Skip to content

GitLab

Commit 1bfbcff0 authored by wanglch's avatar wanglch
Browse files

Initial commit

parents
Pipeline #1204 canceled with stages
Show whitespace changes
Inline Side-by-side
Showing with 2903 additions and 0 deletions
.github/ISSUE_TEMPLATE/bug_report.yaml 0 → 100644
View file @ 1bfbcff0
name: "\U0001F41B Bug Report"
description: Submit a bug report to help us improve GLM-4-9B / 提交一个 Bug 问题报告来帮助我们改进 GLM-4-9B
body:
- type: textarea
id: system-info
attributes:
label: System Info / 系統信息
description: Your operating environment / 您的运行环境信息
placeholder: Includes Cuda version, Transformers version, Python version, operating system, hardware information (if you suspect a hardware problem)... / 包括Cuda版本，Transformers版本，Python版本，操作系统，硬件信息(如果您怀疑是硬件方面的问题)...
validations:
required: true
- type: textarea
id: who-can-help
attributes:
label: Who can help? / 谁可以帮助到您？
description: |
Your issue will be replied to more quickly if you can figure out the right person to tag with @
All issues are read by one of the maintainers, so if you don't know who to tag, just leave this blank and our maintainer will ping the right person.
Please tag fewer than 3 people.
如果您能找到合适的标签 @，您的问题会更快得到回复。
所有问题都会由我们的维护者阅读，如果您不知道该标记谁，只需留空，我们的维护人员会找到合适的开发组成员来解决问题。
标记的人数应该不超过 3 个人。
If it's not a bug in these three subsections, you may not specify the helper. Our maintainer will find the right person in the development group to solve the problem.
如果不是这三个子版块的bug，您可以不指明帮助者，我们的维护人员会找到合适的开发组成员来解决问题。
placeholder: "@Username ..."
- type: checkboxes
id: information-scripts-examples
attributes:
label: Information / 问题信息
description: 'The problem arises when using: / 问题出现在'
options:
- label: "The official example scripts / 官方的示例脚本"
- label: "My own modified scripts / 我自己修改的脚本和任务"
- type: textarea
id: reproduction
validations:
required: true
attributes:
label: Reproduction / 复现过程
description: |
Please provide a code example that reproduces the problem you encountered, preferably with a minimal reproduction unit.
If you have code snippets, error messages, stack traces, please provide them here as well.
Please format your code correctly using code tags. See https://help.github.com/en/github/writing-on-github/creating-and-highlighting-code-blocks#syntax-highlighting
Do not use screenshots, as they are difficult to read and (more importantly) do not allow others to copy and paste your code.
请提供能重现您遇到的问题的代码示例,最好是最小复现单元。
如果您有代码片段、错误信息、堆栈跟踪，也请在此提供。
请使用代码标签正确格式化您的代码。请参见 https://help.github.com/en/github/writing-on-github/creating-and-highlighting-code-blocks#syntax-highlighting
请勿使用截图，因为截图难以阅读，而且（更重要的是）不允许他人复制粘贴您的代码。
placeholder: |
Steps to reproduce the behavior/复现Bug的步骤:
1.
2.
3.
- type: textarea
id: expected-behavior
validations:
required: true
attributes:
label: Expected behavior / 期待表现
description: "A clear and concise description of what you would expect to happen. /简单描述您期望发生的事情。"
\ No newline at end of file
.github/ISSUE_TEMPLATE/feature-request.yaml 0 → 100644
View file @ 1bfbcff0
name: "\U0001F680 Feature request"
description: Submit a request for a new GLM-4-9B feature / 提交一个新的 GLM-4-9B 的功能建议
labels: [ "feature" ]
body:
- type: textarea
id: feature-request
validations:
required: true
attributes:
label: Feature request / 功能建议
description: |
A brief description of the functional proposal. Links to corresponding papers and code are desirable.
对功能建议的简述。最好提供对应的论文和代码链接
- type: textarea
id: motivation
validations:
required: true
attributes:
label: Motivation / 动机
description: |
Your motivation for making the suggestion. If that motivation is related to another GitHub issue, link to it here.
