From 8c36b5bddee0d16b747255e673499856554db799 Mon Sep 17 00:00:00 2001 From: Yuwen Hu <54161268+Oscilloscope98@users.noreply.github.com> Date: Fri, 7 Jun 2024 10:29:33 +0800 Subject: [PATCH] Add qwen2 example (#11252) * Add GPU example for Qwen2 * Update comments in README * Update README for Qwen2 GPU example * Add CPU example for Qwen2 Sample Output under README pending * Update generate.py and README for CPU Qwen2 * Update GPU example for Qwen2 * Small update * Small fix * Add Qwen2 table * Update README for Qwen2 CPU and GPU Update sample output under README --------- Co-authored-by: Zijie Li --- README.md | 1 + docs/readthedocs/source/index.rst | 7 + .../Model/qwen2/README.md | 83 +++++++++++ .../Model/qwen2/generate.py | 80 +++++++++++ .../CPU/PyTorch-Models/Model/qwen2/README.md | 84 +++++++++++ .../PyTorch-Models/Model/qwen2/generate.py | 82 +++++++++++ .../Model/qwen2/README.md | 134 ++++++++++++++++++ .../Model/qwen2/generate.py | 92 ++++++++++++ .../GPU/PyTorch-Models/Model/qwen2/README.md | 134 ++++++++++++++++++ .../PyTorch-Models/Model/qwen2/generate.py | 91 ++++++++++++ 10 files changed, 788 insertions(+) create mode 100644 python/llm/example/CPU/HF-Transformers-AutoModels/Model/qwen2/README.md create mode 100644 python/llm/example/CPU/HF-Transformers-AutoModels/Model/qwen2/generate.py create mode 100644 python/llm/example/CPU/PyTorch-Models/Model/qwen2/README.md create mode 100644 python/llm/example/CPU/PyTorch-Models/Model/qwen2/generate.py create mode 100644 python/llm/example/GPU/HF-Transformers-AutoModels/Model/qwen2/README.md create mode 100644 python/llm/example/GPU/HF-Transformers-AutoModels/Model/qwen2/generate.py create mode 100644 python/llm/example/GPU/PyTorch-Models/Model/qwen2/README.md create mode 100644 python/llm/example/GPU/PyTorch-Models/Model/qwen2/generate.py diff --git a/README.md b/README.md index c996d430de2..adc1eac3791 100644 --- a/README.md +++ b/README.md @@ -169,6 +169,7 @@ Over 50 models have been optimized/verified on `ipex-llm`, including *LLaMA/LLaM | InternLM | [link](python/llm/example/CPU/HF-Transformers-AutoModels/Model/internlm) | [link](python/llm/example/GPU/HF-Transformers-AutoModels/Model/internlm) | | Qwen | [link](python/llm/example/CPU/HF-Transformers-AutoModels/Model/qwen) | [link](python/llm/example/GPU/HF-Transformers-AutoModels/Model/qwen) | | Qwen1.5 | [link](python/llm/example/CPU/HF-Transformers-AutoModels/Model/qwen1.5) | [link](python/llm/example/GPU/HF-Transformers-AutoModels/Model/qwen1.5) | +| Qwen2 | [link](python/llm/example/CPU/HF-Transformers-AutoModels/Model/qwen2) | [link](python/llm/example/GPU/HF-Transformers-AutoModels/Model/qwe2) | | Qwen-VL | [link](python/llm/example/CPU/HF-Transformers-AutoModels/Model/qwen-vl) | [link](python/llm/example/GPU/HF-Transformers-AutoModels/Model/qwen-vl) | | Aquila | [link](python/llm/example/CPU/HF-Transformers-AutoModels/Model/aquila) | [link](python/llm/example/GPU/HF-Transformers-AutoModels/Model/aquila) | | Aquila2 | [link](python/llm/example/CPU/HF-Transformers-AutoModels/Model/aquila2) | [link](python/llm/example/GPU/HF-Transformers-AutoModels/Model/aquila2) | diff --git a/docs/readthedocs/source/index.rst b/docs/readthedocs/source/index.rst index 7c4d77cb2ac..fb883121dac 100644 --- a/docs/readthedocs/source/index.rst +++ b/docs/readthedocs/source/index.rst @@ -363,6 +363,13 @@ Verified Models link + + Qwen2 + + link + + link + Qwen-VL diff --git a/python/llm/example/CPU/HF-Transformers-AutoModels/Model/qwen2/README.md b/python/llm/example/CPU/HF-Transformers-AutoModels/Model/qwen2/README.md new file mode 100644 index 00000000000..423de9b429f --- /dev/null +++ b/python/llm/example/CPU/HF-Transformers-AutoModels/Model/qwen2/README.md @@ -0,0 +1,83 @@ +# Qwen2 + +In this directory, you will find examples on how you could apply IPEX-LLM INT4 optimizations on Qwen2 models. For illustration purposes, we utilize the [Qwen/Qwen2-7B-Instruct](https://huggingface.co/Qwen/Qwen2-7B-Instruct) as a reference Qwen2 model. + +## 0. Requirements +To run these examples with IPEX-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 Qwen model to predict the next N tokens using `generate()` API, with IPEX-LLM INT4 optimizations. +### 1. Install +We suggest using conda to manage environment: + +On Linux: + +```bash +conda create -n llm python=3.11 +conda activate llm + +# install ipex-llm with 'all' option +pip install --pre --upgrade ipex-llm[all] --extra-index-url https://download.pytorch.org/whl/cpu +pip install transformers==4.37.0 # install the transformers which support Qwen2 +``` + +On Windows: + +```cmd +conda create -n llm python=3.11 +conda activate llm + +pip install --pre --upgrade ipex-llm[all] +pip install transformers==4.37.0 +``` + +### 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 Qwen model to be downloaded, or the path to the huggingface checkpoint folder. It is default to be `'Qwen/Qwen2-7B-Instruct'`. +- `--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, IPEX-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 Qwen 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: +```cmd +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 IPEX-LLM env variables +source ipex-llm-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 +##### [Qwen/Qwen2-7B-Instruct](https://huggingface.co/Qwen/Qwen2-7B-Instruct) +```log +Inference time: xxxx s +-------------------- Prompt -------------------- +AI是什么？ +-------------------- Output -------------------- +AI，即人工智能（Artificial Intelligence），是一种计算机科学领域，旨在开发能够模拟、延伸和增强人类智能的算法和系统。人工 智能涉及许多 +``` + +```log +Inference time: xxxx s +-------------------- Prompt -------------------- +What is AI? +-------------------- Output -------------------- +AI stands for Artificial Intelligence, which is the simulation of human intelligence in machines that are programmed to think and learn like humans and mimic their actions. The term may +``` \ No newline at end of file diff --git a/python/llm/example/CPU/HF-Transformers-AutoModels/Model/qwen2/generate.py b/python/llm/example/CPU/HF-Transformers-AutoModels/Model/qwen2/generate.py new file mode 100644 index 00000000000..90626539427 --- /dev/null +++ b/python/llm/example/CPU/HF-Transformers-AutoModels/Model/qwen2/generate.py @@ -0,0 +1,80 @@ +# +# 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 transformers import AutoTokenizer + + +if __name__ == '__main__': + parser = argparse.ArgumentParser(description='Qwen2-7B-Instruct') + parser.add_argument('--repo-id-or-model-path', type=str, default="Qwen/Qwen2-7B-Instruct", + help='The huggingface repo id for the Qwen2 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 + + + from ipex_llm.transformers import AutoModelForCausalLM + model = AutoModelForCausalLM.from_pretrained(model_path, + load_in_4bit=True, + optimize_model=True, + trust_remote_code=True, + use_cache=True) + + # Load tokenizer + tokenizer = AutoTokenizer.from_pretrained(model_path, + trust_remote_code=True) + + prompt = args.prompt + + # Generate predicted tokens + with torch.inference_mode(): + # The following code for generation is adapted from https://huggingface.co/Qwen/Qwen2-7B-Instruct#quickstart + messages = [ + {"role": "system", "content": "You are a helpful assistant."