From 55ce091242179b936450dba1091c616d40fc6293 Mon Sep 17 00:00:00 2001 From: "Xu, Shuo" <100334393+ATMxsp01@users.noreply.github.com> Date: Fri, 27 Dec 2024 09:40:29 +0800 Subject: [PATCH] Add GLM4-Edge-V GPU example (#12596) * Add GLM4-Edge-V examples * polish readme * revert wrong changes * polish readme * polish readme * little polish in reference info and indent * Small fix and sample output updates * Update main readme --------- Co-authored-by: Yuwen Hu --- README.md | 1 + README.zh-CN.md | 1 + .../Multimodal/glm-edge-v/README.md | 164 ++++++++++++++++++ .../Multimodal/glm-edge-v/generate.py | 122 +++++++++++++ 4 files changed, 288 insertions(+) create mode 100644 python/llm/example/GPU/HuggingFace/Multimodal/glm-edge-v/README.md create mode 100644 python/llm/example/GPU/HuggingFace/Multimodal/glm-edge-v/generate.py diff --git a/README.md b/README.md index 4fc6c6b6..f46fed04 100644 --- a/README.md +++ b/README.md @@ -270,6 +270,7 @@ Over 70 models have been optimized/verified on `ipex-llm`, including *LLaMA/LLaM | 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 | | [link](python/llm/example/GPU/HuggingFace/LLM/glm-edge) | [Python link](python/llm/example/NPU/HF-Transformers-AutoModels/LLM) | +| GLM-Edge-V | | [link](python/llm/example/GPU/HuggingFace/Multimodal/glm-edge-v) | | 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 2bd80347..0567189d 100644 --- a/README.zh-CN.md +++ b/README.zh-CN.md @@ -270,6 +270,7 @@ See the demo of running [*Text-Generation-WebUI*](https://ipex-llm.readthedocs.i | 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 | | [link](python/llm/example/GPU/HuggingFace/LLM/glm-edge) | [Python link](python/llm/example/NPU/HF-Transformers-AutoModels/LLM) | +| GLM-Edge-V | | [link](python/llm/example/GPU/HuggingFace/Multimodal/glm-edge-v) | | 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/Multimodal/glm-edge-v/README.md b/python/llm/example/GPU/HuggingFace/Multimodal/glm-edge-v/README.md new file mode 100644 index 00000000..46962220 --- /dev/null +++ b/python/llm/example/GPU/HuggingFace/Multimodal/glm-edge-v/README.md @@ -0,0 +1,164 @@ +# GLM-Edge-V +In this directory, you will find examples on how you could apply IPEX-LLM INT4 optimizations on GLM-Edge-V models on [Intel GPUs](../../../README.md). For illustration purposes, we utilize the [THUDM/glm-edge-v-5b](https://huggingface.co/THUDM/glm-edge-v-5b) and [THUDM/glm-edge-v-2b](https://huggingface.co/THUDM/glm-edge-v-2b) (or [ZhipuAI/glm-edge-v-5b](https://www.modelscope.cn/models/ZhipuAI/glm-edge-v-5b) and [ZhipuAI/glm-edge-v-2b](https://www.modelscope.cn/models/ZhipuAI/glm-edge-v-2b) for ModelScope) as reference GLM-Edge-V 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. + +## Example: Predict Tokens using `generate()` API +In the example [generate.py](./generate.py), we show a basic use case for a GLM-Edge-V model to predict the next N tokens using `generate()` API, with IPEX-LLM FP8 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/ + +# install packages required for GLM-Edge-V +pip install transformers==4.47.0 +pip install accelerate==0.33.0 +pip install "trl<0.12.0" + +# [optional] only needed if you would like to use ModelScope as model hub +pip install modelscope==1.11.0 +``` + +#### 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-V +pip install transformers==4.47.0 +pip install accelerate==0.33.0 +pip install "trl<0.12.0" + +# [optional] only needed if you would like to use ModelScope as model hub +pip install modelscope==1.11.0 +``` + +### 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 + +```bash +# for Hugging Face model hub +python ./generate.py --repo-id-or-model-path REPO_ID_OR_MODEL_PATH --prompt PROMPT --n-predict N_PREDICT --image-url-or-path IMAGE_URL_OR_PATH + +# for ModelScope model hub +python ./generate.py --repo-id-or-model-path REPO_ID_OR_MODEL_PATH --prompt PROMPT --n-predict N_PREDICT --image-url-or-path IMAGE_URL_OR_PATH --modelscope +``` + +Arguments info: +- `--repo-id-or-model-path REPO_ID_OR_MODEL_PATH`: argument defining the **Hugging Face** (e.g. `THUDM/glm-edge-v-5b` or `THUDM/glm-edge-v-2b`) or **ModelScope** (e.g. `ZhipuAI/glm-edge-v-5b` or `ZhipuAI/glm-edge-v-2b`) repo id for the GLM-Edge-V model to be downloaded, or the path to the checkpoint folder. It is default to be `'THUDM/glm-edge-v-5b'` for **Hugging Face** or `'ZhipuAI/glm-edge-v-5b'` for **ModelScope**. +- `--image-url-or-path IMAGE_URL_OR_PATH`: argument defining the image to be infered. It is default to be `'http://farm6.staticflickr.com/5268/5602445367_3504763978_z.jpg'`. +- `--prompt PROMPT`: argument defining the prompt to be infered (with integrated prompt format for chat). It is default to be `'What is in the image?'