您提出建议的动机。如果该动机与另一个 GitHub 问题有关，请在此处提供对应的链接。
- type: textarea
id: contribution
validations:
required: true
attributes:
label: Your contribution / 您的贡献
description: |
Your PR link or any other link you can help with.
您的PR链接或者其他您能提供帮助的链接。
\ No newline at end of file
.github/PULL_REQUEST_TEMPLATE/pr_template.md 0 → 100644
View file @ 1bfbcff0
# Raise valuable PR / 提出有价值的PR
## Caution/ 注意事项:
Users should keep the following points in mind when submitting PRs:
1. The proposed PR should be about this project.
2. the proposed PR should be relevant, if there are multiple ideas and optimizations, they should be assigned to different PRs.
用户在提交PR时候应该注意以下几点:
1. 提出的PR应该是关于本项目的。
2. 提出的PR应该具有针对性，如果具有多个不同的想法和优化方案，应该分配到不同的PR中。
## 不应该提出的PR / PRs that should not be proposed
If a developer proposes a PR about any of the following, it may be closed or Rejected.
1. those that don't describe improvement options.
2. multiple issues of different types combined in one PR.
3. The proposed PR is highly duplicative of already existing PRs.
如果开发者提出关于以下方面的PR，则可能会被直接关闭或拒绝通过。
1. 没有说明改进方案的。
2. 多个不同类型的问题合并在一个PR中的。
3. 提出的PR与已经存在的PR高度重复的。
# 检查您的PR
- [ ] Have you read the Contributor Guidelines, Pull Request section? / 您是否阅读了贡献者指南、Pull Request 部分？
- [ ] Has this been discussed/approved via a Github issue or forum? If so, add a link. / 是否通过 Github 问题或论坛讨论/批准过？如果是，请添加链接。
- [ ] Did you make sure you updated the documentation with your changes? Here are the Documentation Guidelines, and here are the Documentation Formatting Tips. /您是否确保根据您的更改更新了文档？这里是文档指南，这里是文档格式化技巧。
- [ ] Did you write new required tests? / 您是否编写了新的必要测试？
- [ ] Are your PRs for only one issue / 您的PR是否仅针对一个问题
\ No newline at end of file
.gitignore 0 → 100644
View file @ 1bfbcff0
*venv
*.DS_Store
*.idea/
LICENSE 0 → 100644
View file @ 1bfbcff0
Apache License
Version 2.0, January 2004
http://www.apache.org/licenses/
TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
1. Definitions.
"License" shall mean the terms and conditions for use, reproduction,
and distribution as defined by Sections 1 through 9 of this document.
"Licensor" shall mean the copyright owner or entity authorized by
the copyright owner that is granting the License.
"Legal Entity" shall mean the union of the acting entity and all
other entities that control, are controlled by, or are under common
control with that entity. For the purposes of this definition,
"control" means (i) the power, direct or indirect, to cause the
direction or management of such entity, whether by contract or
otherwise, or (ii) ownership of fifty percent (50%) or more of the
outstanding shares, or (iii) beneficial ownership of such entity.
"You" (or "Your") shall mean an individual or Legal Entity
exercising permissions granted by this License.
"Source" form shall mean the preferred form for making modifications,
including but not limited to software source code, documentation
source, and configuration files.
"Object" form shall mean any form resulting from mechanical
transformation or translation of a Source form, including but
not limited to compiled object code, generated documentation,
and conversions to other media types.
"Work" shall mean the work of authorship, whether in Source or
Object form, made available under the License, as indicated by a
copyright notice that is included in or attached to the work
(an example is provided in the Appendix below).
"Derivative Works" shall mean any work, whether in Source or Object
form, that is based on (or derived from) the Work and for which the
editorial revisions, annotations, elaborations, or other modifications
represent, as a whole, an original work of authorship. For the purposes
of this License, Derivative Works shall not include works that remain
separable from, or merely link (or bind by name) to the interfaces of,
the Work and Derivative Works thereof.
