From 8153c3008e40c11b7b6d2d15941df6232aee4137 Mon Sep 17 00:00:00 2001 From: Yang Wang Date: Thu, 18 Apr 2024 11:01:33 -0700 Subject: [PATCH] Initial llama3 example (#10799) * Add initial hf huggingface GPU example * Small fix * Add llama3 gpu pytorch model example * Add llama 3 hf transformers CPU example * Add llama 3 pytorch model CPU example * Fixes * Small fix * Small fixes * Small fix * Small fix * Add links * update repo id * change prompt tuning url * remove system header if there is no system prompt --------- Co-authored-by: Yuwen Hu Co-authored-by: Yuwen Hu <54161268+Oscilloscope98@users.noreply.github.com> --- README.md | 1 + docs/readthedocs/source/index.rst | 7 + .../Model/llama3/README.md | 68 +++++++++ .../Model/llama3/generate.py | 81 +++++++++++ .../CPU/PyTorch-Models/Model/llama3/README.md | 68 +++++++++ .../PyTorch-Models/Model/llama3/generate.py | 82 +++++++++++ .../Model/llama3/README.md | 136 +++++++++++++++++ .../Model/llama3/generate.py | 90 ++++++++++++ .../GPU/PyTorch-Models/Model/llama3/README.md | 137 ++++++++++++++++++ .../PyTorch-Models/Model/llama3/generate.py | 93 ++++++++++++ 10 files changed, 763 insertions(+) create mode 100644 python/llm/example/CPU/HF-Transformers-AutoModels/Model/llama3/README.md create mode 100644 python/llm/example/CPU/HF-Transformers-AutoModels/Model/llama3/generate.py create mode 100644 python/llm/example/CPU/PyTorch-Models/Model/llama3/README.md create mode 100644 python/llm/example/CPU/PyTorch-Models/Model/llama3/generate.py create mode 100644 python/llm/example/GPU/HF-Transformers-AutoModels/Model/llama3/README.md create mode 100644 python/llm/example/GPU/HF-Transformers-AutoModels/Model/llama3/generate.py create mode 100644 python/llm/example/GPU/PyTorch-Models/Model/llama3/README.md create mode 100644 python/llm/example/GPU/PyTorch-Models/Model/llama3/generate.py diff --git a/README.md b/README.md index 35ccdea6..b2b50e6e 100644 --- a/README.md +++ b/README.md @@ -128,6 +128,7 @@ Over 50 models have been optimized/verified on `ipex-llm`, including *LLaMA/LLaM |------------|----------------------------------------------------------------|-----------------------------------------------------------------| | LLaMA *(such as Vicuna, Guanaco, Koala, Baize, WizardLM, etc.)* | [link1](python/llm/example/CPU/Native-Models), [link2](python/llm/example/CPU/HF-Transformers-AutoModels/Model/vicuna) |[link](python/llm/example/GPU/HF-Transformers-AutoModels/Model/vicuna)| | LLaMA 2 | [link1](python/llm/example/CPU/Native-Models), [link2](python/llm/example/CPU/HF-Transformers-AutoModels/Model/llama2) | [link](python/llm/example/GPU/HF-Transformers-AutoModels/Model/llama2) | +| LLaMA 3 | [link](python/llm/example/CPU/HF-Transformers-AutoModels/Model/llama3) | [link](python/llm/example/GPU/HF-Transformers-AutoModels/Model/llama3) | | ChatGLM | [link](python/llm/example/CPU/HF-Transformers-AutoModels/Model/chatglm) | | | ChatGLM2 | [link](python/llm/example/CPU/HF-Transformers-AutoModels/Model/chatglm2) | [link](python/llm/example/GPU/HF-Transformers-AutoModels/Model/chatglm2) | | ChatGLM3 | [link](python/llm/example/CPU/HF-Transformers-AutoModels/Model/chatglm3) | [link](python/llm/example/GPU/HF-Transformers-AutoModels/Model/chatglm3) | diff --git a/docs/readthedocs/source/index.rst b/docs/readthedocs/source/index.rst index efe4728f..27d80445 100644 --- a/docs/readthedocs/source/index.rst +++ b/docs/readthedocs/source/index.rst @@ -219,6 +219,13 @@ Verified Models link link + + LLaMA 3 + + link + + link + ChatGLM diff --git a/python/llm/example/CPU/HF-Transformers-AutoModels/Model/llama3/README.md b/python/llm/example/CPU/HF-Transformers-AutoModels/Model/llama3/README.md new file mode 100644 index 00000000..50502469 --- /dev/null +++ b/python/llm/example/CPU/HF-Transformers-AutoModels/Model/llama3/README.md @@ -0,0 +1,68 @@ +# Llama3 +In this directory, you will find examples on how you could apply IPEX-LLM INT4 optimizations on Llama3 models. For illustration purposes, we utilize the [meta-llama/Meta-Llama-3-8B-Instruct](https://huggingface.co/meta-llama/Meta-Llama-3-8B-Instruct) as a reference Llama3 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 Llama3 model to predict the next N tokens using `generate()` API, with IPEX-LLM INT4 optimizations. +### 1. Install +We suggest using conda to manage environment: +```bash +conda create -n llm python=3.11 +conda activate llm + +pip install --pre --upgrade ipex-llm[all] # install ipex-llm with 'all' option + +# transformers>=4.33.0 is required for Llama3 with IPEX-LLM optimizations +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 Llama3 model (e.g. `meta-llama/Meta-Llama-3-8B-Instruct`) to be downloaded, or the path to the huggingface checkpoint folder. It is default to be `'meta-llama/Meta-Llama-3-8B-Instruct'`. +- `--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`. + +> **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 Llama3 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: +```powershell +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 +#### [meta-llama/Meta-Llama-3-8B-Instruct](https://huggingface.co/meta-llama/Meta-Llama-3-8B-Instruct) +```log +Inference time: xxxx s +-------------------- Prompt -------------------- +<|begin_of_text|><|start_header_id|>user<|end_header_id|> +What is AI?<|eot_id|><|start_header_id|>assistant<|end_header_id|> + +-------------------- Output (skip_special_tokens=False) -------------------- +<|begin_of_text|><|start_header_id|>user<|end_header_id|> +What is AI?<|eot_id|><|start_header_id|>assistant<|end_header_id|> +A question that gets to the heart of the 21st century! + +Artificial Intelligence (AI) refers to the development of computer systems that can perform tasks that +``` diff --git a/python/llm/example/CPU/HF-Transformers-AutoModels/Model/llama3/generate.py b/python/llm/example/CPU/HF-Transformers-AutoModels/Model/llama3/generate.py new file mode 100644 index 00000000..604bfe17 --- /dev/null +++ b/python/llm/example/CPU/HF-Transformers-AutoModels/Model/llama3/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 + +# you could tune the prompt based on your own model, +# here the prompt tuning refers to https://llama.meta.com/docs/model-cards-and-prompt-formats/meta-llama-3 +DEFAULT_SYSTEM_PROMPT = """\ +""" + +def get_prompt(user_input: str, chat_history: list[tuple[str, str]], + system_prompt: str) -> str: + prompt_texts = [f'<|begin_of_text|>'] + + if system_prompt != '': + prompt_texts.append(f'<|start_header_id|>system<|end_header_id|>\n{system_prompt}<|eot_id|>') + + for history_input, history_response in chat_history: + prompt_texts.append(f'<|start_header_id|>user<|end_header_id|>\n{history_input.strip()}<|eot_id|>') + prompt_texts.append(f'<|start_header_id|>assistant<|end_header_id|>\n{history_response.strip()}<|eot_id|>') + + prompt_texts.append(f'<|start_header_id|>user<|end_header_id|>\n{user_input.strip()}<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n') + return ''.join(prompt_texts) + +if __name__ == '__main__': + parser = argparse.ArgumentParser(description='Predict Tokens using `generate()` API for Llama3 model') + parser.add_argument('--repo-id-or-model-path', type=str, default="meta-llama/Meta-Llama-3-8B-Instruct", + help='The huggingface repo id for the Llama3 (e.g. `meta-llama/Meta-Llama-3-8B-Instruct`) 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 + 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) + + # Generate predicted tokens + with torch.inference_mode(): + prompt = get_prompt(args.prompt, [], system_prompt=DEFAULT_SYSTEM_PROMPT) + input_ids = tokenizer.encode(prompt, return_tensors="pt") + st = time.time() + output = model.generate(input_ids, + max_new_tokens=args.n_predict) + end = time.time() + output_str = tokenizer.decode(output[0], skip_special_tokens=False) + print(f'Inference time: {end-st} s') + print('-'*20, 'Prompt', '-'*20) + print(prompt) + print('-'*20, 'Output (skip_special_tokens=False)', '-'*20) + print(output_str) + diff --git a/python/llm/example/CPU/PyTorch-Models/Model/llama3/README.md b/python/llm/example/CPU/PyTorch-Models/Model/llama3/README.md new file mode 100644 index 00000000..83f214e3 --- /dev/null +++ b/python/llm/example/CPU/PyTorch-Models/Model/llama3/README.md @@ -0,0 +1,68 @@ +# Llama3 +In this directory, you will find examples on how you could use IPEX-LLM `optimize_model` API to accelerate Llama3 models. For illustration purposes, we utilize the [meta-llama/Meta-Llama-3-8B-Instruct](https://huggingface.co/meta-llama/Meta-Llama-3-8B-Instruct) as a reference Llama3 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 Llama3 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://docs.conda.io/en/latest/miniconda.html#). + +After installing conda, create a Python environment for IPEX-LLM: +```bash +conda create -n llm python=3.11 # recommend to use Python 3.11 +conda activate llm + +pip install --pre --upgrade ipex-llm[all] # install the latest ipex-llm nightly build with 'all' option + +# transformers>=4.33.0 is required for Llama3 with IPEX-LLM optimizations +pip install transformers==4.37.0 +``` + +### 2. Run +After setting up the Python environment, you could run the example by following steps. + +#### 2.1 Client +On client Windows machines, it is recommended to run directly with full utilization of all cores: +```powershell +python ./generate.py --prompt 'What is AI?' +``` +More information about arguments can be found in [Arguments Info](#23-arguments-info) section. The expected output can be found in [Sample Output](#24-sample-output) section. + +#### 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?' +``` +More information about arguments can be found in [Arguments Info](#23-arguments-info) section. The expected output can be found in [Sample Output](#24-sample-output) section. + +#### 2.3 Arguments Info +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 Llama3 model (e.g. `meta-llama/Meta-Llama-3-8B-Instruct`) to be downloaded, or the path to the huggingface checkpoint folder. It is default to be `'meta-llama/Meta-Llama-3-8B-Instruct'`. +- `--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`. + +#### 2.3 Sample Output +#### [meta-llama/Meta-Llama-3-8B-Instruct](https://huggingface.co/meta-llama/Meta-Llama-3-8B-Instruct) +```log +Inference time: xxxx s +-------------------- Prompt -------------------- +<|begin_of_text|><|start_header_id|>user<|end_header_id|> +What is AI?<|eot_id|><|start_header_id|>assistant<|end_header_id|> + +-------------------- Output (skip_special_tokens=False) -------------------- +<|begin_of_text|><|start_header_id|>user<|end_header_id|> +What is AI?<|eot_id|><|start_header_id|>assistant<|end_header_id|> +A question that gets to the heart of the 21st century! + +Artificial Intelligence (AI) refers to the development of computer systems that can perform tasks that +``` diff --git a/python/llm/example/CPU/PyTorch-Models/Model/llama3/generate.py b/python/llm/example/CPU/PyTorch-Models/Model/llama3/generate.py new file mode 100644 index 00000000..9e82eb81 --- /dev/null +++ b/python/llm/example/CPU/PyTorch-Models/Model/llama3/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 + +from transformers import AutoModelForCausalLM, AutoTokenizer +from ipex_llm import optimize_model + +# you could tune the prompt based on your own model, +# here the prompt tuning refers