Add --modelscope in GPU examples for glm4, codegeex2, qwen2 and qwen2.5 (#12561)
* Add --modelscope for more models * imporve readme --------- Co-authored-by: ATMxsp01 <shou.xu@intel.com>
This commit is contained in:
Xu, Shuo
GitHub
parent
28e81fda8e
commit
47e90a362f
10 changed files with 125 additions and 125 deletions
|
|
@ -108,7 +108,7 @@ python ./generate.py --repo-id-or-model-path REPO_ID_OR_MODEL_PATH --prompt PROM
|
|||
```
|
||||
|
||||
Arguments info:
|
||||
- `--repo-id-or-model-path REPO_ID_OR_MODEL_PATH`: argument defining the **Hugging Face** or **ModelScope** repo id for the ChatGLM3 model to be downloaded, or the path to the checkpoint folder. It is default to be `'THUDM/chatglm3-6b'` for **Hugging Face** or `ZhipuAI/chatglm3-6b` for **ModelScope**.
|
||||
- `--repo-id-or-model-path REPO_ID_OR_MODEL_PATH`: argument defining the **Hugging Face** or **ModelScope** repo id for the ChatGLM3 model to be downloaded, or the path to the checkpoint folder. It is default to be `'THUDM/chatglm3-6b'` for **Hugging Face** or `'ZhipuAI/chatglm3-6b'` for **ModelScope**.
|
||||
- `--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`.
|
||||
- `--modelscope`: using **ModelScope** as model hub instead of **Hugging Face**.
|
||||
|
|
@ -162,7 +162,7 @@ python ./streamchat.py --repo-id-or-model-path REPO_ID_OR_MODEL_PATH --question
|
|||
```
|
||||
|
||||
Arguments info:
|
||||
- `--repo-id-or-model-path REPO_ID_OR_MODEL_PATH`: argument defining the **Hugging Face** or **ModelScope** repo id for the ChatGLM3 model to be downloaded, or the path to the checkpoint folder. It is default to be `'THUDM/chatglm3-6b'` for **Hugging Face** or `ZhipuAI/chatglm3-6b` for **ModelScope**.
|
||||
- `--repo-id-or-model-path REPO_ID_OR_MODEL_PATH`: argument defining the **Hugging Face** or **ModelScope** repo id for the ChatGLM3 model to be downloaded, or the path to the checkpoint folder. It is default to be `'THUDM/chatglm3-6b'` for **Hugging Face** or `'ZhipuAI/chatglm3-6b'` for **ModelScope**.
|
||||
- `--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.
|
||||
- `--modelscope`: using **ModelScope** as model hub instead of **Hugging Face**.
|
||||
|
|
|
|||
|
|
@ -1,6 +1,6 @@
|
|||
# CodeGeeX2
|
||||
|
||||
In this directory, you will find examples on how you could apply IPEX-LLM INT4 optimizations on CodeGeeX2 models which is implemented based on the ChatGLM2 architecture trained on more code data on [Intel GPUs](../../../README.md). For illustration purposes, we utilize the [THUDM/codegeex-6b](https://huggingface.co/THUDM/codegeex2-6b) as a reference CodeGeeX2 model.
|
||||
In this directory, you will find examples on how you could apply IPEX-LLM INT4 optimizations on CodeGeeX2 models which is implemented based on the ChatGLM2 architecture trained on more code data on [Intel GPUs](../../../README.md). For illustration purposes, we utilize the [THUDM/codegeex2-6b](https://huggingface.co/THUDM/codegeex2-6b) (or [ZhipuAI/codegeex2-6b](https://www.modelscope.cn/models/ZhipuAI/codegeex2-6b) for ModelScope) as a reference CodeGeeX2 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.