"Contribution" shall mean any work of authorship, including
the original version of the Work and any modifications or additions
to that Work or Derivative Works thereof, that is intentionally
submitted to Licensor for inclusion in the Work by the copyright owner
or by an individual or Legal Entity authorized to submit on behalf of
the copyright owner. For the purposes of this definition, "submitted"
means any form of electronic, verbal, or written communication sent
to the Licensor or its representatives, including but not limited to
communication on electronic mailing lists, source code control systems,
and issue tracking systems that are managed by, or on behalf of, the
Licensor for the purpose of discussing and improving the Work, but
excluding communication that is conspicuously marked or otherwise
designated in writing by the copyright owner as "Not a Contribution."
"Contributor" shall mean Licensor and any individual or Legal Entity
on behalf of whom a Contribution has been received by Licensor and
subsequently incorporated within the Work.
2. Grant of Copyright License. Subject to the terms and conditions of
this License, each Contributor hereby grants to You a perpetual,
worldwide, non-exclusive, no-charge, royalty-free, irrevocable
copyright license to reproduce, prepare Derivative Works of,
publicly display, publicly perform, sublicense, and distribute the
Work and such Derivative Works in Source or Object form.
3. Grant of Patent License. Subject to the terms and conditions of
this License, each Contributor hereby grants to You a perpetual,
worldwide, non-exclusive, no-charge, royalty-free, irrevocable
(except as stated in this section) patent license to make, have made,
use, offer to sell, sell, import, and otherwise transfer the Work,
where such license applies only to those patent claims licensable
by such Contributor that are necessarily infringed by their
Contribution(s) alone or by combination of their Contribution(s)
with the Work to which such Contribution(s) was submitted. If You
institute patent litigation against any entity (including a
cross-claim or counterclaim in a lawsuit) alleging that the Work
or a Contribution incorporated within the Work constitutes direct
or contributory patent infringement, then any patent licenses
granted to You under this License for that Work shall terminate
as of the date such litigation is filed.
4. Redistribution. You may reproduce and distribute copies of the
Work or Derivative Works thereof in any medium, with or without
modifications, and in Source or Object form, provided that You
meet the following conditions:
(a) You must give any other recipients of the Work or
Derivative Works a copy of this License; and
(b) You must cause any modified files to carry prominent notices
stating that You changed the files; and
(c) You must retain, in the Source form of any Derivative Works
that You distribute, all copyright, patent, trademark, and
attribution notices from the Source form of the Work,
excluding those notices that do not pertain to any part of
the Derivative Works; and
(d) If the Work includes a "NOTICE" text file as part of its
distribution, then any Derivative Works that You distribute must
include a readable copy of the attribution notices contained
within such NOTICE file, excluding those notices that do not
pertain to any part of the Derivative Works, in at least one
of the following places: within a NOTICE text file distributed
as part of the Derivative Works; within the Source form or
documentation, if provided along with the Derivative Works; or,
within a display generated by the Derivative Works, if and
wherever such third-party notices normally appear. The contents
of the NOTICE file are for informational purposes only and
do not modify the License. You may add Your own attribution
notices within Derivative Works that You distribute, alongside
or as an addendum to the NOTICE text from the Work, provided
that such additional attribution notices cannot be construed
as modifying the License.
You may add Your own copyright statement to Your modifications and
may provide additional or different license terms and conditions
for use, reproduction, or distribution of Your modifications, or
for any such Derivative Works as a whole, provided Your use,
reproduction, and distribution of the Work otherwise complies with
the conditions stated in this License.
5. Submission of Contributions. Unless You explicitly state otherwise,
any Contribution intentionally submitted for inclusion in the Work
by You to the Licensor shall be under the terms and conditions of
this License, without any additional terms or conditions.
Notwithstanding the above, nothing herein shall supersede or modify
the terms of any separate license agreement you may have executed
with Licensor regarding such Contributions.
6. Trademarks. This License does not grant permission to use the trade
names, trademarks, service marks, or product names of the Licensor,
except as required for reasonable and customary use in describing the
origin of the Work and reproducing the content of the NOTICE file.
7. Disclaimer of Warranty. Unless required by applicable law or
agreed to in writing, Licensor provides the Work (and each
Contributor provides its Contributions) on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
implied, including, without limitation, any warranties or conditions
of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
PARTICULAR PURPOSE. You are solely responsible for determining the
appropriateness of using or redistributing the Work and assume any
risks associated with Your exercise of permissions under this License.