to https://llama.meta.com/docs/model-cards-and-prompt-formats/meta-llama-3 +DEFAULT_SYSTEM_PROMPT = """\ +""" + +def get_prompt(user_input: str, chat_history: list[tuple[str, str]], + system_prompt: str) -> str: + prompt_texts = [f'<|begin_of_text|>'] + + if system_prompt != '': + prompt_texts.append(f'<|start_header_id|>system<|end_header_id|>\n{system_prompt}<|eot_id|>') + + for history_input, history_response in chat_history: + prompt_texts.append(f'<|start_header_id|>user<|end_header_id|>\n{history_input.strip()}<|eot_id|>') + prompt_texts.append(f'<|start_header_id|>assistant<|end_header_id|>\n{history_response.strip()}<|eot_id|>') + + prompt_texts.append(f'<|start_header_id|>user<|end_header_id|>\n{user_input.strip()}<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n') + return ''.join(prompt_texts) + +if __name__ == '__main__': + parser = argparse.ArgumentParser(description='Predict Tokens using `generate()` API for Llama3 model') + parser.add_argument('--repo-id-or-model-path', type=str, default="meta-llama/Meta-Llama-3-8B-Instruct", + help='The huggingface repo id for the Llama3 (e.g. `meta-llama/Meta-Llama-3-8B-Instruct`) 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 + model = AutoModelForCausalLM.from_pretrained(model_path, + trust_remote_code=True, + torch_dtype='auto', + low_cpu_mem_usage=True, + use_cache=True) + + # With only one line to enable IPEX-LLM optimization on model + model = optimize_model(model) + + # Load tokenizer + tokenizer = AutoTokenizer.from_pretrained(model_path, trust_remote_code=True) + + # Generate predicted tokens + with torch.inference_mode(): + prompt = get_prompt(args.prompt, [], system_prompt=DEFAULT_SYSTEM_PROMPT) + input_ids = tokenizer.encode(prompt, return_tensors="pt") + st = time.time() + output = model.generate(input_ids, + max_new_tokens=args.n_predict) + end = time.time() + output_str = tokenizer.decode(output[0], skip_special_tokens=False) + print(f'Inference time: {end-st} s') + print('-'*20, 'Prompt', '-'*20) + print(prompt) + print('-'*20, 'Output (skip_special_tokens=False)', '-'*20) + print(output_str) diff --git a/python/llm/example/GPU/HF-Transformers-AutoModels/Model/llama3/README.md b/python/llm/example/GPU/HF-Transformers-AutoModels/Model/llama3/README.md new file mode 100644 index 00000000..89697ac0 --- /dev/null +++ b/python/llm/example/GPU/HF-Transformers-AutoModels/Model/llama3/README.md @@ -0,0 +1,136 @@ +# Llama3 +In this directory, you will find examples on how you could apply IPEX-LLM INT4 optimizations on Llama3 models on [Intel GPUs](../../../README.md). For illustration purposes, we utilize the [meta-llama/Meta-Llama-3-8B-Instruct](https://huggingface.co/meta-llama/Meta-Llama-3-8B-Instruct) as a reference Llama3 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 Llama3 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/ + +# transformers>=4.33.0 is required for Llama3 with IPEX-LLM optimizations +pip install transformers==4.37.0 +``` + +#### 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 use pip to install the Intel oneAPI Base Toolkit 2024.0 +pip install dpcpp-cpp-rt==2024.0.2 mkl-dpcpp==2024.0.0 onednn==2024.0.0 + +# 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/ + +# transformers>=4.33.0 is required for Llama3 with IPEX-LLM optimizations +pip install transformers==4.37.0 +``` + +### 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 +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 Llama3 model (e.g. `meta-llama/Meta-Llama-3-8B-Instruct`) to be downloaded, or the path to the huggingface checkpoint folder. It is default to be `'meta-llama/Meta-Llama-3-8B-Instruct'`. +- `--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 +#### [meta-llama/Meta-Llama-3-8B-Instruct](https://huggingface.co/meta-llama/Meta-Llama-3-8B-Instruct) +```log +Inference time: xxxx s +-------------------- Prompt -------------------- +<|begin_of_text|><|start_header_id|>user<|end_header_id|> +What is AI?