|
||||
|
|
@ -16,6 +16,9 @@ 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/
|
||||
|
||||
# [optional] only needed if you would like to use ModelScope as model hub
|
||||
pip install modelscope==1.11.0
|
||||
```
|
||||
|
||||
#### 1.2 Installation on Windows
|
||||
|
|
@ -26,10 +29,13 @@ 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/
|
||||
|
||||
# [optional] only needed if you would like to use ModelScope as model hub
|
||||
pip install modelscope==1.11.0
|
||||
```
|
||||
|
||||
### 2. Download Model and Replace File
|
||||
If you select the codegeex2-6b model ([THUDM/codegeex-6b](https://huggingface.co/THUDM/codegeex2-6b)), please note that their code (`tokenization_chatglm.py`) initialized tokenizer after the call of `__init__` of its parent class, which may result in error during loading tokenizer. To address issue, we have provided an updated file ([tokenization_chatglm.py](./codegeex2-6b/tokenization_chatglm.py))
|
||||
If you select the codegeex2-6b model ([THUDM/codegeex2-6b](https://huggingface.co/THUDM/codegeex2-6b) (for **Hugging Face**) or [ZhipuAI/codegeex2-6b](https://www.modelscope.cn/models/ZhipuAI/codegeex2-6b) (for **ModelScope**)), please note that their code (`tokenization_chatglm.py`) initialized tokenizer after the call of `__init__` of its parent class, which may result in error during loading tokenizer. To address issue, we have provided an updated file ([tokenization_chatglm.py](./codegeex2-6b/tokenization_chatglm.py))
|
||||
|
||||
```python
|
||||
def __init__(self, vocab_file, padding_side="left", clean_up_tokenization_spaces=False, **kwargs):
|
||||
|
|
@ -37,7 +43,7 @@ def __init__(self, vocab_file, padding_side="left", clean_up_tokenization_spaces
|
|||
super().__init__(padding_side=padding_side, clean_up_tokenization_spaces=clean_up_tokenization_spaces, **kwargs)
|
||||
```
|
||||
|
||||
You could download the model from [THUDM/codegeex-6b](https://huggingface.co/THUDM/codegeex2-6b), and replace the file `tokenization_chatglm.py` with [tokenization_chatglm.py](./codegeex2-6b/tokenization_chatglm.py).
|
||||
You could download the model from [THUDM/codegeex2-6b](https://huggingface.co/THUDM/codegeex2-6b) (for **Hugging Face**) or [ZhipuAI/codegeex2-6b](https://www.modelscope.cn/models/ZhipuAI/codegeex2-6b) (for **ModelScope**), and replace the file `tokenization_chatglm.py` with [tokenization_chatglm.py](./codegeex2-6b/tokenization_chatglm.py).
|
||||
|
||||
### 3. Configures OneAPI environment variables for Linux
|
||||
|
||||
|
|
@ -104,17 +110,22 @@ set SYCL_CACHE_PERSISTENT=1
|
|||
> 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.
|
||||
|
||||
### 5. Running examples
|
||||
```
|
||||
```bash
|
||||
# for Hugging Face model hub
|
||||
python ./generate.py --repo-id-or-model-path REPO_ID_OR_MODEL_PATH --prompt PROMPT --n-predict N_PREDICT
|
||||
|
||||
# for ModelScope model hub
|
||||
python ./generate.py --repo-id-or-model-path REPO_ID_OR_MODEL_PATH --prompt PROMPT --n-predict N_PREDICT --modelscope
|
||||
```
|
||||
|
||||
Arguments info:
|
||||
- `--repo-id-or-model-path REPO_ID_OR_MODEL_PATH`: argument defining the huggingface repo id for the CodeGeeX2 model to be downloaded, or the path to the huggingface checkpoint folder. It is default to be `'THUDM/codegeex-6b'`.
|
||||
- `--repo-id-or-model-path REPO_ID_OR_MODEL_PATH`: argument defining the **Hugging Face** or **ModelScope** repo id for the CodeGeeX2 model to be downloaded, or the path to the checkpoint folder. It is default to be `'THUDM/codegeex2-6b'` for **Hugging Face** or `'ZhipuAI/codegeex-6b'` for **ModelScope**.
|
||||
- `--prompt PROMPT`: argument defining the prompt to be infered (with integrated prompt format for chat). It is default to be `'# language: Python\n# write a bubble sort function\n'`.
|
||||
- `--n-predict N_PREDICT`: argument defining the max number of tokens to predict. It is default to be `128`.
|
||||
- `--modelscope`: using **ModelScope** as model hub instead of **Hugging Face**.