8. Limitation of Liability. In no event and under no legal theory,
whether in tort (including negligence), contract, or otherwise,
unless required by applicable law (such as deliberate and grossly
negligent acts) or agreed to in writing, shall any Contributor be
liable to You for damages, including any direct, indirect, special,
incidental, or consequential damages of any character arising as a
result of this License or out of the use or inability to use the
Work (including but not limited to damages for loss of goodwill,
work stoppage, computer failure or malfunction, or any and all
other commercial damages or losses), even if such Contributor
has been advised of the possibility of such damages.
9. Accepting Warranty or Additional Liability. While redistributing
the Work or Derivative Works thereof, You may choose to offer,
and charge a fee for, acceptance of support, warranty, indemnity,
or other liability obligations and/or rights consistent with this
License. However, in accepting such obligations, You may act only
on Your own behalf and on Your sole responsibility, not on behalf
of any other Contributor, and only if You agree to indemnify,
defend, and hold each Contributor harmless for any liability
incurred by, or claims asserted against, such Contributor by reason
of your accepting any such warranty or additional liability.
END OF TERMS AND CONDITIONS
APPENDIX: How to apply the Apache License to your work.
To apply the Apache License to your work, attach the following
boilerplate notice, with the fields enclosed by brackets "[]"
replaced with your own identifying information. (Don't include
the brackets!) The text should be enclosed in the appropriate
comment syntax for the file format. We also recommend that a
file or class name and description of purpose be included on the
same "printed page" as the copyright notice for easier
identification within third-party archives.
Copyright 2024 GLM-4-9B Model Team @ Zhipu AI
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.
README.md 0 → 100644
View file @ 1bfbcff0
# GLM-4V
**GLM-4V-9B** 具备 1120 * 1120 高分辨率下的中英双语多轮对话能力，在中英文综合能力、感知推理、文字识别、图表理解等多方面多模态评测中，GLM-4V-9B 表现出超越 GPT-4-turbo-2024-04-09、Gemini 1.0 Pro、Qwen-VL-Max 和 Claude 3 Opus 的卓越性能。
## 论文
- [GLM: General Language Model Pretraining with Autoregressive Blank Infilling](https://arxiv.org/abs/2103.10360)
## 模型结构
GLM-4-9B 是智谱 AI 推出的最新一代预训练模型 GLM-4 系列中的开源版本。
<div align="center">
<img src="./images/GLM.png"/>
</div>
## 算法原理
在强化文本能力的同时，我们首次推出了基于GLM基座的开源多模态模型GLM-4V-9B。这一模型采用了与CogVLM2相似的架构设计，能够处理高达1120 x 1120分辨率的输入，并通过降采样技术有效减少了token的开销。为了减小部署与计算开销，GLM-4V-9B没有引入额外的视觉专家模块，采用了直接混合文本和图片数据的方式进行训练，在保持文本性能的同时提升多模态能力。
<div align=center>
<img src="./images/mt.png"/>
</div>
## 环境配置
### Docker（方法一）
[光源](https://www.sourcefind.cn/#/service-details)拉取docker镜像的地址与使用步骤
```
docker pull image.sourcefind.cn:5000/dcu/admin/base/pytorch:2.1.0-ubuntu22.04-dtk23.10.1-py310
docker run -it -v /path/your_code_data/:/path/your_code_data/ -v /opt/hyhal/:/opt/hyhal/:ro --shm-size=64G --privileged=true --device=/dev/kfd --device=/dev/dri/ --group-add video --name glm-4v <your imageID> bash
cd /path/your_code_data/
pip install -r requirements.txt -i http://mirrors.aliyun.com/pypi/simple/ --trusted-host mirrors.aliyun.com
```
### Dockerfile（方法二）
```
cd /path/your_code_data/docker
docker build --no-cache -t glm-4v:latest .
