From 1a1ddc4144ee63f178fd2255424975d5f0ee57d5 Mon Sep 17 00:00:00 2001 From: JIN Qiao <89779290+JinBridger@users.noreply.github.com> Date: Thu, 12 Oct 2023 13:42:14 +0800 Subject: [PATCH] LLM: Add Replit CPU and GPU example (#9028) --- .../Model/README.md | 1 + .../Model/replit/README.md | 66 ++++++++++++++++ .../Model/replit/generate.py | 67 ++++++++++++++++ .../Model/README.md | 3 + .../Model/replit/README.md | 64 +++++++++++++++ .../Model/replit/generate.py | 78 +++++++++++++++++++ 6 files changed, 279 insertions(+) create mode 100644 python/llm/example/CPU/HF-Transformers-AutoModels/Model/replit/README.md create mode 100644 python/llm/example/CPU/HF-Transformers-AutoModels/Model/replit/generate.py create mode 100644 python/llm/example/GPU/HF-Transformers-AutoModels/Model/replit/README.md create mode 100644 python/llm/example/GPU/HF-Transformers-AutoModels/Model/replit/generate.py diff --git a/python/llm/example/CPU/HF-Transformers-AutoModels/Model/README.md b/python/llm/example/CPU/HF-Transformers-AutoModels/Model/README.md index 3d55bbe0..55d4c41b 100644 --- a/python/llm/example/CPU/HF-Transformers-AutoModels/Model/README.md +++ b/python/llm/example/CPU/HF-Transformers-AutoModels/Model/README.md @@ -22,6 +22,7 @@ You can use BigDL-LLM to run any Huggingface Transformer models with INT4 optimi | Whisper | [link](whisper) | | Qwen | [link](qwen) | | Aquila | [link](aquila) | +| Replit | [link](replit) | | Mistral | [link](mistral) | ## Recommended Requirements diff --git a/python/llm/example/CPU/HF-Transformers-AutoModels/Model/replit/README.md b/python/llm/example/CPU/HF-Transformers-AutoModels/Model/replit/README.md new file mode 100644 index 00000000..b191e32d --- /dev/null +++ b/python/llm/example/CPU/HF-Transformers-AutoModels/Model/replit/README.md @@ -0,0 +1,66 @@ +# Replit +In this directory, you will find examples on how you could apply BigDL-LLM INT4 optimizations on Replit models. For illustration purposes, we utilize the [replit/replit-code-v1-3b](https://huggingface.co/replit/replit-code-v1-3b) as a reference Replit model. + +## 0. Requirements +To run these examples with BigDL-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 an Replit model to predict the next N tokens using `generate()` API, with BigDL-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 BigDL-LLM: +```bash +conda create -n llm python=3.9 # recommend to use Python 3.9 +conda activate llm + +pip install --pre --upgrade bigdl-llm[all] # install the latest bigdl-llm nightly build with 'all' option +``` + +### 2. Run +After setting up the Python environment, you could run the example by following steps. + +#### 2.1 Client +On client Windows machine, it is recommended to run directly with full utilization of all cores: +```powershell +python ./generate.py --prompt 'def print_hello_world():' +``` +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 BigDL-Nano env variables +source bigdl-nano-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 +``` +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 Replit model to be downloaded, or the path to the huggingface checkpoint folder. It is default to be `'replit/replit-code-v1-3b'`. +- `--prompt PROMPT`: argument defining the prompt to be infered (with integrated prompt format for chat). It is default to be `'def print_hello_world():'`. +- `--n-predict N_PREDICT`: argument defining the max number of tokens to predict. It is default to be `32`. + +#### 2.4 Sample Output +#### [replit/replit-code-v1-3b](https://huggingface.co/bigcode/replit/replit-code-v1-3b) +```log +-------------------- Prompt -------------------- +def print_hello_world(): +-------------------- Output -------------------- +def print_hello_world(): + print("Hello") + print("World") + +print_hello_world() + + +def print_hello_world(): + print +``` \ No