From bfa13671494da4b738458333f07e8228ef603c13 Mon Sep 17 00:00:00 2001 From: Zijie Li Date: Wed, 5 Jun 2024 18:09:53 +0800 Subject: [PATCH] Add CPU and GPU example for MiniCPM (#11202) * Change installation address Change former address: "https://docs.conda.io/en/latest/miniconda.html#" to new address: "https://conda-forge.org/download/" for 63 occurrences under python\llm\example * Change Prompt Change "Anaconda Prompt" to "Miniforge Prompt" for 1 occurrence * Create and update model minicpm * Update model minicpm Update model minicpm under GPU/PyTorch-Models * Update readme and generate.py change "prompt = tokenizer.apply_chat_template(chat, tokenize=False, add_generation_prompt=False)" and delete "pip install transformers==4.37.0 " * Update comments for minicpm GPU Update comments for generate.py at minicpm GPU * Add CPU example for MiniCPM * Update minicpm README for CPU * Update README for MiniCPM and Llama3 * Update Readme for Llama3 CPU Pytorch * Update and fix comments for MiniCPM --- README.md | 1 + docs/readthedocs/source/index.rst | 7 + .../Model/minicpm/README.md | 71 ++++++++++ .../Model/minicpm/generate.py | 72 ++++++++++ .../Model/phi-3/README.md | 1 + .../CPU/PyTorch-Models/Model/llama3/README.md | 4 +- .../PyTorch-Models/Model/minicpm/README.md | 74 +++++++++++ .../PyTorch-Models/Model/minicpm/generate.py | 74 +++++++++++ .../CPU/PyTorch-Models/Model/phi-3/README.md | 1 + .../Model/minicpm/README.md | 123 ++++++++++++++++++ .../Model/minicpm/generate.py | 80 ++++++++++++ .../PyTorch-Models/Model/minicpm/README.md | 123 ++++++++++++++++++ .../PyTorch-Models/Model/minicpm/generate.py | 81 ++++++++++++ 13 files changed, 710 insertions(+), 2 deletions(-) create mode 100644 python/llm/example/CPU/HF-Transformers-AutoModels/Model/minicpm/README.md create mode 100644 python/llm/example/CPU/HF-Transformers-AutoModels/Model/minicpm/generate.py create mode 100644 python/llm/example/CPU/PyTorch-Models/Model/minicpm/README.md create mode 100644 python/llm/example/CPU/PyTorch-Models/Model/minicpm/generate.py create mode 100644 python/llm/example/GPU/HF-Transformers-AutoModels/Model/minicpm/README.md create mode 100644 python/llm/example/GPU/HF-Transformers-AutoModels/Model/minicpm/generate.py create mode 100644 python/llm/example/GPU/PyTorch-Models/Model/minicpm/README.md create mode 100644 python/llm/example/GPU/PyTorch-Models/Model/minicpm/generate.py diff --git a/README.md b/README.md index 8901f4bb636..c996d430de2 100644 --- a/README.md +++ b/README.md @@ -207,6 +207,7 @@ Over 50 models have been optimized/verified on `ipex-llm`, including *LLaMA/LLaM | CodeGemma | [link](python/llm/example/CPU/HF-Transformers-AutoModels/Model/codegemma) | [link](python/llm/example/GPU/HF-Transformers-AutoModels/Model/codegemma) | | Command-R/cohere | [link](python/llm/example/CPU/HF-Transformers-AutoModels/Model/cohere) | [link](python/llm/example/GPU/HF-Transformers-AutoModels/Model/cohere) | | CodeGeeX2 | [link](python/llm/example/CPU/HF-Transformers-AutoModels/Model/codegeex2) | [link](python/llm/example/GPU/HF-Transformers-AutoModels/Model/codegeex2) | +| MiniCPM | [link](python/llm/example/CPU/HF-Transformers-AutoModels/Model/minicpm) | [link](python/llm/example/GPU/HF-Transformers-AutoModels/Model/minicpm) | ## Get Support - Please report a bug or raise a feature request by opening a [Github Issue](https://github.com/intel-analytics/ipex-llm/issues) diff --git a/docs/readthedocs/source/index.rst b/docs/readthedocs/source/index.rst index 1071b5692ab..7c4d77cb2ac 100644 --- a/docs/readthedocs/source/index.rst +++ b/docs/readthedocs/source/index.rst @@ -618,6 +618,13 @@ Verified Models link + + MiniCPM + + link + + link + diff --git a/python/llm/example/CPU/HF-Transformers-AutoModels/Model/minicpm/README.md