<|eot_id|><|start_header_id|>assistant<|end_header_id|> + +-------------------- Output (skip_special_tokens=False) -------------------- +<|begin_of_text|><|start_header_id|>user<|end_header_id|> +What is AI?<|eot_id|><|start_header_id|>assistant<|end_header_id|> +A question that gets to the heart of the 21st century! + +Artificial Intelligence (AI) refers to the development of computer systems that can perform tasks that +``` diff --git a/python/llm/example/GPU/HF-Transformers-AutoModels/Model/llama3/generate.py b/python/llm/example/GPU/HF-Transformers-AutoModels/Model/llama3/generate.py new file mode 100644 index 00000000..b58cb7a6 --- /dev/null +++ b/python/llm/example/GPU/HF-Transformers-AutoModels/Model/llama3/generate.py @@ -0,0 +1,90 @@ +# +# 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 + +# you could tune the prompt based on your own model, +# here the prompt tuning refers to https://llama.meta.com/docs/model-cards-and-prompt-formats/meta-llama-3 +DEFAULT_SYSTEM_PROMPT = """\ +""" + +def get_prompt(user_input: str, chat_history: list[tuple[str, str]], + system_prompt: str) -> str: + prompt_texts = [f'<|begin_of_text|>'] + + if system_prompt != '': + prompt_texts.append(f'<|start_header_id|>system<|end_header_id|>\n{system_prompt}<|eot_id|>') + + for history_input, history_response in chat_history: + prompt_texts.append(f'<|start_header_id|>user<|end_header_id|>\n{history_input.strip()}<|eot_id|>') + prompt_texts.append(f'<|start_header_id|>assistant<|end_header_id|>\n{history_response.strip()}<|eot_id|>') + + prompt_texts.append(f'<|start_header_id|>user<|end_header_id|>\n{user_input.strip()}<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n') + return ''.join(prompt_texts) + +if __name__ == '__main__': + parser = argparse.ArgumentParser(description='Predict Tokens using `generate()` API for Llama3 model') + parser.add_argument('--repo-id-or-model-path', type=str, default="meta-llama/Meta-Llama-3-8B-Instruct", + help='The huggingface repo id for the Llama3 (e.g. `meta-llama/Meta-Llama-3-8B-Instruct`) 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, + 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(): + prompt = get_prompt(args.prompt, [], system_prompt=DEFAULT_SYSTEM_PROMPT) + 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() + 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=False) + print(f'Inference time: {end-st} s') + print('-'*20, 'Prompt', '-'*20) + print(prompt) + print('-'*20, 'Output (skip_special_tokens=False)', '-'*20) + print(output_str) diff --git a/python/llm/example/GPU/PyTorch-Models/Model/llama3/README.md b/python/llm/example/GPU/PyTorch-Models/Model/llama3/README.md new file mode 100644 index 00000000..6dc5e082 --- /dev/null +++ b/python/llm/example/GPU/PyTorch-Models/Model/llama3/README.md @@ -0,0 +1,137 @@ +# Llama3 +In this directory, you will find examples on how you could use IPEX-LLM `optimize_model` API to accelerate Llama3 models. For illustration purposes, we utilize the [meta-llama/Meta-Llama-3-8B-Instruct](https://huggingface.co/meta-llama/Meta-Llama-3-8B-Instruct) as a reference Llama3 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. + +## Example: Predict Tokens using `generate()` API +In the example [generate.py](./generate.py), we show a basic use case for a Llama3 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/ + +# transformers>=4.33.0 is required for Llama3 with IPEX-LLM optimizations +pip install transformers==4.37.0 +``` + +#### 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 use pip to install the Intel oneAPI Base Toolkit 2024.0 +pip install