diff --git a/README.md b/README.md index 8901f4bb636..ff4a50ecbd3 100644 --- a/README.md +++ b/README.md @@ -154,6 +154,7 @@ Over 50 models have been optimized/verified on `ipex-llm`, including *LLaMA/LLaM | 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) | +| GLM-4 | [link](python/llm/example/CPU/HF-Transformers-AutoModels/Model/glm4) | [link](python/llm/example/GPU/HF-Transformers-AutoModels/Model/glm4) | | Mistral | [link](python/llm/example/CPU/HF-Transformers-AutoModels/Model/mistral) | [link](python/llm/example/GPU/HF-Transformers-AutoModels/Model/mistral) | | Mixtral | [link](python/llm/example/CPU/HF-Transformers-AutoModels/Model/mixtral) | [link](python/llm/example/GPU/HF-Transformers-AutoModels/Model/mixtral) | | Falcon | [link](python/llm/example/CPU/HF-Transformers-AutoModels/Model/falcon) | [link](python/llm/example/GPU/HF-Transformers-AutoModels/Model/falcon) | diff --git a/docs/readthedocs/source/index.rst b/docs/readthedocs/source/index.rst index 1071b5692ab..1a5c6055f3f 100644 --- a/docs/readthedocs/source/index.rst +++ b/docs/readthedocs/source/index.rst @@ -257,6 +257,13 @@ Verified Models link + + GLM-4 + + link + + link + Mistral diff --git a/python/llm/example/CPU/HF-Transformers-AutoModels/Model/glm4/README.md b/python/llm/example/CPU/HF-Transformers-AutoModels/Model/glm4/README.md new file mode 100644 index 00000000000..d3d2966159c --- /dev/null +++ b/python/llm/example/CPU/HF-Transformers-AutoModels/Model/glm4/README.md @@ -0,0 +1,166 @@ +# GLM-4 + +In this directory, you will find examples on how you could apply IPEX-LLM INT4 optimizations on GLM-4 models. For illustration purposes, we utilize the [THUDM/glm-4-9b-chat](https://huggingface.co/THUDM/glm-4-9b-chat) as a reference GLM-4 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 1: Predict Tokens using `generate()` API +In the example [generate.py](./generate.py), we show a basic use case for a GLM-4 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 # 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 + +# install tiktoken required for GLM-4 +pip install tiktoken +``` + +On Windows: + +```cmd +conda create -n llm python=3.11 +conda activate llm + +pip install --pre --upgrade ipex-llm[all] + +pip install tiktoken +``` + +### 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 GLM-4 model to be downloaded, or the path to the huggingface checkpoint folder. It is default to be `'THUDM/glm-4-9b-chat'`. +- `--prompt PROMPT`: argument defining the prompt to be infered (with integrated prompt format for chat). It is default to be `'AI是什么?'`. +- `--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 GLM-4 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 +##### [THUDM/glm-4-9b-chat](https://huggingface.co/THUDM/glm-4-9b-chat) +```log +Inference time: xxxx s +-------------------- Prompt -------------------- +<|user|> +AI是什么? +<|assistant|> +-------------------- Output -------------------- + +AI是什么? + +AI,即人工智能(Artificial Intelligence),是指由人创造出来的,能够模拟、延伸和扩展人的智能的计算机系统或机器。人工智能技术 +``` + +```log +Inference time: xxxx s +-------------------- Prompt -------------------- +<|user|> +What is AI? +<|assistant|> +-------------------- Output -------------------- + +What is AI? + +Artificial Intelligence (AI) refers to the simulation of human intelligence in machines that are programmed to think like humans and mimic their actions. The term "art +``` + +## Example 2: Stream Chat using `stream_chat()` API +In the example [streamchat.py](./streamchat.py), we show a basic use case for a GLM-4 model to stream chat, with IPEX-LLM INT4 optimizations. +### 1. Install +We suggest using conda to manage environment: + +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 + +# install tiktoken required for GLM-4 +pip install tiktoken +``` + +On Windows: + +```cmd +conda create -n llm python=3.11 +conda activate llm + +pip install --pre --upgrade ipex-llm[all] + +pip install tiktoken +``` + +### 2. Run +**Stream