b/python/llm/example/CPU/HF-Transformers-AutoModels/Model/minicpm/README.md new file mode 100644 index 00000000000..34a4aced900 --- /dev/null +++ b/python/llm/example/CPU/HF-Transformers-AutoModels/Model/minicpm/README.md @@ -0,0 +1,71 @@ +# MiniCPM +In this directory, you will find examples on how you could apply IPEX-LLM INT4 optimizations on MiniCPM models. For illustration purposes, we utilize the [openbmb/MiniCPM-2B-sft-bf16](https://huggingface.co/openbmb/MiniCPM-2B-sft-bf16) as a reference MiniCPM 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 MiniCPM model to predict the next N tokens using `generate()` API, with IPEX-LLM INT4 optimizations. +### 1. Install +We suggest using conda to manage environment: + +On Linux: + +```bash +conda create -n llm python=3.11 +conda activate llm + +# install ipex-llm with 'all' option +pip install --pre --upgrade ipex-llm[all] --extra-index-url https://download.pytorch.org/whl/cpu +``` +On Windows: + +```cmd +conda create -n llm python=3.11 +conda activate llm + +pip install --pre --upgrade ipex-llm[all] +``` + +### 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 MiniCPM model (e.g. `openbmb/MiniCPM-2B-sft-bf16`) to be downloaded, or the path to the huggingface checkpoint folder. It is default to be `'openbmb/MiniCPM-2B-sft-bf16'`. +- `--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 MiniCPM 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: +```cmd +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 +#### [openbmb/MiniCPM-2B-sft-bf16](https://huggingface.co/openbmb/MiniCPM-2B-sft-bf16) +```log +Inference time: xxxx s +-------------------- Prompt -------------------- +<用户>what is AI? +-------------------- Output -------------------- + <用户>what is AI? AI, or Artificial Intelligence, refers to the simulation of human intelligence in machines that are programmed to think and learn like humans. It is a broad field of computer +``` \ No newline at end of file diff --git a/python/llm/example/CPU/HF-Transformers-AutoModels/Model/minicpm/generate.py b/python/llm/example/CPU/HF-Transformers-AutoModels/Model/minicpm/generate.py new file mode 100644 index 00000000000..8bdb2fcb09c --- /dev/null +++ b/python/llm/example/CPU/HF-Transformers-AutoModels/Model/minicpm/generate.py @@ -0,0 +1,72 @@ +# +# 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 MiniCPM model') + parser.add_argument('--repo-id-or-model-path', type=str, default="openbmb/MiniCPM-2B-sft-bf16", + help='The huggingface repo id for the MiniCPM model 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(): + + # here the prompt formatting refers to: https://huggingface.co/openbmb/MiniCPM-2B-sft-bf16/blob/79fbb1db171e6d8bf77cdb0a94076a43003abd9e/modeling_minicpm.py#L1320 + chat = [ + { "role": "user", "content": args.prompt }, + ] + prompt = tokenizer.apply_chat_template(chat, tokenize=False, add_generation_prompt=False) + input_ids = tokenizer.encode(prompt, return_tensors="pt") + + # start inference + st = time.time() + + output = model.generate(input_ids, + do_sample=False, + 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', '-'*20) + print(output_str) diff --git a/python/llm/example/CPU/HF-Transformers-AutoModels/Model/phi-3/README.md b/python/llm/example/CPU/HF-Transformers-AutoModels/Model/phi-3/README.md index 8f4135ec48a..b3b7dc5f5b2 100644 --- a/python/llm/example/CPU/HF-Transformers-AutoModels/Model/phi-3/README.md +++ b/python/llm/example/CPU/HF-Transformers-AutoModels/Model/phi-3/README.md @@ -76,6 +76,7 @@ In the example, several arguments can be passed to satisfy your requirements: #### 2.4 Sample Output #### [microsoft/Phi-3-mini-4k-instruct](https://huggingface.co/microsoft/Phi-3-mini-4k-instruct) ```log +Inference time: xxxx s -------------------- Prompt -------------------- <|user|> What is AI?