}, + {"role": "user", "content": prompt} + ] + text = tokenizer.apply_chat_template( + messages, + tokenize=False, + add_generation_prompt=True + ) + model_inputs = tokenizer([text], return_tensors="pt") + st = time.time() + generated_ids = model.generate( + model_inputs.input_ids, + max_new_tokens=args.n_predict + ) + end = time.time() + generated_ids = [ + output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids) + ] + + response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0] + print(f'Inference time: {end-st} s') + print('-'*20, 'Prompt', '-'*20) + print(prompt) + print('-'*20, 'Output', '-'*20) + print(response) diff --git a/python/llm/example/CPU/PyTorch-Models/Model/qwen2/README.md b/python/llm/example/CPU/PyTorch-Models/Model/qwen2/README.md new file mode 100644 index 00000000000..84d7cc1293f --- /dev/null +++ b/python/llm/example/CPU/PyTorch-Models/Model/qwen2/README.md @@ -0,0 +1,84 @@ +# Qwen2 +In this directory, you will find examples on how you could use IPEX-LLM `optimize_model` API to accelerate Qwen2 models. For illustration purposes, we utilize the [Qwen/Qwen2-7B-Instruct](https://huggingface.co/Qwen/Qwen2-7B-Instruct) as reference Qwen2 model. + +## Requirements +To run these examples with IPEX-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 Qwen2 model to predict the next N tokens using `generate()` API, with IPEX-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://conda-forge.org/download/). + +After installing conda, create a Python environment for IPEX-LLM: + +On Linux: + +```bash +conda create -n llm python=3.11 # recommend to use Python 3.11 +conda activate llm + +# install the latest ipex-llm nightly build with 'all' option +pip install --pre --upgrade ipex-llm[all] --extra-index-url https://download.pytorch.org/whl/cpu +pip install transformers==4.37.0 # install transformers which supports Qwen2 +``` + +On Windows: + +```cmd +conda create -n llm python=3.11 +conda activate llm + +pip install --pre --upgrade ipex-llm[all] +pip install transformers==4.37.0 +``` + +### 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 Qwen2 to be downloaded, or the path to the huggingface checkpoint folder. It is default to be `'Qwen/Qwen2-7B-Instruct'`. +- `--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, IPEX-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 Qwen model based on the capabilities of your machine. + +#### 2.1 Client +On client Windows machines, it is recommended to run directly with full utilization of all cores: +```cmd +python ./generate.py --prompt 'What is AI?' +``` + +#### 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 IPEX-LLM env variables +source ipex-llm-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?' +``` + +#### 2.3 Sample Output +##### [Qwen/Qwen2-7B-Instruct](https://huggingface.co/Qwen/Qwen2-7B-Instruct) +```log +Inference time: xxxx s +-------------------- Prompt -------------------- +AI是什么？ +-------------------- Output -------------------- +AI，即人工智能（Artificial Intelligence），是一门研究、开发用于模拟、延伸和扩展人的智能的理论、方法、技术及应用系统的学科 +``` + +```log +Inference time: xxxx s +-------------------- Prompt -------------------- +What is AI? +-------------------- Output -------------------- +AI stands for Artificial Intelligence, which refers to the development of computer systems that can perform tasks that typically require human intelligence. These tasks may include learning from experience, +``` diff --git a/python/llm/example/CPU/PyTorch-Models/Model/qwen2/generate.py b/python/llm/example/CPU/PyTorch-Models/Model/qwen2/generate.py new file mode 100644 index 00000000000..0c9b428abd6 --- /dev/null +++ b/python/llm/example/CPU/PyTorch-Models/Model/qwen2/generate.py @@ -0,0 +1,82 @@ +# +# 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 transformers import AutoTokenizer + + +if __name__ == '__main__': + parser = argparse.ArgumentParser(description='Qwen2-7B-Instruct') + parser.add_argument('--repo-id-or-model-path', type=str, default="Qwen/Qwen2-7B-Instruct", + help='The huggingface repo id for the Qwen2 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 + + + from transformers import AutoModelForCausalLM + model = AutoModelForCausalLM.from_pretrained(model_path, + trust_remote_code=True, + torch_dtype='auto', + low_cpu_mem_usage=True, + use_cache=True) + + + # Load tokenizer + tokenizer = AutoTokenizer.from_pretrained(model_path, + trust_remote_code=True) + from ipex_llm import optimize_model + model = optimize_model(model) + + prompt = args.prompt + # Generate predicted tokens + with torch.inference_mode(): + # The following code for generation is adapted from https://huggingface.co/Qwen/Qwen2-7B-Instruct#quickstart + messages = [ + {"role": "system", "content": "You are a helpful assistant."}, + {"role": "user", "content": prompt} + ] + text = tokenizer.apply_chat_template( + messages, + tokenize=False, + add_generation_prompt=True + ) + model_inputs = tokenizer([text], return_tensors="pt") + st = time.time() + generated_ids = model.generate( + model_inputs.input_ids, + max_new_tokens=args.n_predict + ) + end = time.time() + generated_ids = [ + output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids) + ] + + response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0] + print(f'Inference time: {end-st} s') + print('-'*20, 'Prompt', '-'*20) + print(prompt) + print('-'*20, 'Output', '-'*20) + print(response) diff --git a/python/llm/example/GPU/HF-Transformers-AutoModels/Model/qwen2/README.md b/python/llm/example/GPU/HF-Transformers-AutoModels/Model/qwen2/README.md new file mode 100644 index 00000000000..274b0b47f74 --- /dev/null +++ b/python/llm/example/GPU/HF-Transformers-AutoModels/Model/qwen2/README.md @@ -0,0 +1,134 @@ +# Qwen2 +In this directory, you will find examples on how you could apply IPEX-LLM INT4 optimizations on Qwen2 models on [Intel GPUs](../../../README.md). For illustration purposes, we utilize the [Qwen/Qwen2-7B-Instruct](https://huggingface.co/Qwen/Qwen2-7B-Instruct) as a reference InternLM model. + +## 0. Requirements +To run these examples with IPEX-LLM on Intel GPUs, we have some recommended requirements for your machine, please refer to [here](../../../README.md#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 Qwen2 model to predict the next N tokens using `generate()` API, with IPEX-LLM INT4 optimizations on Intel GPUs. +### 1. Install +#### 1.1 Installation on Linux +We suggest using conda to manage environment: +```bash +conda create -n llm python=3.11 +conda activate llm +# below command will install intel_extension_for_pytorch==2.1.10+xpu as default +pip install --pre --upgrade ipex-llm[xpu] --extra-index-url https://pytorch-extension.intel.com/release-whl/stable/xpu/us/ + +pip install transformers==4.37.0 # install transformers which supports Qwen2 +``` + +#### 1.2 Installation on Windows +We suggest using conda to manage environment: +```bash +conda create -n llm python=3.11 libuv +conda activate llm + +# below command will install intel_extension_for_pytorch==2.1.10+xpu as default +pip install --pre --upgrade ipex-llm[xpu] --extra-index-url https://pytorch-extension.intel.com/release-whl/stable/xpu/us/ + +pip install transformers==4.37.0 # install transformers which supports Qwen2 +``` + +### 2. Configures OneAPI environment variables for Linux + +> [!NOTE] +> Skip this step if you are running on Windows. + +This is a required step on Linux for APT or offline installed oneAPI. Skip this step for PIP-installed oneAPI. + +```bash +source /opt/intel/oneapi/setvars.sh +``` + +### 3. Runtime Configurations +For optimal performance, it is recommended to set several environment variables. Please check out the suggestions based on your device. +#### 3.1 Configurations for Linux +