|
||||
|
||||
#### Sample Output
|
||||
#### [THUDM/codegeex-6b](https://huggingface.co/THUDM/codegeex-6b)
|
||||
#### [THUDM/codegeex2-6b](https://huggingface.co/THUDM/codegeex2-6b)
|
||||
```log
|
||||
Inference time: xxxx s
|
||||
-------------------- Prompt --------------------
|
||||
|
|
|
|||
|
|
@ -28,18 +28,29 @@ CODEGEEX_PROMPT_FORMAT = "{prompt}"
|
|||
|
||||
|
||||
if __name__ == '__main__':
|
||||
parser = argparse.ArgumentParser(description='Predict Tokens using `generate()` API for ChatGLM2 model')
|
||||
parser.add_argument('--repo-id-or-model-path', type=str, default="THUDM/codegeex2-6b",
|
||||
help='The huggingface repo id for the CodeGeeX2 model to be downloaded'
|
||||
', or the path to the huggingface checkpoint folder')
|
||||
parser = argparse.ArgumentParser(description='Predict Tokens using `generate()` API for CodeGeeX2 model')
|
||||
parser.add_argument('--repo-id-or-model-path', type=str,
|
||||
help='The Hugging Face or ModelScope repo id for the CodeGeeX2 model to be downloaded'
|
||||
', or the path to the checkpoint folder')
|
||||
parser.add_argument('--prompt', type=str, default="# language: Python\n# write a bubble sort function\n",
|
||||
help='Prompt to infer')
|
||||
parser.add_argument('--n-predict', type=int, default=128,
|
||||
help='Max tokens to predict')
|
||||
parser.add_argument('--modelscope', action="store_true", default=False,
|
||||
help="Use models from modelscope")
|
||||
|
||||
args = parser.parse_args()
|
||||
model_path = args.repo_id_or_model_path
|
||||
|
||||
if args.modelscope:
|
||||
from modelscope import AutoTokenizer
|
||||
model_hub = 'modelscope'
|
||||
else:
|
||||
from transformers import AutoTokenizer
|
||||
model_hub = 'huggingface'
|
||||
|
||||
model_path = args.repo_id_or_model_path if args.repo_id_or_model_path else \
|
||||
("ZhipuAI/codegeex2-6b" if args.modelscope else "THUDM/codegeex2-6b")
|
||||
|
||||
# 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.
|
||||
|
|
@ -48,7 +59,8 @@ if __name__ == '__main__':
|
|||
load_in_4bit=True,
|
||||
optimize_model=True,
|
||||
trust_remote_code=True,
|
||||
use_cache=True)
|
||||
use_cache=True,
|
||||
model_hub=model_hub)
|
||||
model = model.half().to('xpu')
|
||||
|
||||
# Load tokenizer
|
||||
|
|
|
|||
|
|
@ -1,5 +1,5 @@
|
|||
# GLM-4
|
||||
In this directory, you will find examples on how you could apply IPEX-LLM INT4 optimizations on GLM-4 models on [Intel GPUs](../../../README.md). For illustration purposes, we utilize the [THUDM/glm-4-9b-chat](https://huggingface.co/THUDM/glm-4-9b-chat) as a reference InternLM model.
|
||||
In this directory, you will find examples on how you could apply IPEX-LLM INT4 optimizations on GLM-4 models on [Intel GPUs](../../../README.md). For illustration purposes, we utilize the [THUDM/glm-4-9b-chat](https://huggingface.co/THUDM/glm-4-9b-chat) (or [ZhipuAI/glm4-9b-chat](https://www.modelscope.cn/models/ZhipuAI/glm4-9b-chat) for ModelScope) as a reference InternLM 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.