<|end|> diff --git a/python/llm/example/CPU/PyTorch-Models/Model/llama3/README.md b/python/llm/example/CPU/PyTorch-Models/Model/llama3/README.md index f518d8873b4..ceadb191ab7 100644 --- a/python/llm/example/CPU/PyTorch-Models/Model/llama3/README.md +++ b/python/llm/example/CPU/PyTorch-Models/Model/llama3/README.md @@ -66,7 +66,7 @@ In the example, several arguments can be passed to satisfy your requirements: - `--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 +#### 2.4 Sample Output #### [meta-llama/Meta-Llama-3-8B-Instruct](https://huggingface.co/meta-llama/Meta-Llama-3-8B-Instruct) ```log Inference time: xxxx s @@ -84,4 +84,4 @@ What is AI?<|eot_id|><|start_header_id|>assistant<|end_header_id|> Artificial Intelligence (AI) refers to the development of computer systems that can perform tasks that would typically require human intelligence, such as: 1. Learning: AI -``` +``` \ No newline at end of file diff --git a/python/llm/example/CPU/PyTorch-Models/Model/minicpm/README.md b/python/llm/example/CPU/PyTorch-Models/Model/minicpm/README.md new file mode 100644 index 00000000000..6a14ffc2212 --- /dev/null +++ b/python/llm/example/CPU/PyTorch-Models/Model/minicpm/README.md @@ -0,0 +1,74 @@ +# MiniCPM +In this directory, you will find examples on how you could use IPEX-LLM `optimize_model` API to accelerate MiniCPM models. For illustration purposes, we utilize the [openbmb/MiniCPM-2B-sft-bf16](https://huggingface.co/openbmb/MiniCPM-2B-sft-bf16) as a reference MiniCPM 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 MiniCPM 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://conda-forge.org/download/). + +After installing conda, create a Python environment for IPEX-LLM: + +On Linux: + +```bash +conda create -n llm python=3.11 # recommend to use Python 3.11 +conda activate llm + +# install the latest ipex-llm nightly build with 'all' option +pip install --pre --upgrade ipex-llm[all] --extra-index-url https://download.pytorch.org/whl/cpu +``` + +On Windows: + +```cmd +conda create -n llm python=3.11 +conda activate llm + +pip install --pre --upgrade ipex-llm[all] +``` + +### 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 MiniCPM model (e.g. `openbmb/MiniCPM-2B-sft-bf16`) to be downloaded, or the path to the huggingface checkpoint folder. It is default to be `'openbmb/MiniCPM-2B-sft-bf16'`. +- `--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 MiniCPM model based on the capabilities of your machine. + +#### 2.1 Client +On client Windows machines, it is recommended to run directly with full utilization of all cores: +```cmd +python ./generate.py --prompt 'What is AI?' +``` + +#### 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?' +``` + +#### 2.3 Sample Output +#### [openbmb/MiniCPM-2B-sft-bf16](https://huggingface.co/openbmb/MiniCPM-2B-sft-bf16) +```log +Inference time: xxxx s +-------------------- Prompt -------------------- +<用户>what is AI? +-------------------- Output -------------------- + <用户>what is AI? AI, or Artificial Intelligence, refers to the simulation of human intelligence in machines that are programmed to think and learn like humans. It is a broad field of computer +``` diff --git a/python/llm/example/CPU/PyTorch-Models/Model/minicpm/generate.py b/python/llm/example/CPU/PyTorch-Models/Model/minicpm/generate.py new file mode 100644 index 00000000000..50ea8ff61d0 --- /dev/null +++ b/python/llm/example/CPU/PyTorch-Models/Model/minicpm/generate.py @@ -0,0 +1,74 @@ +# +# 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 AutoTokenizer, AutoModelForCausalLM +from ipex_llm import optimize_model + + +if __name__ == '__main__': + parser = argparse.ArgumentParser(description='Predict Tokens using `generate()` API for MiniCPM