`. +- `--n-predict N_PREDICT`: argument defining the max number of tokens to predict. It is default to be `32`. +- `--modelscope`: using **ModelScope** as model hub instead of **Hugging Face**. + +#### Sample Output +#### [THUDM/glm-edge-v-2b](https://huggingface.co/THUDM/glm-edge-v-2b) + +```log +Inference time: xxxx s +-------------------- Prompt -------------------- +图片里有什么？ +-------------------- Output -------------------- +这张图片中，一个穿着粉色和白色条纹衣服的小女孩举着一只穿着粉色裙子的白色小熊玩偶。背景是砖墙 +``` + +```log +Inference time: xxxx s +-------------------- Prompt -------------------- +What is in the image? +-------------------- Output -------------------- +A white teddy bear with pink accents is being held up by a person wearing a pink and white striped shirt. The bear is dressed in a +``` + +#### [THUDM/glm-edge-v-5b](https://huggingface.co/THUDM/glm-edge-v-5b) + +```log +Inference time: xxxx s +-------------------- Prompt -------------------- +图片里有什么？ +-------------------- Output -------------------- +这张图片展示了一个穿着红白条纹衣服的小女孩，她正在展示她手中的一个粉红色的白熊玩偶。背景是砖墙和一丛 +``` + +```log +Inference time: xxxx s +-------------------- Prompt -------------------- +What is in the image? +-------------------- Output -------------------- +The image shows a young girl holding a small, white teddy bear. The girl is wearing a pink striped dress with a red bow at the neckline +``` + +The sample input image is (which is fetched from [COCO dataset](https://cocodataset.org/#explore?id=264959)): + + diff --git a/python/llm/example/GPU/HuggingFace/Multimodal/glm-edge-v/generate.py b/python/llm/example/GPU/HuggingFace/Multimodal/glm-edge-v/generate.py new file mode 100644 index 00000000..ce96bd09 --- /dev/null +++ b/python/llm/example/GPU/HuggingFace/Multimodal/glm-edge-v/generate.py @@ -0,0 +1,122 @@ +# +# 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 os +import torch +import time +import argparse +import requests + +from ipex_llm.transformers import AutoModelForCausalLM + +from PIL import Image + +if __name__ == '__main__': + parser = argparse.ArgumentParser(description='Predict Tokens using `generate()` API for glm-edge-v model') + parser.add_argument('--repo-id-or-model-path', type=str, + help='The Hugging Face or ModelScope repo id for the glm-edge-v model to be downloaded' + ', or the path to the checkpoint folder') + parser.add_argument('--image-url-or-path', type=str, + default='http://farm6.staticflickr.com/5268/5602445367_3504763978_z.jpg', + help='The URL or path to the image to infer') + parser.add_argument('--prompt', type=str, default="What is in the image?", + help='Prompt to infer') + parser.add_argument('--n-predict', type=int, default=32, + help='Max tokens to predict') + parser.add_argument('--modelscope', action="store_true", default=False, + help="Use models from modelscope") + + args = parser.parse_args() + + if args.modelscope: + from modelscope import AutoTokenizer, AutoImageProcessor + model_hub = 'modelscope' + else: + from transformers import AutoTokenizer, AutoImageProcessor + model_hub = 'huggingface' + + model_path = args.repo_id_or_model_path if args.repo_id_or_model_path else \ + ("ZhipuAI/glm-edge-v-5b" if args.modelscope else "THUDM/glm-edge-v-5b") + image_path = args.image_url_or_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, + modules_to_not_convert=["vision"], + use_cache=True, + model_hub=model_hub) + model = model.half().to('xpu') + + tokenizer = AutoTokenizer.from_pretrained(model_path, trust_remote_code=True) + + query = args.prompt + if os.path.exists(image_path): + image = Image.open(image_path) + else: + image = Image.open(requests.get(image_path, stream=True).raw) + + image_processor = AutoImageProcessor.from_pretrained(model_path, trust_remote_code=True) + + with torch.inference_mode(): + pixel_values = image_processor(images=[image], return_tensors='pt').pixel_values + pixel_values = pixel_values.to('xpu') + + # The following code for generation is adapted from https://huggingface.co/THUDM/glm-edge-v-5b#inference + messages = [{ + "role": "user", + "content": [{"type": "image"}, + {"type": "text", + "text": args.prompt}] + }] + + inputs = tokenizer.apply_chat_template( + messages, + add_generation_prompt=True, + return_dict=True, + tokenize=True, + return_tensors="pt" + ) + inputs = inputs.to('xpu') + + generate_kwargs = { + **inputs, + "pixel_values": pixel_values, + "max_new_tokens": args.n_predict, + } + + # 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][len(inputs["input_ids"][0]):], + 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)