newline at end of file diff --git a/python/llm/example/CPU/HF-Transformers-AutoModels/Model/replit/generate.py b/python/llm/example/CPU/HF-Transformers-AutoModels/Model/replit/generate.py new file mode 100644 index 00000000..014d5f13 --- /dev/null +++ b/python/llm/example/CPU/HF-Transformers-AutoModels/Model/replit/generate.py @@ -0,0 +1,67 @@ +# +# 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 bigdl.llm.transformers import AutoModelForCausalLM +from transformers import AutoTokenizer + +# you could tune the prompt based on your own model, +REPLIT_PROMPT_FORMAT = "{prompt}" + +if __name__ == '__main__': + parser = argparse.ArgumentParser(description='Predict Tokens using `generate()` API for Replit model') + parser.add_argument('--repo-id-or-model-path', type=str, default="replit/replit-code-v1-3b", + help='The huggingface repo id for the Replit to be downloaded' + ', or the path to the huggingface checkpoint folder') + parser.add_argument('--prompt', type=str, default="def print_hello_world():", + 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, + trust_remote_code=True) + + # Load tokenizer + tokenizer = AutoTokenizer.from_pretrained(model_path, + trust_remote_code=True) + + # Generate predicted tokens + with torch.inference_mode(): + prompt = REPLIT_PROMPT_FORMAT.format(prompt=args.prompt) + input_ids = tokenizer.encode(prompt, return_tensors="pt") + 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 BigDL-LLM INT4 optimizations + output = model.generate(input_ids, + max_new_tokens=args.n_predict) + end = time.time() + output_str = tokenizer.decode(output[0], skip_special_tokens=True) + print(f'Inference time: {end-st} s') + print('-'*20, 'Prompt', '-'*20) + print(prompt) + print('-'*20, 'Output', '-'*20) + print(output_str) \ No newline at end of file diff --git a/python/llm/example/GPU/HF-Transformers-AutoModels/Model/README.md b/python/llm/example/GPU/HF-Transformers-AutoModels/Model/README.md index b122285d..6f8fc22b 100644 --- a/python/llm/example/GPU/HF-Transformers-AutoModels/Model/README.md +++ b/python/llm/example/GPU/HF-Transformers-AutoModels/Model/README.md @@ -2,6 +2,7 @@ You can use BigDL-LLM to run almost every Huggingface Transformer models with INT4 optimizations on your laptops with Intel GPUs. This directory contains example scripts to help you quickly get started using BigDL-LLM to run some popular open-source models in the community. Each model has its own dedicated folder, where you can find detailed instructions on how to install and run it. ## Verified models + | Model | Example | |----------------|----------------------------------------------------------| | Aquila | [link](aquila) | @@ -21,6 +22,8 @@ You can use BigDL-LLM to run almost every Huggingface Transformer models with IN | StarCoder | [link](starcoder) | | Vicuna | [link](vicuna) | | Whisper | [link](whisper) | +| Replit | [link](replit) | + ## Verified Hardware Platforms diff --git a/python/llm/example/GPU/HF-Transformers-AutoModels/Model/replit/README.md b/python/llm/example/GPU/HF-Transformers-AutoModels/Model/replit/README.md new file mode 100644 index 00000000..a1f631a4 --- /dev/null +++ b/python/llm/example/GPU/HF-Transformers-AutoModels/Model/replit/README.md @@ -0,0 +1,64 @@ +# Replit +In this directory, you will find examples on how you could apply BigDL-LLM INT4 optimizations on Replit models on [Intel GPUs](../README.md). For illustration purposes, we utilize the [replit/replit-code-v1-3b](https://huggingface.co/replit/replit-code-v1-3b) as a reference Replit model. + +## 0. Requirements +To run these examples with BigDL-LLM on Intel GPUs, 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 an Replit model to predict the next N tokens using `generate()` API, with BigDL-LLM INT4 optimizations on Intel GPUs. +### 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 