|
||||
|
|
@ -15,6 +15,9 @@ pip install --pre --upgrade ipex-llm[xpu] --extra-index-url https://pytorch-exte
|
|||
|
||||
# install packages required for GLM-4, it is recommended to use transformers>=4.44 for THUDM/glm-4-9b-chat updated after August 12, 2024
|
||||
pip install "tiktoken>=0.7.0" transformers==4.44 "trl<0.12.0"
|
||||
|
||||
# [optional] only needed if you would like to use ModelScope as model hub
|
||||
pip install modelscope==1.11.0
|
||||
```
|
||||
|
||||
### 1.2 Installation on Windows
|
||||
|
|
@ -28,6 +31,9 @@ pip install --pre --upgrade ipex-llm[xpu] --extra-index-url https://pytorch-exte
|
|||
|
||||
# install packages required for GLM-4, it is recommended to use transformers>=4.44 for THUDM/glm-4-9b-chat updated after August 12, 2024
|
||||
pip install "tiktoken>=0.7.0" transformers==4.44 "trl<0.12.0"
|
||||
|
||||
# [optional] only needed if you would like to use ModelScope as model hub
|
||||
pip install modelscope==1.11.0
|
||||
```
|
||||
|
||||
## 2. Configures OneAPI environment variables for Linux
|
||||
|
|
@ -98,14 +104,19 @@ set SYCL_CACHE_PERSISTENT=1
|
|||
### 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 on Intel GPUs.
|
||||
|
||||
```
|
||||
```bash
|
||||
# for Hugging Face model hub
|
||||
python ./generate.py --repo-id-or-model-path REPO_ID_OR_MODEL_PATH --prompt PROMPT --n-predict N_PREDICT
|
||||
|
||||
# for ModelScope model hub
|
||||
python ./generate.py --repo-id-or-model-path REPO_ID_OR_MODEL_PATH --prompt PROMPT --n-predict N_PREDICT --modelscope
|
||||
```
|
||||
|
||||
Arguments info:
|
||||
- `--repo-id-or-model-path REPO_ID_OR_MODEL_PATH`: argument defining the huggingface repo id for the GLM-4 model (e.g. `THUDM/glm-4-9b-chat`) to be downloaded, or the path to the huggingface checkpoint folder. It is default to be `'THUDM/glm-4-9b-chat'`.
|
||||
- `--repo-id-or-model-path REPO_ID_OR_MODEL_PATH`: argument defining the **Hugging Face** or **ModelScope** repo id for the GLM-4 model (e.g. `THUDM/glm-4-9b-chat`) to be downloaded, or the path to the checkpoint folder. It is default to be `'THUDM/glm-4-9b-chat'` for **Hugging Face** or `'ZhipuAI/glm-4-9b-chat'` for **ModelScope**.
|
||||
- `--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`.
|
||||
- `--modelscope`: using **ModelScope** as model hub instead of **Hugging Face**.
|
||||
|
||||
#### Sample Output
|
||||
#### [THUDM/glm-4-9b-chat](https://huggingface.co/THUDM/glm-4-9b-chat)
|
||||
|
|
@ -134,21 +145,3 @@ 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.
|
||||
|
||||
**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 `"AI是什么？"`.
|
||||
- `--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.
|
||||
|
|
|
|||
|
|
@ -20,7 +20,6 @@ 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
|
||||
|
|
@ -28,16 +27,27 @@ 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('--repo-id-or-model-path', type=str,
|
||||
help='The Hugging Face or ModelScope repo id for GLM-4 model model to be downloaded'
|
||||
', or the path to the 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')
|
||||
parser.add_argument('--modelscope', action="store_true", default=False,
|
||||
help="Use models from modelscope")
|
||||
|
||||
args = parser.parse_args()
|
||||
model_path = args.repo_id_or_model_path
|
||||
|
||||
if args.modelscope:
|
||||
from modelscope import AutoTokenizer
|
||||
model_hub = 'modelscope'
|
||||
else:
|
||||
from transformers import AutoTokenizer
|
||||
model_hub = 'huggingface'
|
||||
|
||||
model_path = args.repo_id_or_model_path if args.repo_id_or_model_path else \
|
||||
