/llm/example/pytorch-model/openai-whisper/readme.md b/python/llm/example/pytorch-models/openai-whisper/readme.md similarity index 100% rename from python/llm/example/pytorch-model/openai-whisper/readme.md rename to python/llm/example/pytorch-models/openai-whisper/readme.md diff --git a/python/llm/example/pytorch-model/openai-whisper/recognize.py b/python/llm/example/pytorch-models/openai-whisper/recognize.py similarity index 100% rename from python/llm/example/pytorch-model/openai-whisper/recognize.py rename to python/llm/example/pytorch-models/openai-whisper/recognize.py