BigDL-LLM: +```bash +conda create -n llm python=3.9 +conda activate llm +# below command will install intel_extension_for_pytorch==2.0.110+xpu as default +# you can install specific ipex/torch version for your need +pip install --pre --upgrade bigdl-llm[xpu] -f https://developer.intel.com/ipex-whl-stable-xpu +``` + +### 2. Configures OneAPI environment variables +```bash +source /opt/intel/oneapi/setvars.sh +``` + +### 3. Run + +For optimal performance on Arc, it is recommended to set several environment variables. + +```bash +export USE_XETLA=OFF +export SYCL_PI_LEVEL_ZERO_USE_IMMEDIATE_COMMANDLISTS=1 +``` + +``` +python ./generate.py --prompt 'def print_hello_world():' +``` +More information about arguments can be found in [Arguments Info](#31-arguments-info) section. The expected output can be found in [Sample Output](#32-sample-output) section. + +#### 3.1 Arguments Info +In the example, several arguments can be passed to satisfy your requirements: + +Arguments info: +- `--repo-id-or-model-path REPO_ID_OR_MODEL_PATH`: argument defining the huggingface repo id for the Replit model to be downloaded, or the path to the huggingface checkpoint folder. It is default to be `'replit/replit-code-v1-3b'`. +- `--prompt PROMPT`: argument defining the prompt to be infered (with integrated prompt format for chat). It is default to be `'def print_hello_world():'`. +- `--n-predict N_PREDICT`: argument defining the max number of tokens to predict. It is default to be `32`. + +#### 3.2 Sample Output +#### [replit/replit-code-v1-3b](https://huggingface.co/replit/replit-code-v1-3b) +```log +Inference time: xxxx s +-------------------- Prompt -------------------- +def print_hello_world(): +-------------------- Output -------------------- +def print_hello_world(): + print("Hello") + print("World") + +print_hello_world() + + +def print_hello_world(): + print +``` \ No newline at end of file diff --git a/python/llm/example/GPU/HF-Transformers-AutoModels/Model/replit/generate.py b/python/llm/example/GPU/HF-Transformers-AutoModels/Model/replit/generate.py new file mode 100644 index 00000000..9934eda4 --- /dev/null +++ b/python/llm/example/GPU/HF-Transformers-AutoModels/Model/replit/generate.py @@ -0,0 +1,78 @@ +# +# 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 intel_extension_for_pytorch as ipex +import time +import argparse + +from bigdl.llm.transformers import AutoModelForCausalLM +from transformers import AutoTokenizer + +# you could tune the prompt based on your own model, +REPLIT_PROMPT_FORMAT = "{prompt}" + +if __name__ == '__main__': + parser = argparse.ArgumentParser(description='Predict Tokens using `generate()` API for Replit model') + parser.add_argument('--repo-id-or-model-path', type=str, default="replit/replit-code-v1-3b", + help='The huggingface repo id for the Replit to be downloaded' + ', or the path to the huggingface checkpoint folder') + parser.add_argument('--prompt', type=str, default="def print_hello_world():", + 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) + model = model.to('xpu') + + # Load tokenizer + tokenizer = AutoTokenizer.from_pretrained(model_path, + trust_remote_code=True) + + # Generate predicted tokens + with torch.inference_mode(): + prompt = REPLIT_PROMPT_FORMAT.format(prompt=args.prompt) + input_ids = tokenizer.encode(prompt, return_tensors="pt").to('xpu') + + # ipex 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 BigDL-LLM INT4 optimizations + output = model.generate(input_ids, + max_new_tokens=args.n_predict) + torch.xpu.synchronize() + end = time.time() + output_str = tokenizer.decode(output[0], skip_special_tokens=True) + print(f'Inference time: {end-st} s') + print('-'*20, 'Prompt', '-'*20) + print(prompt) + print('-'*20, 'Output', '-'*20) + print(output_str) \ No newline at end of file