+ +For Intel Arc™ A-Series Graphics and Intel Data Center GPU Flex Series + +```bash +export USE_XETLA=OFF +export SYCL_PI_LEVEL_ZERO_USE_IMMEDIATE_COMMANDLISTS=1 +export SYCL_CACHE_PERSISTENT=1 +``` + +
+ +
+ +For Intel Data Center GPU Max Series + +```bash +export LD_PRELOAD=${LD_PRELOAD}:${CONDA_PREFIX}/lib/libtcmalloc.so +export SYCL_PI_LEVEL_ZERO_USE_IMMEDIATE_COMMANDLISTS=1 +export SYCL_CACHE_PERSISTENT=1 +export ENABLE_SDP_FUSION=1 +``` +> Note: Please note that `libtcmalloc.so` can be installed by `conda install -c conda-forge -y gperftools=2.10`. +
+ +
+ +For Intel iGPU + +```bash +export SYCL_CACHE_PERSISTENT=1 +export BIGDL_LLM_XMX_DISABLED=1 +``` + +
+ +#### 3.2 Configurations for Windows +
+ +For Intel iGPU + +```cmd +set SYCL_CACHE_PERSISTENT=1 +set BIGDL_LLM_XMX_DISABLED=1 +``` + +
+ +
+ +For Intel Arc™ A-Series Graphics + +```cmd +set SYCL_CACHE_PERSISTENT=1 +``` + +
+ +> [!NOTE] +> For the first time that each model runs on Intel iGPU/Intel Arc™ A300-Series or Pro A60, it may take several minutes to compile. +### 4. Running examples + +``` +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 Qwen2 model (e.g. `Qwen/Qwen2-7B-Instruct`) to be downloaded, or the path to the huggingface checkpoint folder. It is default to be `'Qwen/Qwen2-7B-Instruct'`. +- `--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`. + +#### Sample Output +##### [Qwen/Qwen2-7B-Instruct](https://huggingface.co/Qwen/Qwen2-7B-Instruct) +```log +Inference time: xxxx s +-------------------- Prompt -------------------- +AI是什么？ +-------------------- Output -------------------- +AI，即人工智能（Artificial Intelligence），是一种模拟人类智能行为的技术。它使计算机系统能够执行类似于人类的思维活动，例 如学习、推理 +``` + +```log +Inference time: xxxx s +-------------------- Prompt -------------------- +What is AI? +-------------------- Output -------------------- +AI, or Artificial Intelligence, refers to the simulation of human intelligence in machines that are programmed to think and learn like humans and mimic their actions. The term may +``` \ No newline at end of file diff --git a/python/llm/example/GPU/HF-Transformers-AutoModels/Model/qwen2/generate.py b/python/llm/example/GPU/HF-Transformers-AutoModels/Model/qwen2/generate.py new file mode 100644 index 00000000000..25fdaeec16a --- /dev/null +++ b/python/llm/example/GPU/HF-Transformers-AutoModels/Model/qwen2/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 time +import argparse + +from transformers import AutoTokenizer +from ipex_llm import optimize_model +import numpy as np + + +if __name__ == '__main__': + parser = argparse.ArgumentParser(description='Qwen2-7B-Instruct') + parser.add_argument('--repo-id-or-model-path', type=str, default="Qwen/Qwen2-7B-Instruct", + help='The huggingface repo id for the Qwen2 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 + + + from ipex_llm.transformers import AutoModelForCausalLM + # 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, + optimize_model=True, + trust_remote_code=True, + use_cache=True) + model = model.to("xpu") + + # Load tokenizer + tokenizer = AutoTokenizer.from_pretrained(model_path, + trust_remote_code=True) + + prompt = args.prompt + + # Generate predicted tokens + with torch.inference_mode(): + # The following code for generation is adapted from https://huggingface.co/Qwen/Qwen2-7B-Instruct#quickstart + messages = [ + {"role": "system", "content": "You are a helpful assistant."