docker run --shm-size=64G --name qwen-vl -v /opt/hyhal:/opt/hyhal:ro --privileged=true --device=/dev/kfd --device=/dev/dri/ --group-add video -v /path/your_code_data/:/path/your_code_data/ -it glm-4v bash
```
### Anaconda（方法三）
关于本项目DCU显卡所需的特殊深度学习库可从[光合](https://developer.hpccube.com/tool/)开发者社区下载安装。
```
DTK驱动：dtk23.10
python：python3.10
torch:2.1
torchvision: 0.16.0
```
`Tips：以上dtk驱动、python、paddle等DCU相关工具版本需要严格一一对应`
关于本项目DCU显卡所需的特殊深度学习库可从[光合](https://developer.hpccube.com/tool/)开发者社区下载安装。
```
conda create -n minicpm-v python=3.10
conda activate minicpm-v
cd /path/your_code_data/
pip install -r requirements.txt -i http://mirrors.aliyun.com/pypi/simple
```
## 数据集
迷你数据集 [coco测试数据集](./data/train.json)
[coco图像描述完整数据集](https://modelscope.cn/datasets/modelscope/coco_2014_caption/summary?spm=a2c6h.12873639.article-detail.17.70117774diTYiv)
本仓库提供测试数据集用于微调代码测试，需要可自行下载。预训练需要准备你的训练数据，需要将所有样本放到一个列表中并存入json文件中。自定义数据集支持json和jsonl样式。glm-4v-9b支持多轮对话，但总的对话轮次中需包含一张图片，支持传入本地路径或URL。以下是自定义数据集的示例：。用于正常训练的完整数据集请按此目录结构进行制备：
```
{"query": "55555", "response": "66666", "images": ["image_path"]}
{"query": "eeeee", "response": "fffff", "history": [], "images": ["image_path"]}
{"query": "EEEEE", "response": "FFFFF", "history": [["AAAAA", "BBBBB"], ["CCCCC", "DDDDD"]], "images": ["image_path"]}
```
## 训练
微调需要安装swift，具体操作如下：
```
git clone https://github.com/modelscope/swift.git
cd swift
pip install -e .
```
训练需在[finetune_lora.sh](finetune_lora.sh)中修改以下参数，其他参数可自行修改:
```
--model_id_or_path # or 修改为本地模型地址
--dataset # 训练集文件夹
--output_dir # 训练输出文件夹
```
### 单卡训练
```
sh lora_finetune_single.sh
```
### 多卡训练
```
sh lora_finetune_multi.sh
```
## 推理
若要执行推理需要将本仓库修改后的[visual.py](visual.py)替换模型文件中的**visual.py**文件，并替换模型路径。
`tokenizer = AutoTokenizer.from_pretrained("/home/wanglch/projects/GLM-4V/glm-4v-b", trust_remote_code=True)`
`model = AutoModelForCausalLM.from_pretrained(
"/home/wanglch/projects/GLM-4V/glm-4v-b",
torch_dtype=torch.float16,
low_cpu_mem_usage=True,
trust_remote_code=True
).to(device).eval()`
### 单机单卡
```
python glm-4v_inference.py
```
## result
### OCR
<div align=center>
<img src="./images/result.png"/>
</div>
### 精度
测试数据： [迷你coco图像描述数据集](./data/train.json) ，使用的加速卡:A800/K100。
| device | train_loss | eval_loss |eval_loss |
| :------: | :------: | :------: | :------: |
| A800*2 | 2.493 | 4.25 | 0.302 |
| K100*2 | 2.495 | 4.25 | 0.295 |
## 应用场景
### 算法类别
`ocr`
### 热点应用行业
`金融,教育,政府,科研,制造,能源,交通`
## 预训练权重
- [THUDM/glm-4v-9b](https://huggingface.co/THUDM/glm-4v-9b)
- [ZhipuAI/glm-4v-9b](https://modelscope.cn/models/ZhipuAI/glm-4-9b-chat-1m)
## 源码仓库及问题反馈
- http://developer.hpccube.com/codes/modelzoo/glm-4v-9b_pytorch.git
## 参考资料
- [GLM: General Language Model Pretraining with Autoregressive Blank Infilling](https://arxiv.org/abs/2103.10360)
