diff --git a/README.md b/README.md index 9542fb78..555180c4 100644 --- a/README.md +++ b/README.md @@ -301,6 +301,7 @@ Over 50 models have been optimized/verified on `ipex-llm`, including *LLaMA/LLaM | Phi-3-vision | [link](python/llm/example/CPU/HF-Transformers-AutoModels/Model/phi-3-vision) | [link](python/llm/example/GPU/HuggingFace/Multimodal/phi-3-vision) | | Yuan2 | [link](python/llm/example/CPU/HF-Transformers-AutoModels/Model/yuan2) | [link](python/llm/example/GPU/HuggingFace/LLM/yuan2) | | Gemma | [link](python/llm/example/CPU/HF-Transformers-AutoModels/Model/gemma) | [link](python/llm/example/GPU/HuggingFace/LLM/gemma) | +| Gemma2 | | [link](python/llm/example/GPU/HuggingFace/LLM/gemma2) | | DeciLM-7B | [link](python/llm/example/CPU/HF-Transformers-AutoModels/Model/deciLM-7b) | [link](python/llm/example/GPU/HuggingFace/LLM/deciLM-7b) | | Deepseek | [link](python/llm/example/CPU/HF-Transformers-AutoModels/Model/deepseek) | [link](python/llm/example/GPU/HuggingFace/LLM/deepseek) | | StableLM | [link](python/llm/example/CPU/HF-Transformers-AutoModels/Model/stablelm) | [link](python/llm/example/GPU/HuggingFace/LLM/stablelm) | diff --git a/python/llm/example/GPU/HuggingFace/LLM/gemma2/README.md b/python/llm/example/GPU/HuggingFace/LLM/gemma2/README.md new file mode 100644 index 00000000..c9352357 --- /dev/null +++ b/python/llm/example/GPU/HuggingFace/LLM/gemma2/README.md @@ -0,0 +1,144 @@ +# Gemma2 +In this directory, you will find examples on how you could apply IPEX-LLM INT4 optimizations on Google Gemma2 models on [Intel GPUs](../../../README.md). For illustration purposes, we utilize the [google/gemma-2-9b-it](https://huggingface.co/google/gemma-2-9b-it) and [google/gemma-2-2b-it](https://huggingface.co/google/gemma-2-2b-it) as reference Gemma2 models. + +## 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. + +**Important: According to Gemma2's requirement, please make sure you have installed `transformers==4.43.1` and `trl` to run the example.** + +## Example: Predict Tokens using `generate()` API +In the example [generate.py](./generate.py), we show a basic use case for a Gemma2 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/ + +# According to Gemma2's requirement, please make sure you are using a stable version of Transformers, 4.43.1 or newer. +pip install "transformers>=4.43.1" +pip install trl +``` + +#### 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/ + +# According to Gemma2's requirement, please make sure you are using a stable version of Transformers, 4.43.1 or newer. +pip install "transformers>=4.43.1" +pip install trl +``` + +### 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 + +```bash +python ./generate.py --prompt 'What is AI?' +``` + +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 Gemma model (e.g. `google/gemma-2-9b-it` and `google/gemma-2-2b-it`) to be downloaded, or the path to the huggingface checkpoint folder. It is default to be `'google/gemma-2-9b-it'`. +- `--prompt PROMPT`: argument defining the prompt to be infered (with integrated prompt format for chat). It is default to be `'What is AI?'`. +- `--n-predict N_PREDICT`: argument defining the max number of tokens to predict. It is default to be `32`. + +##### Sample Output +##### [google/gemma-2-9b-it](https://huggingface.co/google/gemma-2-9b-it) +```log +Inference time: xxxx s +-------------------- Output -------------------- +user +What is AI? +model +Artificial intelligence (AI) is a broad field of computer science focused on creating intelligent agents, which are systems that can reason, learn, and act autonomously. +``` + +##### [google/gemma-2-2b-it](https://huggingface.co/google/gemma-2-2b-it) +```log +Inference time: xxxx s +-------------------- Output -------------------- +user +What is AI? +model +AI, or Artificial Intelligence, is a broad field of computer science focused on creating intelligent agents, which are systems that can reason, learn, and act like humans +``` diff --git a/python/llm/example/GPU/HuggingFace/LLM/gemma2/generate.py b/python/llm/example/GPU/HuggingFace/LLM/gemma2/generate.py new file mode 100644 index 00000000..2e24a2b4 --- /dev/null +++ b/python/llm/example/GPU/HuggingFace/LLM/gemma2/generate.py @@ -0,0 +1,81 @@ +# +# 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 + +# The instruction-tuned models use a chat template that must be adhered to for conversational use. +# see https://huggingface.co/google/gemma-2b-it#chat-template. +chat = [ + { "role": "user", "content": "What is AI?" }, +] + +if __name__ == '__main__': + parser = argparse.ArgumentParser(description='Predict Tokens using `generate()` API for Gemma model') + parser.add_argument('--repo-id-or-model-path', type=str, default="google/gemma-2-9b-it", + help='The huggingface repo id for the Gemma2 (e.g. `google/gemma-2-9b-it` and `google/gemma-2-2b-it`) to be downloaded' + ', or the path to the huggingface checkpoint folder') + parser.add_argument('--prompt', type=str, default="What is AI?", + 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, + mixed_precision=True, + use_cache=True) + model = model.to('xpu') + + # Load tokenizer + tokenizer = AutoTokenizer.from_pretrained(model_path, trust_remote_code=True) + + # Generate predicted tokens + with torch.inference_mode(): + chat[0]['content'] = args.prompt + prompt = tokenizer.apply_chat_template(chat, tokenize=False, add_generation_prompt=True) + input_ids = tokenizer.encode(prompt, return_tensors="pt").to('xpu') + # ipex_llm model needs a warmup, then inference time can be accurate + output = model.generate(input_ids, + max_new_tokens=args.n_predict) + + # start inference + 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 IPEX-LLM INT4 optimizations + output = model.generate(input_ids, + max_new_tokens=args.n_predict) + torch.xpu.synchronize() + end = time.time() + output = output.cpu() + output_str = tokenizer.decode(output[0], skip_special_tokens=True) + print(f'Inference time: {end-st} s') + print('-'*20, 'Output', '-'*20) + print(output_str)