model') + parser.add_argument('--repo-id-or-model-path', type=str, default="openbmb/MiniCPM-2B-sft-bf16", + help='The huggingface repo id for the MiniCPM model 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(): + + # here the prompt formatting refers to: https://huggingface.co/openbmb/MiniCPM-2B-sft-bf16/blob/79fbb1db171e6d8bf77cdb0a94076a43003abd9e/modeling_minicpm.py#L1320 + chat = [ + { "role": "user", "content": args.prompt }, + ] + prompt = tokenizer.apply_chat_template(chat, tokenize=False, add_generation_prompt=False) + input_ids = tokenizer.encode(prompt, return_tensors="pt") + + # start inference + st = time.time() + + output = model.generate(input_ids, + do_sample=False, + 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', '-'*20) + print(output_str) diff --git a/python/llm/example/CPU/PyTorch-Models/Model/phi-3/README.md b/python/llm/example/CPU/PyTorch-Models/Model/phi-3/README.md index d20b271ce9d..3cacedff2e3 100644 --- a/python/llm/example/CPU/PyTorch-Models/Model/phi-3/README.md +++ b/python/llm/example/CPU/PyTorch-Models/Model/phi-3/README.md @@ -73,6 +73,7 @@ In the example, several arguments can be passed to satisfy your requirements: #### 2.4 Sample Output #### [microsoft/Phi-3-mini-4k-instruct](https://huggingface.co/microsoft/Phi-3-mini-4k-instruct) ```log +Inference time: xxxx s -------------------- Prompt -------------------- <|user|> What is AI?<|end|> diff --git a/python/llm/example/GPU/HF-Transformers-AutoModels/Model/minicpm/README.md b/python/llm/example/GPU/HF-Transformers-AutoModels/Model/minicpm/README.md new file mode 100644 index 00000000000..45a213ba48e --- /dev/null +++ b/python/llm/example/GPU/HF-Transformers-AutoModels/Model/minicpm/README.md @@ -0,0 +1,123 @@ +# MiniCPM +In this directory, you will find examples on how you could apply IPEX-LLM INT4 optimizations on MiniCPM models on [Intel GPUs](../../../README.md). For illustration purposes, we utilize the [openbmb/MiniCPM-2B-sft-bf16](https://huggingface.co/openbmb/MiniCPM-2B-sft-bf16) as a reference MiniCPM model. + +## 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 MiniCPM 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/ +``` + +#### 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/ +``` + +### 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 --prompt 'What is AI?' +``` + +Arguments info: +- `--repo-id-or-model-path REPO_ID_OR_MODEL_PATH`: argument defining the huggingface repo id for the MiniCPM model (e.g. `openbmb/MiniCPM-2B-sft-bf16`) to be downloaded, or the path to the huggingface checkpoint folder. It is default to be `'openbmb/MiniCPM-2B-sft-bf16'`. +- `--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 +#### [openbmb/MiniCPM-2B-sft-bf16](https://huggingface.co/openbmb/MiniCPM-2B-sft-bf16) + +```log +Inference time: xxxx s +-------------------- Prompt -------------------- +<用户>what is AI? +-------------------- Output -------------------- + <用户>what is AI? AI, or Artificial Intelligence, refers to the simulation of human intelligence in machines that are programmed to think and learn like humans. It is a field of computer science +``` diff --git a/python/llm/example/GPU/HF-Transformers-AutoModels/Model/minicpm/generate.py b/python/llm/example/GPU/HF-Transformers-AutoModels/Model/minicpm/generate.py new file mode 100644 index 00000000000..669162e61a1 --- /dev/null +++ b/python/llm/example/GPU/HF-Transformers-AutoModels/Model/minicpm/generate.py @@ -0,0 +1,80 @@ +# +# 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 MiniCPM model') + parser.add_argument('--repo-id-or-model-path', type=str, default="openbmb/MiniCPM-2B-sft-bf16", + help='The huggingface repo id for the MiniCPM model 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, + trust_remote_code=True, + optimize_model=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(): + + # here