From a86487c539f5e5828495f9d653d04a666d413146 Mon Sep 17 00:00:00 2001 From: "Chu,Youcheng" <1340390339@qq.com> Date: Mon, 16 Dec 2024 14:39:19 +0800 Subject: [PATCH] Add GLM-Edge GPU example (#12483) * feat: initial commit * generate.py and README updates * Update link for main readme * Update based on comments * Small fix --------- Co-authored-by: Yuwen Hu --- README.md | 2 +- README.zh-CN.md | 2 +- .../GPU/HuggingFace/LLM/glm-edge/README.md | 147 ++++++++++++++++++ .../GPU/HuggingFace/LLM/glm-edge/generate.py | 85 ++++++++++ 4 files changed, 234 insertions(+), 2 deletions(-) create mode 100644 python/llm/example/GPU/HuggingFace/LLM/glm-edge/README.md create mode 100644 python/llm/example/GPU/HuggingFace/LLM/glm-edge/generate.py diff --git a/README.md b/README.md index 21886b3d..0fed45f5 100644 --- a/README.md +++ b/README.md @@ -264,7 +264,7 @@ Over 70 models have been optimized/verified on `ipex-llm`, including *LLaMA/LLaM | ChatGLM3 | [link](python/llm/example/CPU/HF-Transformers-AutoModels/Model/chatglm3) | [link](python/llm/example/GPU/HuggingFace/LLM/chatglm3) | | GLM-4 | [link](python/llm/example/CPU/HF-Transformers-AutoModels/Model/glm4) | [link](python/llm/example/GPU/HuggingFace/LLM/glm4) | | GLM-4V | [link](python/llm/example/CPU/HF-Transformers-AutoModels/Model/glm-4v) | [link](python/llm/example/GPU/HuggingFace/Multimodal/glm-4v) | -| GLM-Edge | | | [Python link](python/llm/example/NPU/HF-Transformers-AutoModels/LLM) | +| GLM-Edge | | [link](python/llm/example/GPU/HuggingFace/LLM/glm-edge) | [Python link](python/llm/example/NPU/HF-Transformers-AutoModels/LLM) | | Mistral | [link](python/llm/example/CPU/HF-Transformers-AutoModels/Model/mistral) | [link](python/llm/example/GPU/HuggingFace/LLM/mistral) | | Mixtral | [link](python/llm/example/CPU/HF-Transformers-AutoModels/Model/mixtral) | [link](python/llm/example/GPU/HuggingFace/LLM/mixtral) | | Falcon | [link](python/llm/example/CPU/HF-Transformers-AutoModels/Model/falcon) | [link](python/llm/example/GPU/HuggingFace/LLM/falcon) | diff --git a/README.zh-CN.md b/README.zh-CN.md index 93c8b25d..86293807 100644 --- a/README.zh-CN.md +++ b/README.zh-CN.md @@ -264,7 +264,7 @@ See the demo of running [*Text-Generation-WebUI*](https://ipex-llm.readthedocs.i | ChatGLM3 | [link](python/llm/example/CPU/HF-Transformers-AutoModels/Model/chatglm3) | [link](python/llm/example/GPU/HuggingFace/LLM/chatglm3) | | GLM-4 | [link](python/llm/example/CPU/HF-Transformers-AutoModels/Model/glm4) | [link](python/llm/example/GPU/HuggingFace/LLM/glm4) | | GLM-4V | [link](python/llm/example/CPU/HF-Transformers-AutoModels/Model/glm-4v) | [link](python/llm/example/GPU/HuggingFace/Multimodal/glm-4v) | -| GLM-Edge | | | [Python link](python/llm/example/NPU/HF-Transformers-AutoModels/LLM) | +| GLM-Edge | | [link](python/llm/example/GPU/HuggingFace/LLM/glm-edge) | [Python link](python/llm/example/NPU/HF-Transformers-AutoModels/LLM) | | Mistral | [link](python/llm/example/CPU/HF-Transformers-AutoModels/Model/mistral) | [link](python/llm/example/GPU/HuggingFace/LLM/mistral) | | Mixtral | [link](python/llm/example/CPU/HF-Transformers-AutoModels/Model/mixtral) | [link](python/llm/example/GPU/HuggingFace/LLM/mixtral) | | Falcon | [link](python/llm/example/CPU/HF-Transformers-AutoModels/Model/falcon) | [link](python/llm/example/GPU/HuggingFace/LLM/falcon) | diff --git a/python/llm/example/GPU/HuggingFace/LLM/glm-edge/README.md b/python/llm/example/GPU/HuggingFace/LLM/glm-edge/README.md new file mode 100644 index 00000000..a7038c0e --- /dev/null +++ b/python/llm/example/GPU/HuggingFace/LLM/glm-edge/README.md @@ -0,0 +1,147 @@ +# GLM-Edge +In this directory, you will find examples on how you could apply IPEX-LLM INT4 optimizations on GLM-Edge models on [Intel GPUs](../../../README.md). For illustration purposes, we utilize the [THUDM/glm-edge-1.5b-chat](https://huggingface.co/THUDM/glm-edge-1.5b-chat) and [THUDM/glm-edge-4b-chat](https://huggingface.co/THUDM/glm-edge-4b-chat) as reference GLM-Edge models. + +## 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. + +## 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/ + +# install packages required for GLM-Edge +pip install transformers==4.47.0 +pip install accelerate==0.33.0 +pip install "trl<0.12.0" +``` + +### 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/ + +# install packages required for GLM-Edge +pip install transformers==4.47.0 +pip install accelerate==0.33.0 +pip install "trl<0.12.0" +``` + +## 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 +``` + +