- [GLM4v github](https://github.com/THUDM/GLM-4)
- [swift github](https://github.com/modelscope/swift/blob/main/docs/source/Multi-Modal/glm4v%E6%9C%80%E4%BD%B3%E5%AE%9E%E8%B7%B5.md)
README_base.md 0 → 100644
View file @ 1bfbcff0
# GLM-4
<p align="center">
🤗 <a href="https://huggingface.co/collections/THUDM/glm-4-665fcf188c414b03c2f7e3b7" target="_blank">HF Repo</a> • 🤖 <a href="https://modelscope.cn/models/ZhipuAI/glm-4-9b-chat" target="_blank">ModelScope</a> • 🐦 <a href="https://twitter.com/thukeg" target="_blank">Twitter</a> • 👋 加入我们的 <a href="https://discord.gg/fK2dz4bg" target="_blank">Discord</a> 和 <a href="resources/WECHAT.md" target="_blank">微信</a>
</p>
<p align="center">
📍在 <a href="https://open.bigmodel.cn/?utm_campaign=open&_channel_track_key=OWTVNma9">智谱AI开放平台</a> 体验和使用更大规模的 GLM 商业模型。
</p>
Read this in [English](README_en.md)
## 模型介绍
GLM-4-9B 是智谱 AI 推出的最新一代预训练模型 GLM-4 系列中的开源版本。 在语义、数学、推理、代码和知识等多方面的数据集测评中，
**GLM-4-9B** 及其人类偏好对齐的版本 **GLM-4-9B-Chat** 均表现出超越 Llama-3-8B 的卓越性能。除了能进行多轮对话，GLM-4-9B-Chat
还具备网页浏览、代码执行、自定义工具调用（Function Call）和长文本推理（支持最大 128K 上下文）等高级功能。本代模型增加了多语言支持，支持包括日语，韩语，德语在内的
26 种语言。我们还推出了支持 1M 上下文长度（约 200 万中文字符）的 **GLM-4-9B-Chat-1M** 模型和基于 GLM-4-9B 的多模态模型
GLM-4V-9B。**GLM-4V-9B** 具备 1120 * 1120 高分辨率下的中英双语多轮对话能力，在中英文综合能力、感知推理、文字识别、图表理解等多方面多模态评测中，GLM-4V-9B
表现出超越 GPT-4-turbo-2024-04-09、Gemini
1.0 Pro、Qwen-VL-Max 和 Claude 3 Opus 的卓越性能。
## Model List
| Model | Type | Seq Length | Download | Online Demo |
|------------------|------|------------|-----------------------------------------------------------------------------------------------------------------------------------------|--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
| GLM-4-9B | Base | 8K | [🤗 Huggingface](https://huggingface.co/THUDM/glm-4-9b) [🤖 ModelScope](https://modelscope.cn/models/ZhipuAI/glm-4-9b) | / |
| GLM-4-9B-Chat | Chat | 128K | [🤗 Huggingface](https://huggingface.co/THUDM/glm-4-9b-chat) [🤖 ModelScope](https://modelscope.cn/models/ZhipuAI/glm-4-9b-chat) | [🤖 ModelScope CPU](https://modelscope.cn/studios/dash-infer/GLM-4-Chat-DashInfer-Demo/summary)<br> [🤖 ModelScope vLLM](https://modelscope.cn/studios/ZhipuAI/glm-4-9b-chat-vllm/summary) |
| GLM-4-9B-Chat-1M | Chat | 1M | [🤗 Huggingface](https://huggingface.co/THUDM/glm-4-9b-chat-1m) [🤖 ModelScope](https://modelscope.cn/models/ZhipuAI/glm-4-9b-chat-1m) | / |
| GLM-4V-9B | Chat | 8K | [🤗 Huggingface](https://huggingface.co/THUDM/glm-4v-9b) [🤖 ModelScope](https://modelscope.cn/models/ZhipuAI/glm-4v-9b) | / |
## 评测结果
### 对话模型典型任务
| Model | AlignBench | MT-Bench | IFEval | MMLU | C-Eval | GSM8K | MATH | HumanEval | NaturalCodeBench |
|:--------------------|:----------:|:--------:|:------:|:----:|:------:|:-----:|:----:|:---------:|:----------------:|