Chat using `stream_chat()` API**: +``` +python ./streamchat.py --repo-id-or-model-path REPO_ID_OR_MODEL_PATH --question QUESTION +``` + +**Chat using `chat()` API**: +``` +python ./streamchat.py --repo-id-or-model-path REPO_ID_OR_MODEL_PATH --question QUESTION --disable-stream +``` + +Arguments info: +- `--repo-id-or-model-path REPO_ID_OR_MODEL_PATH`: argument defining the huggingface repo id for the GLM-4 model to be downloaded, or the path to the huggingface checkpoint folder. It is default to be `'THUDM/glm-4-9b-chat'`. +- `--question QUESTION`: argument defining the question to ask. It is default to be `"晚上睡不着应该怎么办"`. +- `--disable-stream`: argument defining whether to stream chat. If include `--disable-stream` when running the script, the stream chat is disabled and `chat()` API is used. + +> **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 GLM-4 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 +$env:PYTHONUNBUFFERED=1 # ensure stdout and stderr streams are sent straight to terminal without being first buffered +python ./streamchat.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 +export PYTHONUNBUFFERED=1 # ensure stdout and stderr streams are sent straight to terminal without being first buffered +numactl -C 0-47 -m 0 python ./streamchat.py +``` diff --git a/python/llm/example/CPU/HF-Transformers-AutoModels/Model/glm4/generate.py b/python/llm/example/CPU/HF-Transformers-AutoModels/Model/glm4/generate.py new file mode 100644 index 00000000000..9ec89f2c8b6 --- /dev/null +++ b/python/llm/example/CPU/HF-Transformers-AutoModels/Model/glm4/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 +import numpy as np + +from ipex_llm.transformers import AutoModel +from transformers import AutoTokenizer + +# you could tune the prompt based on your own model, +# here the prompt tuning refers to https://huggingface.co/THUDM/glm-4-9b-chat/blob/main/tokenization_chatglm.py +GLM4_PROMPT_FORMAT = "<|user|>\n{prompt}\n<|assistant|>" + +if __name__ == '__main__': + parser = argparse.ArgumentParser(description='Predict Tokens using `generate()` API for GLM-4 model') + parser.add_argument('--repo-id-or-model-path', type=str, default="THUDM/glm-4-9b-chat", + help='The huggingface repo id for the GLM-4 model to be downloaded' + ', or the path to the huggingface checkpoint folder') + parser.add_argument('--prompt', type=str, default="AI是什么?", + 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 = AutoModel.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 = GLM4_PROMPT_FORMAT.format(prompt=args.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=True) + 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/glm4/streamchat.py b/python/llm/example/CPU/HF-Transformers-AutoModels/Model/glm4/streamchat.py new file mode 100644 index 00000000000..79243fde5f1 --- /dev/null +++ b/python/llm/example/CPU/HF-Transformers-AutoModels/Model/glm4/streamchat.py @@ -0,0 +1,62 @@ +# +# 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 +import numpy as np + +from ipex_llm.transformers import AutoModel +from transformers import AutoTokenizer + + +if __name__ == '__main__': + parser = argparse.ArgumentParser(description='Stream Chat for GLM-4 model') + parser.add_argument('--repo-id-or-model-path', type=str, default="THUDM/glm-4-9b-chat", + help='The huggingface repo id for the GLM-4 model to be downloaded' + ', or the path to the huggingface checkpoint folder') + parser.add_argument('--question', type=str, default="晚上睡不着应该怎么办", + help='Qustion you want to ask') + parser.add_argument('--disable-stream', action="store_true", + help='Disable stream chat') + + args = parser.parse_args() + model_path = args.repo_id_or_model_path + disable_stream = args.disable_stream + + # Load model in 4 bit, + # which convert the relevant layers in the model into INT4 format + model = AutoModel.from_pretrained(model_path, + load_in_4bit=True, + trust_remote_code=True) + + # Load tokenizer + tokenizer = AutoTokenizer.from_pretrained(model_path, + trust_remote_code=True) + + with torch.inference_mode(): + if disable_stream: + # Chat + response, history = model.chat(tokenizer, args.question, history=[]) + print('-'*20, 'Chat Output', '-'*20) + print(response) + else: + # Stream chat + response_ = "" + print('-'*20, 'Stream Chat Output', '-'*20) + for response, history in model.stream_chat(tokenizer, args.question, history=[]): + print(response.replace(response_, ""), end="") + response_ = response diff --git a/python/llm/example/CPU/PyTorch-Models/Model/glm4/README.md b/python/llm/example/CPU/PyTorch-Models/Model/glm4/README.md new file mode 100644 index 00000000000..9a1cf56069d --- /dev/null +++ b/python/llm/example/CPU/PyTorch-Models/Model/glm4/README.md @@ -0,0 +1,87 @@ +# GLM-4 +In this directory, you will find examples on how you could use IPEX-LLM `optimize_model` API to accelerate GLM-4 models. For illustration purposes, we utilize the [THUDM/glm-4-9b-chat](https://huggingface.co/THUDM/glm-4-9b-chat) as a reference GLM-4 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 GLM-4 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 + +# install tiktoken required for GLM-4 +pip install tiktoken +``` + +On Windows: + +```cmd +conda create -n llm python=3.11 +conda activate llm + +pip install --pre --upgrade ipex-llm[all] + +pip install tiktoken +``` + +### 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: +```cmd +python ./generate.py --prompt 'AI是什么?' +``` +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 'AI是什么?' +``` +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`: str, argument defining the huggingface repo id for the GLM-4 model to be downloaded, or the path to the huggingface checkpoint folder. It is default to be `'THUDM/glm-4-9b-chat'`. +- `--prompt`: str, argument defining the prompt to be inferred (with integrated prompt format for chat). It is default to be `'AI是什么?'`. +- `--n-predict`: int, argument defining the max number of tokens to predict. It is default to be `32`. + +#### 2.4 Sample Output +##### [THUDM/glm-4-9b-chat](https://huggingface.co/THUDM/glm-4-9b-chat) +```log +Inference time: xxxx s +-------------------- Output -------------------- + +AI是什么? + +AI,即人工智能(Artificial Intelligence),是指由人创造出来的,能够模拟、延伸和扩展人的智能的计算机系统或机器。人工智能技术 +``` + +``` +Inference time: xxxx s +-------------------- Output -------------------- + +What is AI? + +Artificial Intelligence (AI) refers to the simulation of human intelligence in machines that are programmed to think like humans and mimic their actions. The term "art +``` diff --git a/python/llm/example/CPU/PyTorch-Models/Model/glm4/generate.py b/python/llm/example/CPU/PyTorch-Models/Model/glm4/generate.py new file mode 100644 index 00000000000..e09ab4a43c7 --- /dev/null +++ b/python/llm/example/CPU/PyTorch-Models/Model/glm4/generate.py @@ -0,0 +1,65 @@ +# +# 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 AutoModel, AutoTokenizer +from ipex_llm import optimize_model + +# you could tune the prompt based on your own model, +# here the prompt tuning refers to https://huggingface.co/THUDM/glm-4-9b-chat/blob/main/tokenization_chatglm.py +GLM4_PROMPT_FORMAT = "<|user|>\n{prompt}\n<|assistant|>" + +if __name__ == '__main__': + parser = argparse.ArgumentParser(description='Predict Tokens using `generate()` API for GLM-4 model') + parser.add_argument('--repo-id-or-model-path', type=str, default="THUDM/glm-4-9b-chat", + help='The huggingface repo id for the GLM-4 model to be downloaded' + ', or the path to the huggingface checkpoint folder') + parser.add_argument('--prompt', type=str, default="AI是什么?", + 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 = AutoModel.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 = GLM4_PROMPT_FORMAT.format(prompt=args.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=True) + print(f'Inference time: {end-st} s') + print('-'*20, 'Output', '-'*20) + print(output_str)