From 457571b44e7e1c0171fa8bd4e90e020e8c1666bf Mon Sep 17 00:00:00 2001 From: binbin Deng <108676127+plusbang@users.noreply.github.com> Date: Wed, 19 Jul 2023 15:15:38 +0800 Subject: [PATCH] [LLM] Add more transformers int4 example (InternLM) (#8557) --- python/llm/README.md | 1 + .../transformers/transformers_int4/README.md | 1 + .../transformers_int4/internlm/README.md | 72 +++++++++++++++++++ .../transformers_int4/internlm/generate.py | 70 ++++++++++++++++++ 4 files changed, 144 insertions(+) create mode 100644 python/llm/example/transformers/transformers_int4/internlm/README.md create mode 100644 python/llm/example/transformers/transformers_int4/internlm/generate.py diff --git a/python/llm/README.md b/python/llm/README.md index 9aa00e19..71220572 100644 --- a/python/llm/README.md +++ b/python/llm/README.md @@ -27,6 +27,7 @@ We may use any Hugging Face Transfomer models on `bigdl-llm`, and the following | RedPajama | [link1](example/transformers/native_int4), [link2](example/transformers/transformers_int4/redpajama) | | Phoenix | [link1](example/transformers/native_int4), [link2](example/transformers/transformers_int4/phoenix) | | StarCoder | [link1](example/transformers/native_int4), [link2](example/transformers/transformers_int4/starcoder) | +| InternLM | [link](example/transformers/transformers_int4/internlm) | ### Working with `bigdl-llm` diff --git a/python/llm/example/transformers/transformers_int4/README.md b/python/llm/example/transformers/transformers_int4/README.md index 9c207a7d..fea6cb53 100644 --- a/python/llm/example/transformers/transformers_int4/README.md +++ b/python/llm/example/transformers/transformers_int4/README.md @@ -15,6 +15,7 @@ You can use BigDL-LLM to run any Huggingface Transformer models with INT4 optimi | RedPajama | [link](redpajama) | | Phoenix | [link](phoenix) | | StarCoder | [link](starcoder) | +| InternLM | [link](internlm) | ## Recommended Requirements To run the examples, we recommend using Intel® Xeon® processors (server), or >= 12th Gen Intel® Core™ processor (client). diff --git a/python/llm/example/transformers/transformers_int4/internlm/README.md b/python/llm/example/transformers/transformers_int4/internlm/README.md new file mode 100644 index 00000000..d9b77293 --- /dev/null +++ b/python/llm/example/transformers/transformers_int4/internlm/README.md @@ -0,0 +1,72 @@ +# InternLM + +In this directory, you will find examples on how you could apply BigDL-LLM INT4 optimizations on InternLM models. For illustration purposes, we utilize the [internlm/internlm-chat-7b-8k](https://huggingface.co/internlm/internlm-chat-7b-8k) as a reference InternLM model. + +## 0. 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 InternLM model to predict the next N tokens using `generate()` API, with BigDL-LLM INT4 optimizations. +### 1. Install +We suggest using conda to manage environment: +```bash +conda create -n llm python=3.9 +conda activate llm + +pip install bigdl-llm[all] # install bigdl-llm with 'all' option +``` + +### 2. Run +``` +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 InternLM model to be downloaded, or the path to the huggingface checkpoint folder. It is default to be `'internlm/internlm-chat-7b-8k'`. +- `--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`. + +> **Note**: When loading the model in 4-bit, BigDL-LLM converts linear layers in the model into INT4 format. In theory, a *X*B model saved in 16-bit will requires approximately 2*X* GB of memory for loading, and ~0.5*X* GB memory for further inference. +> +> Please select the appropriate size of the InternLM model based on the capabilities of your machine. + +#### 2.1 Client +On client Windows machine, it is recommended to run directly with full utilization of all cores: +```powershell +python ./generate.py +``` + +#### 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 +``` + +#### 2.3 Sample Output +#### [internlm/internlm-chat-7b-8k](https://huggingface.co/internlm/internlm-chat-7b-8k) +```log +Inference time: xxxx s +-------------------- Prompt -------------------- +<|User|>:AI是什么？ +<|Bot|>: +-------------------- Output -------------------- +<|User|>:AI是什么？ +<|Bot|>:AI是人工智能的缩写，是计算机科学的一个分支，旨在开发能够执行人类智能任务的算法和计算机程序。AI可以执行各种任务，包括语音 +``` + +```log +Inference time: xxxx s +-------------------- Prompt -------------------- +<|User|>:What is AI? +<|Bot|>: +-------------------- Output -------------------- +<|User|>:What is AI? +<|Bot|>:AI is the ability of machines to perform tasks that would normally require human intelligence, such as perception, reasoning, learning, and decision-making. AI is made possible +``` diff --git a/python/llm/example/transformers/transformers_int4/internlm/generate.py b/python/llm/example/transformers/transformers_int4/internlm/generate.py new file mode 100644 index 00000000..1817079d --- /dev/null +++ b/python/llm/example/transformers/transformers_int4/internlm/generate.py @@ -0,0 +1,70 @@ +# +# 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 +import numpy as np + +from bigdl.llm.transformers import AutoModelForCausalLM +from transformers import AutoTokenizer + +# you could tune the prompt based on your own model, +# here the prompt tuning refers to https://huggingface.co/internlm/internlm-chat-7b-8k/blob/main/modeling_internlm.py#L768 +INTERNLM_PROMPT_FORMAT = "<|User|>:{prompt}\n<|Bot|>:" + +if __name__ == '__main__': + parser = argparse.ArgumentParser(description='Predict Tokens using `generate()` API for InternLM model') + parser.add_argument('--repo-id-or-model-path', type=str, default="internlm/internlm-chat-7b-8k", + help='The huggingface repo id for the InternLM 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 + model = AutoModelForCausalLM.from_pretrained(model_path, + load_in_4bit=True, + trust_remote_code=True) + + # Load tokenizer + tokenizer = AutoTokenizer.from_pretrained(model_path, + trust_remote_code=True) + + # Generate predicted tokens + with torch.inference_mode(): + prompt = INTERNLM_PROMPT_FORMAT.format(prompt=args.prompt) + input_ids = tokenizer.encode(prompt, return_tensors="pt") + 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) + end = time.time() + output_str = tokenizer.decode(output[0], skip_special_tokens=True) + output_str = output_str.split("")[0] + print(f'Inference time: {end-st} s') + print('-'*20, 'Prompt', '-'*20) + print(prompt) + print('-'*20, 'Output', '-'*20) + print(output_str)