the prompt formatting refers to: https://huggingface.co/openbmb/MiniCPM-2B-sft-bf16/blob/79fbb1db171e6d8bf77cdb0a94076a43003abd9e/modeling_minicpm.py#L1320 + chat = [ + { "role": "user", "content": args.prompt }, + ] + prompt = tokenizer.apply_chat_template(chat, tokenize=False, add_generation_prompt=False) + 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, + do_sample=False, + max_new_tokens=args.n_predict) + torch.xpu.synchronize() + 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', '-'*20) + print(output_str) diff --git a/python/llm/example/GPU/PyTorch-Models/Model/minicpm/README.md b/python/llm/example/GPU/PyTorch-Models/Model/minicpm/README.md new file mode 100644 index 00000000000..bce3215ef91 --- /dev/null +++ b/python/llm/example/GPU/PyTorch-Models/Model/minicpm/README.md @@ -0,0 +1,123 @@ +# MiniCPM +In this directory, you will find examples on how you could use IPEX-LLM `optimize_model` API to accelerate MiniCPM models on [Intel GPUs](../../../README.md). For illustration purposes, we utilize the [openbmb/MiniCPM-2B-sft-bf16](https://huggingface.co/openbmb/MiniCPM-2B-sft-bf16) as a reference MiniCPM model. + +## 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 MiniCPM 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/ +``` + +#### 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/ +``` + +### 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 --prompt 'What is AI?' +``` + +Arguments info: +- `--repo-id-or-model-path REPO_ID_OR_MODEL_PATH`: argument defining the huggingface repo id for the MiniCPM model (e.g. `openbmb/MiniCPM-2B-sft-bf16`) to be downloaded, or the path to the huggingface checkpoint folder. It is default to be `'openbmb/MiniCPM-2B-sft-bf16'`. +- `--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 +#### [openbmb/MiniCPM-2B-sft-bf16](https://huggingface.co/openbmb/MiniCPM-2B-sft-bf16) + +```log +Inference time: xxxx s +-------------------- Prompt -------------------- +<用户>what is AI? +-------------------- Output -------------------- + <用户>what is AI? AI, or Artificial Intelligence, refers to the simulation of human intelligence in machines that are programmed to think and learn like humans. It is a field of computer science +``` diff --git a/python/llm/example/GPU/PyTorch-Models/Model/minicpm/generate.py b/python/llm/example/GPU/PyTorch-Models/Model/minicpm/generate.py new file mode 100644 index 00000000000..b6f9a4cf3ca --- /dev/null +++ b/python/llm/example/GPU/PyTorch-Models/Model/minicpm/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 transformers import AutoModelForCausalLM, AutoTokenizer +from ipex_llm import optimize_model + + +if __name__ == '__main__': + parser = argparse.ArgumentParser(description='Predict Tokens using `generate()` API for MiniCPM model') + parser.add_argument('--repo-id-or-model-path', type=str, default="openbmb/MiniCPM-2B-sft-bf16", + help='The huggingface repo id for the MiniCPM model 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.to('xpu') + + # Load tokenizer + tokenizer = AutoTokenizer.from_pretrained(model_path, + trust_remote_code=True) + + # Generate predicted tokens + with torch.inference_mode(): + + # here the prompt formatting refers to: https://huggingface.co/openbmb/MiniCPM-2B-sft-bf16/blob/79fbb1db171e6d8bf77cdb0a94076a43003abd9e/modeling_minicpm.py#L1320 + chat = [ + { "role": "user", "content": args.prompt }, + ] + prompt = tokenizer.apply_chat_template(chat, tokenize=False, add_generation_prompt=False) + 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, + do_sample=False, + max_new_tokens=args.n_predict) + torch.xpu.synchronize() + 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', '-'*20) + print(output_str)