}, + {"role": "user", "content": prompt} + ] + text = tokenizer.apply_chat_template( + messages, + tokenize=False, + add_generation_prompt=True + ) + model_inputs = tokenizer([text], return_tensors="pt").to("xpu") + # warmup + generated_ids = model.generate( + model_inputs.input_ids, + max_new_tokens=args.n_predict + ) + + st = time.time() + generated_ids = model.generate( + model_inputs.input_ids, + max_new_tokens=args.n_predict + ) + torch.xpu.synchronize() + end = time.time() + generated_ids = generated_ids.cpu() + generated_ids = [ + output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids) + ] + + response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0] + print(f'Inference time: {end-st} s') + print('-'*20, 'Prompt', '-'*20) + print(prompt) + print('-'*20, 'Output', '-'*20) + print(response) \ No newline at end of file diff --git a/python/llm/example/GPU/PyTorch-Models/Model/qwen2/README.md b/python/llm/example/GPU/PyTorch-Models/Model/qwen2/README.md new file mode 100644 index 00000000000..9a3e3e03504 --- /dev/null +++ b/python/llm/example/GPU/PyTorch-Models/Model/qwen2/README.md @@ -0,0 +1,134 @@ +# Qwen2 +In this directory, you will find examples on how you could use IPEX-LLM `optimize_model` API to accelerate Qwen2 models on [Intel GPUs](../../../README.md). For illustration purposes, we utilize the [Qwen/Qwen2-7B-Instruct](https://huggingface.co/Qwen/Qwen2-7B-Instruct) as a reference InternLM model. + +## 0. Requirements +To run these examples with IPEX-LLM on Intel GPUs, we have some recommended requirements for your machine, please refer to [here](../../../README.md#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 Qwen2 model to predict the next N tokens using `generate()` API, with IPEX-LLM INT4 optimizations on Intel GPUs. +### 1. Install +#### 1.1 Installation on Linux +We suggest using conda to manage environment: +```bash +conda create -n llm python=3.11 +conda activate llm +# below command will install intel_extension_for_pytorch==2.1.10+xpu as default +pip install --pre --upgrade ipex-llm[xpu] --extra-index-url https://pytorch-extension.intel.com/release-whl/stable/xpu/us/ + +pip install transformers==4.37.0 # install transformers which supports Qwen2 +``` + +#### 1.2 Installation on Windows +We suggest using conda to manage environment: +```bash +conda create -n llm python=3.11 libuv +conda activate llm + +# below command will install intel_extension_for_pytorch==2.1.10+xpu as default +pip install --pre --upgrade ipex-llm[xpu] --extra-index-url https://pytorch-extension.intel.com/release-whl/stable/xpu/us/ + +pip install transformers==4.37.0 # install transformers which supports Qwen2 +``` + +### 2. Configures OneAPI environment variables for Linux + +> [!NOTE] +> Skip this step if you are running on Windows. + +This is a required step on Linux for APT or offline installed oneAPI. Skip this step for PIP-installed oneAPI. + +```bash +source /opt/intel/oneapi/setvars.sh +``` + +### 3. Runtime Configurations +For optimal performance, it is recommended to set several environment variables. Please check out the suggestions based on your device. +#### 3.1 Configurations for Linux +
+ +For Intel Arc™ A-Series Graphics and Intel Data Center GPU Flex Series + +```bash +export USE_XETLA=OFF +export SYCL_PI_LEVEL_ZERO_USE_IMMEDIATE_COMMANDLISTS=1 +export SYCL_CACHE_PERSISTENT=1 +``` + +
+ +
+ +For Intel Data Center GPU Max Series + +```bash +export LD_PRELOAD=${LD_PRELOAD}:${CONDA_PREFIX}/lib/libtcmalloc.so +export SYCL_PI_LEVEL_ZERO_USE_IMMEDIATE_COMMANDLISTS=1 +export SYCL_CACHE_PERSISTENT=1 +export ENABLE_SDP_FUSION=1 +``` +> Note: Please note that `libtcmalloc.so` can be installed by `conda install -c conda-forge -y gperftools=2.10`. +
+ +