+ +### 3.2 Configurations for Windows +
+ +For Intel iGPU and 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 + +### Example 1: Predict Tokens using `generate()` API +In the example [generate.py](./generate.py), we show a basic use case for a GLM-Edge model to predict the next N tokens using `generate()` API, with IPEX-LLM INT4 optimizations on Intel GPUs. + +``` +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 GLM-Edge model (e.g. `THUDM/glm-edge-1.5b-chat` or `THUDM/glm-edge-4b-chat`) to be downloaded, or the path to the huggingface checkpoint folder. It is default to be `'THUDM/glm-edge-4b-chat'`. +- `--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 +#### [THUDM/glm-edge-1.5b-chat](https://huggingface.co/THUDM/glm-edge-1.5b-chat) +```log +Inference time: xxxx s +-------------------- Prompt -------------------- +AI是什么？ +-------------------- Output -------------------- +AI，即人工智能，指的是由人制造出来的系统或机器能够执行通常需要人类智能才能完成的任务。人工智能可以执行多种任务，包括视觉识别、语言 +``` + +```log +Inference time: xxxx s +-------------------- Prompt -------------------- +What is AI? +-------------------- Output -------------------- +Artificial Intelligence, often abbreviated as AI, refers to the simulation of human intelligence in machines that are programmed to think like humans and mimic +``` + +#### [THUDM/glm-edge-4b-chat](https://huggingface.co/THUDM/glm-edge-4b-chat) +```log +Inference time: xxxx s +-------------------- Prompt -------------------- +AI是什么？ +-------------------- Output -------------------- +AI，即人工智能（Artificial Intelligence），是计算机科学的一个分支，旨在开发出一种智能系统，使其能够执行通常需要人类智能才能完成的任务，如视觉 +``` + +```log +Inference time: xxxx s +-------------------- Prompt -------------------- +What is AI? +-------------------- Output -------------------- +Artificial intelligence (AI) refers to the simulation of human intelligence in machines that are programmed to think like humans and mimic their actions. AI systems can +``` diff --git a/python/llm/example/GPU/HuggingFace/LLM/glm-edge/generate.py b/python/llm/example/GPU/HuggingFace/LLM/glm-edge/generate.py new file mode 100644 index 00000000..b02afa18 --- /dev/null +++ b/python/llm/example/GPU/HuggingFace/LLM/glm-edge/generate.py @@ -0,0 +1,85 @@ +# +# 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 ipex_llm.transformers import AutoModelForCausalLM +from transformers import AutoTokenizer + + +if __name__ == '__main__': + parser = argparse.ArgumentParser(description='Predict Tokens using `generate()` API for GLM-Edge model') + parser.add_argument('--repo-id-or-model-path', type=str, default="THUDM/glm-edge-4b-chat", + help='The huggingface repo id for the GLM-Edge 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 + # When running LLMs on Intel iGPUs for Windows users, we recommend setting `cpu_embedding=True` in the from_pretrained function. + # This will allow the memory-intensive embedding layer to utilize the CPU instead of iGPU. + model = AutoModelForCausalLM.from_pretrained(model_path, + load_in_4bit=True, + optimize_model=True, + trust_remote_code=True, + use_cache=True) + model = model.half().to("xpu") + + # Load tokenizer + tokenizer = AutoTokenizer.from_pretrained(model_path, + trust_remote_code=True) + + # Generate predicted tokens + with torch.inference_mode(): + # The following code for generation is adapted from https://huggingface.co/THUDM/glm-edge-1.5b-chat#inference + message = [{"role": "user", "content": args.prompt}] + + inputs = tokenizer.apply_chat_template( + message, + return_tensors="pt", + add_generation_prompt=True, + return_dict=True, + ).to("xpu") + + generate_kwargs = { + "input_ids": inputs["input_ids"], + "attention_mask": inputs["attention_mask"], + "max_new_tokens": args.n_predict, + "do_sample": False, + } + + # ipex_llm model needs a warmup, then inference time can be accurate + output = model.generate(**generate_kwargs) + + st = time.time() + output = model.generate(**generate_kwargs) + torch.xpu.synchronize() + end = time.time() + + output_str = tokenizer.decode(output[0][inputs.input_ids.shape[1]:], skip_special_tokens=True) + print(f'Inference time: {end-st} s') + print('-'*20, 'Prompt', '-'*20) + print(args.prompt) + print('-'*20, 'Output', '-'*20) + print(output_str)