dpcpp-cpp-rt==2024.0.2 mkl-dpcpp==2024.0.0 onednn==2024.0.0 + +# 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/ + +# transformers>=4.33.0 is required for Llama3 with IPEX-LLM optimizations +pip install transformers==4.37.0 +``` + +### 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 +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 Llama3 model (e.g. `meta-llama/Meta-Llama-3-8B-Instruct`) to be downloaded, or the path to the huggingface checkpoint folder. It is default to be `'meta-llama/Meta-Llama-3-8B-Instruct'`. +- `--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 +#### [meta-llama/Meta-Llama-3-8B-Instruct](https://huggingface.co/meta-llama/Meta-Llama-3-8B-Instruct) +```log +Inference time: xxxx s +-------------------- Prompt -------------------- +<|begin_of_text|><|start_header_id|>user<|end_header_id|> +What is AI?<|eot_id|><|start_header_id|>assistant<|end_header_id|> + +-------------------- Output (skip_special_tokens=False) -------------------- +<|begin_of_text|><|start_header_id|>user<|end_header_id|> +What is AI?<|eot_id|><|start_header_id|>assistant<|end_header_id|> +A question that gets to the heart of the 21st century! + +Artificial Intelligence (AI) refers to the development of computer systems that can perform tasks that +``` diff --git a/python/llm/example/GPU/PyTorch-Models/Model/llama3/generate.py b/python/llm/example/GPU/PyTorch-Models/Model/llama3/generate.py new file mode 100644 index 00000000..0b533ff8 --- /dev/null +++ b/python/llm/example/GPU/PyTorch-Models/Model/llama3/generate.py @@ -0,0 +1,93 @@ +# +# 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 AutoModelForCausalLM, AutoTokenizer +from ipex_llm import optimize_model + +# you could tune the prompt based on your own model, +# here the prompt tuning refers to https://llama.meta.com/docs/model-cards-and-prompt-formats/meta-llama-3 +DEFAULT_SYSTEM_PROMPT = """\ +""" + +def get_prompt(user_input: str, chat_history: list[tuple[str, str]], + system_prompt: str) -> str: + prompt_texts = [f'<|begin_of_text|>'] + + if system_prompt != '': + prompt_texts.append(f'<|start_header_id|>system<|end_header_id|>\n{system_prompt}<|eot_id|>') + + for history_input, history_response in chat_history: + prompt_texts.append(f'<|start_header_id|>user<|end_header_id|>\n{history_input.strip()}<|eot_id|>') + prompt_texts.append(f'<|start_header_id|>assistant<|end_header_id|>\n{history_response.strip()}<|eot_id|>') + + prompt_texts.append(f'<|start_header_id|>user<|end_header_id|>\n{user_input.strip()}<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n') + return ''.join(prompt_texts) + +if __name__ == '__main__': + parser = argparse.ArgumentParser(description='Predict Tokens using `generate()` API for Llama3 model') + parser.add_argument('--repo-id-or-model-path', type=str, default="meta-llama/Meta-Llama-3-8B-Instruct", + help='The huggingface repo id for the Llama3 (e.g. `meta-llama/Meta-Llama-3-8B-Instruct`) 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 + model = AutoModelForCausalLM.from_pretrained(model_path, + trust_remote_code=True, + torch_dtype='auto', + low_cpu_mem_usage=True, + use_cache=True) + + # With only one line to enable IPEX-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.half().to('xpu') + + # Load tokenizer + tokenizer = AutoTokenizer.from_pretrained(model_path, trust_remote_code=True) + + # Generate predicted tokens + with torch.inference_mode(): + prompt = get_prompt(args.prompt, [], system_prompt=DEFAULT_SYSTEM_PROMPT) + 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() + 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=False) + print(f'Inference time: {end-st} s') + print('-'*20, 'Prompt', '-'*20) + print(prompt) + print('-'*20, 'Output (skip_special_tokens=False)', '-'*20) + print(output_str)