("ZhipuAI/glm-4-9b-chat" if args.modelscope else "THUDM/glm-4-9b-chat")
|
||||
|
||||
# Load model in 4 bit,
|
||||
# which convert the relevant layers in the model into INT4 format
|
||||
|
|
@ -47,8 +57,9 @@ if __name__ == '__main__':
|
|||
load_in_4bit=True,
|
||||
optimize_model=True,
|
||||
trust_remote_code=True,
|
||||
use_cache=True)
|
||||
model = model.to("xpu")
|
||||
use_cache=True,
|
||||
model_hub=model_hub)
|
||||
model = model.half().to("xpu")
|
||||
|
||||
# Load tokenizer
|
||||
tokenizer = AutoTokenizer.from_pretrained(model_path,
|
||||
|
|
|
|||
|
|
@ -1,69 +0,0 @@
|
|||
#
|
||||
# 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
|
||||
# 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 = AutoModel.from_pretrained(model_path,
|
||||
trust_remote_code=True,
|
||||
load_in_4bit=True,
|
||||
optimize_model=True,
|
||||
use_cache=True,
|
||||
cpu_embedding=True)
|
||||
|
||||
model = model.to('xpu')
|
||||
|
||||
# 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
|
||||
|
|
@ -1,5 +1,5 @@
|
|||
# Qwen2.5
|
||||
In this directory, you will find examples on how you could apply IPEX-LLM INT4 optimizations on Qwen2.5 models on [Intel GPUs](../../../README.md). For illustration purposes, we utilize [Qwen/Qwen2.5-3B-Instruct](https://huggingface.co/Qwen/Qwen2.5-3B-Instruct), [Qwen/Qwen2.5-7B-Instruct](https://huggingface.co/Qwen/Qwen2.5-7B-Instruct) and [Qwen/Qwen2.5-14B-Instruct](https://huggingface.co/Qwen/Qwen2.5-14B-Instruct) as reference Qwen2.5 models.
|
||||
In this directory, you will find examples on how you could apply IPEX-LLM INT4 optimizations on Qwen2.5 models on [Intel GPUs](../../../README.md). For illustration purposes, we utilize [Qwen/Qwen2.5-3B-Instruct](https://huggingface.co/Qwen/Qwen2.5-3B-Instruct), [Qwen/Qwen2.5-7B-Instruct](https://huggingface.co/Qwen/Qwen2.5-7B-Instruct) and [Qwen/Qwen2.5-14B-Instruct](https://huggingface.co/Qwen/Qwen2.5-14B-Instruct) (or [Qwen/Qwen2.5-3B-Instruct](https://www.modelscope.cn/models/Qwen/Qwen2.5-3B-Instruct), [Qwen/Qwen2.5-7B-Instruct](https://www.modelscope.cn/models/Qwen/Qwen2.5-7B-Instruct) and [Qwen/Qwen2.5-14B-Instruct](https://www.modelscope.cn/models/Qwen/Qwen2.5-14B-Instruct) for ModelScope) as reference Qwen2.5 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.
|
||||
|
|
@ -14,6 +14,9 @@ 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/
|
||||
|
||||
# [optional] only needed if you would like to use ModelScope as model hub
|
||||
pip install modelscope==1.11.0
|
||||
```
|
||||
|
||||
#### 1.2 Installation on Windows
|
||||
|
|
@ -24,6 +27,9 @@ 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/
|
||||
|
||||
# [optional] only needed if you would like to use ModelScope as model hub
|
||||
pip install modelscope==1.11.0
|
||||
```
|
||||
|
||||
### 2. Configures OneAPI environment variables for Linux
|
||||
|
|
@ -91,14 +97,19 @@ set SYCL_CACHE_PERSISTENT=1
|
|||
> 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
|
||||
# for Hugging Face model hub
|
||||
python ./generate.py --repo-id-or-model-path REPO_ID_OR_MODEL_PATH --prompt PROMPT --n-predict N_PREDICT
|
||||
|
||||
# for ModelScope model hub
|
||||
python ./generate.py --repo-id-or-model-path REPO_ID_OR_MODEL_PATH --prompt PROMPT --n-predict N_PREDICT --modelscope
|
||||
```
|
||||
|
||||
Arguments info:
|
||||
- `--repo-id-or-model-path REPO_ID_OR_MODEL_PATH`: argument defining the huggingface repo id for the Qwen2.5 model (e.g. `Qwen/Qwen2.5-7B-Instruct`) to be downloaded, or the path to the huggingface checkpoint folder. It is default to be `'Qwen/Qwen2.5-7B-Instruct'`.
