From 63b2556ce2d86160107f77696d4f1e009aaa9b46 Mon Sep 17 00:00:00 2001 From: "Zheng, Yi" <107673535+yzheng124@users.noreply.github.com> Date: Thu, 2 Nov 2023 15:10:45 +0800 Subject: [PATCH] Add cpu examples of skywork (#9340) --- README.md | 1 + python/llm/README.md | 1 + .../Model/skywork/README.md | 60 +++++++++++++++++ .../Model/skywork/generate.py | 67 +++++++++++++++++++ .../PyTorch-Models/Model/skywork/README.md | 60 +++++++++++++++++ .../PyTorch-Models/Model/skywork/generate.py | 64 ++++++++++++++++++ 6 files changed, 253 insertions(+) create mode 100644 python/llm/example/CPU/HF-Transformers-AutoModels/Model/skywork/README.md create mode 100644 python/llm/example/CPU/HF-Transformers-AutoModels/Model/skywork/generate.py create mode 100644 python/llm/example/CPU/PyTorch-Models/Model/skywork/README.md create mode 100644 python/llm/example/CPU/PyTorch-Models/Model/skywork/generate.py diff --git a/README.md b/README.md index 6ee92894599..e4c295c0df6 100644 --- a/README.md +++ b/README.md @@ -157,6 +157,7 @@ Over 20 models have been optimized/verified on `bigdl-llm`, including *LLaMA/LLa | Flan-t5 | [link](python/llm/example/CPU/HF-Transformers-AutoModels/Model/flan-t5) | [link](python/llm/example/GPU/HF-Transformers-AutoModels/Model/flan-t5) | | Qwen-VL | [link](python/llm/example/CPU/HF-Transformers-AutoModels/Model/qwen-vl) | | | LLaVA | [link](python/llm/example/CPU/PyTorch-Models/Model/llava) | [link](python/llm/example/GPU/PyTorch-Models/Model/llava) | +| Skywork | [link](python/llm/example/CPU/HF-Transformers-AutoModels/Model/skywork) | | ***For more details, please refer to the `bigdl-llm` [Document](https://test-bigdl-llm.readthedocs.io/en/main/doc/LLM/index.html), [Readme](python/llm), [Tutorial](https://github.com/intel-analytics/bigdl-llm-tutorial) and [API Doc](https://bigdl.readthedocs.io/en/latest/doc/PythonAPI/LLM/index.html).*** diff --git a/python/llm/README.md b/python/llm/README.md index 0d4f7111534..58ae2ab9221 100644 --- a/python/llm/README.md +++ b/python/llm/README.md @@ -64,6 +64,7 @@ Over 20 models have been optimized/verified on `bigdl-llm`, including *LLaMA/LLa | Flan-t5 | [link](example/CPU/HF-Transformers-AutoModels/Model/flan-t5) | [link](example/GPU/HF-Transformers-AutoModels/Model/flan-t5) | | Qwen-VL | [link](example/CPU/HF-Transformers-AutoModels/Model/qwen-vl) | | | LLaVA | [link](example/CPU/PyTorch-Models/Model/llava) | [link](example/GPU/PyTorch-Models/Model/llava) | +| Skywork | [link](example/CPU/HF-Transformers-AutoModels/Model/skywork) | | ### Working with `bigdl-llm` diff --git a/python/llm/example/CPU/HF-Transformers-AutoModels/Model/skywork/README.md b/python/llm/example/CPU/HF-Transformers-AutoModels/Model/skywork/README.md new file mode 100644 index 00000000000..7e0ab293fdf --- /dev/null +++ b/python/llm/example/CPU/HF-Transformers-AutoModels/Model/skywork/README.md @@ -0,0 +1,60 @@ +# Skywork +In this directory, you will find examples on how you could apply BigDL-LLM INT4 optimizations on Skywork models. For illustration purposes, we utilize the [Skywork/Skywork-13B-base](https://huggingface.co/Skywork/Skywork-13B-base) as the reference Skywork 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 Skywork 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 --pre --upgrade bigdl-llm[all] # install the latest bigdl-llm nightly build 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 Skywork model (e.g. `Skywork/Skywork-13B-base`) to be downloaded, or the path to the huggingface checkpoint folder. It is default to be `'Skywork/Skywork-13B-base'`. +- `--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 Skywork 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 +#### [Skywork/Skywork-13B-base](https://huggingface.co/Skywork/Skywork-13B-base) +```log +Inference time: xxxx s +-------------------- Prompt -------------------- +AI是什么？ +-------------------- Output -------------------- +AI是什么？ +人工智能（Artificial Intelligence），英文缩写为AI。它是研究、开发用于模拟、延伸和扩展人的智能的理论、 +``` diff --git a/python/llm/example/CPU/HF-Transformers-AutoModels/Model/skywork/generate.py b/python/llm/example/CPU/HF-Transformers-AutoModels/Model/skywork/generate.py new file mode 100644 index 00000000000..7a0ae02495a --- /dev/null +++ b/python/llm/example/CPU/HF-Transformers-AutoModels/Model/skywork/generate.py @@ -0,0 +1,67 @@ +# +# 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 bigdl.llm.transformers import AutoModelForCausalLM +from transformers import AutoTokenizer + +# you could tune the prompt based on your own model, +SKYWORK_PROMPT_FORMAT = "{prompt}" + +if __name__ == '__main__': + parser = argparse.ArgumentParser(description='Predict Tokens using `generate()` API for Skywork model') + parser.add_argument('--repo-id-or-model-path', type=str, default="Skywork/Skywork-13B-base", + help='The huggingface repo id for the Skywork 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 = SKYWORK_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) + print(f'Inference time: {end-st} s') + print('-'*20, 'Prompt', '-'*20) + print(prompt) + print('-'*20, 'Output', '-'*20) + print(output_str) diff --git a/python/llm/example/CPU/PyTorch-Models/Model/skywork/README.md b/python/llm/example/CPU/PyTorch-Models/Model/skywork/README.md new file mode 100644 index 00000000000..70834a77ede --- /dev/null +++ b/python/llm/example/CPU/PyTorch-Models/Model/skywork/README.md @@ -0,0 +1,60 @@ +# Skywork +In this directory, you will find examples on how you could use BigDL-LLM `optimize_model` API to accelerate Skywork models. For illustration purposes, we utilize the [Skywork/Skywork-13B-base](https://huggingface.co/Skywork/Skywork-13B-base) as the reference Skywork 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 Skywork 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 +``` +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 +``` +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 Skywork model (e.g. `Skywork/Skywork-13B-base`) to be downloaded, or the path to the huggingface checkpoint folder. It is default to be `'Skywork/Skywork-13B-base'`. +- `--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 +#### [Skywork/Skywork-13B-base](https://huggingface.co/Skywork/Skywork-13B-base) +```log +Inference time: xxxx s +-------------------- Prompt -------------------- +AI是什么？ +-------------------- Output -------------------- +AI是什么？ +AI（Artificial Intelligence）是人工智能的英文简称，指的是一门研究如何让机器具备人类智能的学科。人工智能的 +``` diff --git a/python/llm/example/CPU/PyTorch-Models/Model/skywork/generate.py b/python/llm/example/CPU/PyTorch-Models/Model/skywork/generate.py new file mode 100644 index 00000000000..8528de69258 --- /dev/null +++ b/python/llm/example/CPU/PyTorch-Models/Model/skywork/generate.py @@ -0,0 +1,64 @@ +# +# 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, AutoTokenizer +from bigdl.llm import optimize_model + +# you could tune the prompt based on your own model, +SKYWORK_PROMPT_FORMAT = "{prompt}" + +if __name__ == '__main__': + parser = argparse.ArgumentParser(description='Predict Tokens using `generate()` API for Skywork model') + parser.add_argument('--repo-id-or-model-path', type=str, default="Skywork/Skywork-13B-base", + help='The huggingface repo id for the Skywork 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) + + # 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 = SKYWORK_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, 'Prompt', '-'*20) + print(prompt) + print('-'*20, 'Output', '-'*20) + print(output_str)