From 2f4c754759fc4ab42912eafe46ae961b9571f86f Mon Sep 17 00:00:00 2001 From: Jin Qiao <89779290+JinBridger@users.noreply.github.com> Date: Wed, 7 Feb 2024 10:47:11 +0800 Subject: [PATCH] LLM: add bark gpu example (#10091) * add bark gpu example * fix * fix license * add bark * add example * fix * another way --- .../GPU/PyTorch-Models/Model/bark/README.md | 120 ++++++++++++++++++ .../Model/bark/synthesize_speech.py | 69 ++++++++++ 2 files changed, 189 insertions(+) create mode 100644 python/llm/example/GPU/PyTorch-Models/Model/bark/README.md create mode 100644 python/llm/example/GPU/PyTorch-Models/Model/bark/synthesize_speech.py diff --git a/python/llm/example/GPU/PyTorch-Models/Model/bark/README.md b/python/llm/example/GPU/PyTorch-Models/Model/bark/README.md new file mode 100644 index 00000000..0ec27177 --- /dev/null +++ b/python/llm/example/GPU/PyTorch-Models/Model/bark/README.md @@ -0,0 +1,120 @@ +# Bark +In this directory, you will find examples on how you could use BigDL-LLM `optimize_model` API to accelerate Bark models. For illustration purposes, we utilize the [suno/bark-small](https://huggingface.co/suno/bark-small) as reference Bark 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: Synthesize speech with the given input text +In the example [synthesize_speech.py](./synthesize_speech.py), we show a basic use case for Bark model to synthesize speech based on the given text, with BigDL-LLM INT4 optimizations. +### 1. Install +#### 1.1 Installation on Linux +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==2.1.10+xpu as default +pip install --pre --upgrade bigdl-llm[xpu] -f https://developer.intel.com/ipex-whl-stable-xpu +pip install scipy +``` + +#### 1.2 Installation on Windows +We suggest using conda to manage environment: +```bash +conda create -n llm python=3.9 libuv +conda activate llm + +# below command will install intel_extension_for_pytorch==2.1.10+xpu as default +pip install --pre --upgrade bigdl-llm[xpu] -f https://developer.intel.com/ipex-whl-stable-xpu +pip install scipy +``` + +### 2. Configures OneAPI environment variables +#### 2.1 Configurations for Linux +```bash +source /opt/intel/oneapi/setvars.sh +``` + +#### 2.2 Configurations for Windows +```cmd +call "C:\Program Files (x86)\Intel\oneAPI\setvars.bat" +``` +> Note: Please make sure you are using **CMD** (**Anaconda Prompt** if using conda) to run the command as PowerShell is not supported. +### 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 +``` + +
+ +
+ +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 ENABLE_SDP_FUSION=1 +``` +> Note: Please note that `libtcmalloc.so` can be installed by `conda install -c conda-forge -y gperftools=2.10`. +
+ +#### 3.2 Configurations for Windows +
+ +For Intel iGPU + +```cmd +set SYCL_CACHE_PERSISTENT=1 +set BIGDL_LLM_XMX_DISABLED=1 +``` + +
+ +
+ +For Intel Arc™ A300-Series or Pro A60 + +```cmd +set SYCL_CACHE_PERSISTENT=1 +``` + +
+ +
+ +For other Intel dGPU Series + +There is no need to set further environment variables. + +
+ +> 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 + +```bash +python ./synthesize_speech.py --text 'BigDL-LLM is a library for running large language model on Intel XPU with very low latency.' +``` + +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 Bark model (e.g. `suno/bark-small` and `suno/bark`) to be downloaded, or the path to the huggingface checkpoint folder. It is default to be `'suno/bark-small'`. +- `--voice-preset`: argument defining the voice preset of model. It is default to be `'v2/en_speaker_6'`. +- `--text TEXT`: argument defining the text to synthesize speech. It is default to be `"BigDL-LLM is a library for running large language model on Intel XPU with very low latency."`. + +#### 4.1 Sample Output + +#### [suno/bark-small](https://huggingface.co/suno/bark-small) + +Text: BigDL-LLM is a library for running large language model on Intel XPU with very low latency. + +[Click here to hear sample output.](https://llm-assets.readthedocs.io/en/latest/_downloads/e92874986553193acbd321d1cfe29739/bark-example-output.wav) diff --git a/python/llm/example/GPU/PyTorch-Models/Model/bark/synthesize_speech.py b/python/llm/example/GPU/PyTorch-Models/Model/bark/synthesize_speech.py new file mode 100644 index 00000000..c7c3a3f0 --- /dev/null +++ b/python/llm/example/GPU/PyTorch-Models/Model/bark/synthesize_speech.py @@ -0,0 +1,69 @@ +# +# 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 scipy +import time +import argparse + +from bigdl.llm import optimize_model +from transformers import AutoProcessor, BarkModel + + +if __name__ == '__main__': + parser = argparse.ArgumentParser(description='Synthesize speech with the given input text using Bark model') + parser.add_argument('--repo-id-or-model-path', type=str, default='suno/bark-small', + help='The huggingface repo id for the Bark model to be downloaded' + ', or the path to the huggingface checkpoint folder') + parser.add_argument('--voice-preset', type=str, default='v2/en_speaker_6', + help='The voice preset of model') + parser.add_argument('--text', type=str, default="BigDL-LLM is a library for running large language model on Intel XPU with very low latency.", + help='Text to synthesize speech') + + args = parser.parse_args() + model_path = args.repo_id_or_model_path + voice_preset = args.voice_preset + text = args.text + + # Load model + processor = AutoProcessor.from_pretrained(model_path) + model = BarkModel.from_pretrained(model_path) + + # With only one line to enable BigDL-LLM optimization on model + # When running LLMs on Intel iGPUs for Windows users, we recommend setting `cpu_embedding=True` in the optimize_model function. + # This will allow the memory-intensive embedding layer to utilize the CPU instead of iGPU. + model = optimize_model(model) + model = model.to('xpu') + + inputs = processor(text, voice_preset=voice_preset).to('xpu') + + with torch.inference_mode(): + # ipex model needs a warmup, then inference time can be accurate + audio_array = model.generate(**inputs) + + st = time.time() + audio_array = model.generate(**inputs) + torch.xpu.synchronize() + end = time.time() + + print(f"Inference time: {end-st} s") + + audio_array = audio_array.cpu().numpy().squeeze() + + from scipy.io.wavfile import write as write_wav + sample_rate = model.generation_config.sample_rate + write_wav("bigdl_llm_bark_out.wav", sample_rate, audio_array)