| Llama-3-8B-Instruct | 6.40 | 8.00 | 68.6 | 68.4 | 51.3 | 79.6 | 30.0 | 62.2 | 24.7 |
| ChatGLM3-6B | 5.18 | 5.50 | 28.1 | 61.4 | 69.0 | 72.3 | 25.7 | 58.5 | 11.3 |
| GLM-4-9B-Chat | 7.01 | 8.35 | 69.0 | 72.4 | 75.6 | 79.6 | 50.6 | 71.8 | 32.2 |
### 基座模型典型任务
| Model | MMLU | C-Eval | GPQA | GSM8K | MATH | HumanEval |
|:--------------------|:----:|:------:|:----:|:-----:|:----:|:---------:|
| Llama-3-8B | 66.6 | 51.2 | - | 45.8 | - | 33.5 |
| Llama-3-8B-Instruct | 68.4 | 51.3 | 34.2 | 79.6 | 30.0 | 62.2 |
| ChatGLM3-6B-Base | 61.4 | 69.0 | 26.8 | 72.3 | 25.7 | 58.5 |
| GLM-4-9B | 74.7 | 77.1 | 34.3 | 84.0 | 30.4 | 70.1 |
> 由于 `GLM-4-9B` 在预训练过程中加入了部分数学、推理、代码相关的 instruction 数据，所以将 Llama-3-8B-Instruct 也列入比较范围。
### 长文本
在 1M 的上下文长度下进行[大海捞针实验](https://github.com/LargeWorldModel/LWM/blob/main/scripts/eval_needle.py)，结果如下：
![needle](resources/eval_needle.jpeg)
在 LongBench-Chat 上对长文本能力进行了进一步评测，结果如下:
<p align="center">
<img src="resources/longbench.png" alt="描述文字" style="display: block; margin: auto; width: 65%;">
</p>
### 多语言能力
在六个多语言数据集上对 GLM-4-9B-Chat 和 Llama-3-8B-Instruct 进行了测试，测试结果及数据集对应选取语言如下表
| Dataset | Llama-3-8B-Instruct | GLM-4-9B-Chat | Languages |
|:------------|:-------------------:|:-------------:|:----------------------------------------------------------------------------------------------:|
| M-MMLU | 49.6 | 56.6 | all |
| FLORES | 25.0 | 28.8 | ru, es, de, fr, it, pt, pl, ja, nl, ar, tr, cs, vi, fa, hu, el, ro, sv, uk, fi, ko, da, bg, no |
| MGSM | 54.0 | 65.3 | zh, en, bn, de, es, fr, ja, ru, sw, te, th |
| XWinograd | 61.7 | 73.1 | zh, en, fr, jp, ru, pt |
| XStoryCloze | 84.7 | 90.7 | zh, en, ar, es, eu, hi, id, my, ru, sw, te |
| XCOPA | 73.3 | 80.1 | zh, et, ht, id, it, qu, sw, ta, th, tr, vi |
### 工具调用能力
我们在 [Berkeley Function Calling Leaderboard](https://github.com/ShishirPatil/gorilla/tree/main/berkeley-function-call-leaderboard)
上进行了测试并得到了以下结果：
| Model | Overall Acc. | AST Summary | Exec Summary | Relevance |
|:-----------------------|:------------:|:-----------:|:------------:|:---------:|
| Llama-3-8B-Instruct | 58.88 | 59.25 | 70.01 | 45.83 |
| gpt-4-turbo-2024-04-09 | 81.24 | 82.14 | 78.61 | 88.75 |
| ChatGLM3-6B | 57.88 | 62.18 | 69.78 | 5.42 |
| GLM-4-9B-Chat | 81.00 | 80.26 | 84.40 | 87.92 |
### 多模态能力
GLM-4V-9B 是一个多模态语言模型，具备视觉理解能力，其相关经典任务的评测结果如下：
| | **MMBench-EN-Test** | **MMBench-CN-Test** | **SEEDBench_IMG** | **MMStar** | **MMMU** | **MME** | **HallusionBench** | **AI2D** | **OCRBench** |
|----------------------------|---------------------|---------------------|-------------------|------------|----------|---------|--------------------|----------|--------------|
| **gpt-4o-2024-05-13** | 83.4 | 82.1 | 77.1 | 63.9 | 69.2 | 2310.3 | 55.0 | 84.6 | 736 |
| **gpt-4-turbo-2024-04-09** | 81.0 | 80.2 | 73.0 | 56.0 | 61.7 | 2070.2 | 43.9 | 78.6 | 656 |