+ +For Intel iGPU + +```bash +export SYCL_CACHE_PERSISTENT=1 +export BIGDL_LLM_XMX_DISABLED=1 +``` + +
+ +#### 3.2 Configurations for Windows +
+ +For Intel iGPU + +```cmd +set SYCL_CACHE_PERSISTENT=1 +set BIGDL_LLM_XMX_DISABLED=1 +``` + +
+ +
+ +For Intel Arc™ A-Series Graphics + +```cmd +set SYCL_CACHE_PERSISTENT=1 +``` + +
+ +> [!NOTE] +> For the first time that each model runs on Intel iGPU/Intel Arc™ A300-Series or Pro A60, it may take several minutes to compile. +### 4. Running examples + +``` +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 Qwen2 model (e.g. `Qwen/Qwen2-7B-Instruct`) to be downloaded, or the path to the huggingface checkpoint folder. It is default to be `'Qwen/Qwen2-7B-Instruct'`. +- `--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`. + +#### Sample Output +##### [Qwen/Qwen2-7B-Instruct](https://huggingface.co/Qwen/Qwen2-7B-Instruct) +```log +Inference time: xxxx s +-------------------- Prompt -------------------- +AI是什么？ +-------------------- Output -------------------- +AI是人工智能（Artificial Intelligence）的缩写。它指的是通过计算机程序、算法和模型来模拟、延伸和扩展人类智能的一门学科 +``` + +```log +Inference time: xxxx s +-------------------- Prompt -------------------- +What is AI? +-------------------- Output -------------------- +AI stands for Artificial Intelligence. It refers to the simulation of human intelligence in machines that are programmed to think and work like humans. This includes learning from experience, +``` \ No newline at end of file diff --git a/python/llm/example/GPU/PyTorch-Models/Model/qwen2/generate.py b/python/llm/example/GPU/PyTorch-Models/Model/qwen2/generate.py new file mode 100644 index 00000000000..c3c19253004 --- /dev/null +++ b/python/llm/example/GPU/PyTorch-Models/Model/qwen2/generate.py @@ -0,0 +1,91 @@ +# +# 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 AutoTokenizer +from ipex_llm import optimize_model +import numpy as np + + +if __name__ == '__main__': + parser = argparse.ArgumentParser(description='Qwen2-7B-Instruct') + parser.add_argument('--repo-id-or-model-path', type=str, default="Qwen/Qwen2-7B-Instruct", + help='The huggingface repo id for the Qwen2 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 + + + from transformers import AutoModelForCausalLM + from ipex_llm import optimize_model + model = AutoModelForCausalLM.from_pretrained(model_path, + trust_remote_code=True, + torch_dtype = 'auto', + low_cpu_mem_usage=True, + use_cache=True) + model = optimize_model(model) + model = model.to("xpu") + + # Load tokenizer + tokenizer = AutoTokenizer.from_pretrained(model_path, + trust_remote_code=True) + + prompt = args.prompt + # Generate predicted tokens + with torch.inference_mode(): + # The following code for generation is adapted from https://huggingface.co/Qwen/Qwen2-7B-Instruct#quickstart + messages = [ + {"role": "system", "content": "You are a helpful assistant."}, + {"role": "user", "content": prompt} + ] + text = tokenizer.apply_chat_template( + messages, + tokenize=False, + add_generation_prompt=True + ) + model_inputs = tokenizer([text], return_tensors="pt").to("xpu") + # warmup + generated_ids = model.generate( + model_inputs.input_ids, + max_new_tokens=args.n_predict + ) + + st = time.time() + generated_ids = model.generate( + model_inputs.input_ids, + max_new_tokens=args.n_predict + ) + torch.xpu.synchronize() + end = time.time() + generated_ids = generated_ids.cpu() + generated_ids = [ + output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids) + ] + + response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0] + print(f'Inference time: {end-st} s') + print('-'*20, 'Prompt', '-'*20) + print(prompt) + print('-'*20, 'Output', '-'*20) + print(response) \ No newline at end of file