|
||||
- `--repo-id-or-model-path REPO_ID_OR_MODEL_PATH`: argument defining the **Hugging Face** or **ModelScope** repo id for the Qwen2.5 model (e.g. `Qwen/Qwen2.5-7B-Instruct`) to be downloaded, or the path to the checkpoint folder. It is default to be `'Qwen/Qwen2.5-7B-Instruct'`.
|
||||
- `--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`.
|
||||
- `--modelscope`: using **ModelScope** as model hub instead of **Hugging Face**.
|
||||
|
||||
#### Sample Output
|
||||
##### [Qwen/Qwen2.5-3B-Instruct](https://huggingface.co/Qwen/Qwen2.5-3B-Instruct)
|
||||
|
|
|
|||
|
|
@ -18,20 +18,29 @@ import torch
|
|||
import time
|
||||
import argparse
|
||||
|
||||
from transformers import AutoTokenizer
|
||||
|
||||
|
||||
if __name__ == '__main__':
|
||||
parser = argparse.ArgumentParser(description='Predict Tokens using generate() API for Qwen2.5 model')
|
||||
parser.add_argument('--repo-id-or-model-path', type=str, default="Qwen/Qwen2.5-7B-Instruct",
|
||||
help='The huggingface repo id for the Qwen2.5 model to be downloaded'
|
||||
help='The Hugging Face or ModelScope repo id for the Qwen2.5 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')
|
||||
parser.add_argument('--modelscope', action="store_true", default=False,
|
||||
help="Use models from modelscope")
|
||||
|
||||
args = parser.parse_args()
|
||||
|
||||
if args.modelscope:
|
||||
from modelscope import AutoTokenizer
|
||||
model_hub = 'modelscope'
|
||||
else:
|
||||
from transformers import AutoTokenizer
|
||||
model_hub = 'huggingface'
|
||||
|
||||
model_path = args.repo_id_or_model_path
|
||||
|
||||
|
||||
|
|
@ -42,7 +51,8 @@ if __name__ == '__main__':
|
|||
load_in_4bit=True,
|
||||
optimize_model=True,
|
||||
trust_remote_code=True,
|
||||
use_cache=True)
|
||||
use_cache=True,
|
||||
model_hub=model_hub)
|
||||
model = model.half().to("xpu")
|
||||
|
||||
# Load tokenizer
|
||||
|
|
|
|||
|
|
@ -1,5 +1,5 @@
|
|||
# Qwen2
|
||||
In this directory, you will find examples on how you could apply IPEX-LLM INT4 optimizations on Qwen2 models on [Intel GPUs](../../../README.md). For illustration purposes, we utilize [Qwen/Qwen2-7B-Instruct](https://huggingface.co/Qwen/Qwen2-7B-Instruct) and [Qwen/Qwen2-1.5B-Instruct](https://huggingface.co/Qwen/Qwen2-1.5B-Instruct) as reference Qwen2 models.
|
||||
In this directory, you will find examples on how you could apply IPEX-LLM INT4 optimizations on Qwen2 models on [Intel GPUs](../../../README.md). For illustration purposes, we utilize [Qwen/Qwen2-7B-Instruct](https://huggingface.co/Qwen/Qwen2-7B-Instruct) and [Qwen/Qwen2-1.5B-Instruct](https://huggingface.co/Qwen/Qwen2-1.5B-Instruct) (or [Qwen/Qwen2-7B-Instruct](https://www.modelscope.cn/models/Qwen/Qwen2-7B-Instruct) and [Qwen/Qwen2-1.5B-Instruct](https://www.modelscope.cn/models/Qwen/Qwen2-1.5B-Instruct) for ModelScope) as reference Qwen2 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.
|
||||
|
|
@ -16,6 +16,9 @@ conda activate llm
|
|||
pip install --pre --upgrade ipex-llm[xpu] --extra-index-url https://pytorch-extension.intel.com/release-whl/stable/xpu/us/
|
||||
|
||||
pip install transformers==4.37.0 # install transformers which supports Qwen2
|
||||
|
||||
# [optional] only needed if you would like to use ModelScope as model hub
|
||||
pip install modelscope==1.11.0
|
||||
```
|
||||
|
||||
#### 1.2 Installation on Windows
|
||||
|
|
@ -28,6 +31,9 @@ conda activate llm
|
|||
pip install --pre --upgrade ipex-llm[xpu] --extra-index-url https://pytorch-extension.intel.com/release-whl/stable/xpu/us/
|
||||
|
||||
pip install transformers==4.37.0 # install transformers which supports Qwen2
|
||||
|
||||
# [optional] only needed if you would like to use ModelScope as model hub
|
||||
pip install modelscope==1.11.0
|
||||
```
|
||||
|
||||
### 2. Configures OneAPI environment variables for Linux
|
||||
|
|
@ -95,14 +101,19 @@ set SYCL_CACHE_PERSISTENT=1
|
|||
> 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
|
||||
# for Hugging Face model hub
|
||||
python ./generate.py --repo-id-or-model-path REPO_ID_OR_MODEL_PATH --prompt PROMPT --n-predict N_PREDICT
|
||||
|
||||
# for ModelScope model hub
|
||||
python ./generate.py --repo-id-or-model-path REPO_ID_OR_MODEL_PATH --prompt PROMPT --n-predict N_PREDICT --modelscope
|
||||
```
|
||||
|
||||
Arguments info:
|
||||
- `--repo-id-or-model-path REPO_ID_OR_MODEL_PATH`: argument defining the huggingface repo id for the Qwen2 model (e.g. `Qwen/Qwen2-7B-Instruct`) to be downloaded, or the path to the huggingface checkpoint folder. It is default to be `'Qwen/Qwen2-7B-Instruct'`.
|
||||
- `--repo-id-or-model-path REPO_ID_OR_MODEL_PATH`: argument defining the **Hugging Face** or **ModelScope** repo id for the Qwen2 model (e.g. `Qwen/Qwen2-7B-Instruct`) to be downloaded, or the path to the checkpoint folder. It is default to be `'Qwen/Qwen2-7B-Instruct'`.
|
||||
- `--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`.
|
||||
- `--modelscope`: using **ModelScope** as model hub instead of **Hugging Face**.
|
||||
|
||||
#### Sample Output
|
||||
##### [Qwen/Qwen2-7B-Instruct](https://huggingface.co/Qwen/Qwen2-7B-Instruct)
|
||||
|
|
|
|||
|
|
@ -18,21 +18,30 @@ import torch
|
|||
import time
|
||||
import argparse
|
||||
|
||||
from transformers import AutoTokenizer
|
||||
import numpy as np
|
||||
|
||||
|
||||
if __name__ == '__main__':
|
||||
parser = argparse.ArgumentParser(description='Qwen2-7B-Instruct')
|
||||
parser = argparse.ArgumentParser(description='Predict Tokens using `generate()` API for Qwen2 model')
|
||||
parser.add_argument('--repo-id-or-model-path', type=str, default="Qwen/Qwen2-7B-Instruct",
|
||||
help='The huggingface repo id for the Qwen2 model to be downloaded'
|
||||
', or the path to the huggingface checkpoint folder')
|
||||
help='The Hugging Face or ModelScope repo id for the Qwen2 model to be downloaded'
|
||||
', or the path to the 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')
|
||||
parser.add_argument('--modelscope', action="store_true", default=False,
|
||||
help="Use models from modelscope")
|
||||
|
||||
args = parser.parse_args()
|
||||
|
||||
if args.modelscope:
|
||||
from modelscope import AutoTokenizer
|
||||
model_hub = 'modelscope'
|
||||
else:
|
||||
from transformers import AutoTokenizer
|
||||
model_hub = 'huggingface'
|
||||
|
||||
model_path = args.repo_id_or_model_path
|
||||
|
||||
|
||||
|
|
@ -43,7 +52,8 @@ if __name__ == '__main__':
|
|||
load_in_4bit=True,
|
||||
optimize_model=True,
|
||||
trust_remote_code=True,
|
||||
use_cache=True)
|
||||
use_cache=True,
|
||||
model_hub=model_hub)
|
||||
model = model.half().to("xpu")
|
||||
|
||||
# Load tokenizer
|
||||
|
|
|
|||
Loading…
Reference in a new issue