| **gpt-4-1106-preview** | 77.0 | 74.4 | 72.3 | 49.7 | 53.8 | 1771.5 | 46.5 | 75.9 | 516 |
| **InternVL-Chat-V1.5** | 82.3 | 80.7 | 75.2 | 57.1 | 46.8 | 2189.6 | 47.4 | 80.6 | 720 |
| **LLaVA-Next-Yi-34B** | 81.1 | 79.0 | 75.7 | 51.6 | 48.8 | 2050.2 | 34.8 | 78.9 | 574 |
| **Step-1V** | 80.7 | 79.9 | 70.3 | 50.0 | 49.9 | 2206.4 | 48.4 | 79.2 | 625 |
| **MiniCPM-Llama3-V2.5** | 77.6 | 73.8 | 72.3 | 51.8 | 45.8 | 2024.6 | 42.4 | 78.4 | 725 |
| **Qwen-VL-Max** | 77.6 | 75.7 | 72.7 | 49.5 | 52.0 | 2281.7 | 41.2 | 75.7 | 684 |
| **Gemini 1.0 Pro** | 73.6 | 74.3 | 70.7 | 38.6 | 49.0 | 2148.9 | 45.7 | 72.9 | 680 |
| **Claude 3 Opus** | 63.3 | 59.2 | 64.0 | 45.7 | 54.9 | 1586.8 | 37.8 | 70.6 | 694 |
| **GLM-4V-9B** | 81.1 | 79.4 | 76.8 | 58.7 | 47.2 | 2163.8 | 46.6 | 81.1 | 786 |
## 快速调用
**硬件配置和系统要求，请查看[这里](basic_demo/README.md)。**
### 使用以下方法快速调用 GLM-4-9B-Chat 语言模型
使用 transformers 后端进行推理:
```python
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
device = "cuda"
tokenizer = AutoTokenizer.from_pretrained("THUDM/glm-4-9b-chat", trust_remote_code=True)
query = "你好"
inputs = tokenizer.apply_chat_template([{"role": "user", "content": query}],
add_generation_prompt=True,
tokenize=True,
return_tensors="pt",
return_dict=True
)
inputs = inputs.to(device)
model = AutoModelForCausalLM.from_pretrained(
"THUDM/glm-4-9b-chat",
torch_dtype=torch.bfloat16,
low_cpu_mem_usage=True,
trust_remote_code=True
).to(device).eval()
gen_kwargs = {"max_length": 2500, "do_sample": True, "top_k": 1}
with torch.no_grad():
outputs = model.generate(**inputs, **gen_kwargs)
outputs = outputs[:, inputs['input_ids'].shape[1]:]
print(tokenizer.decode(outputs[0], skip_special_tokens=True))
```
使用 vLLM 后端进行推理:
```python
from transformers import AutoTokenizer
from vllm import LLM, SamplingParams
# GLM-4-9B-Chat-1M
# max_model_len, tp_size = 1048576, 4
# 如果遇见 OOM 现象，建议减少max_model_len，或者增加tp_size
max_model_len, tp_size = 131072, 1
model_name = "THUDM/glm-4-9b-chat"
prompt = [{"role": "user", "content": "你好"}]
tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True)
llm = LLM(
model=model_name,
tensor_parallel_size=tp_size,
max_model_len=max_model_len,
trust_remote_code=True,
enforce_eager=True,
# GLM-4-9B-Chat-1M 如果遇见 OOM 现象，建议开启下述参数
# enable_chunked_prefill=True,
# max_num_batched_tokens=8192
)
stop_token_ids = [151329, 151336, 151338]
sampling_params = SamplingParams(temperature=0.95, max_tokens=1024, stop_token_ids=stop_token_ids)
inputs = tokenizer.apply_chat_template(prompt, tokenize=False, add_generation_prompt=True)
outputs = llm.generate(prompts=inputs, sampling_params=sampling_params)
print(outputs[0].outputs[0].text)
```
### 使用以下方法快速调用 GLM-4V-9B 多模态模型
使用 transformers 后端进行推理:
```python
import torch
from PIL import Image
from transformers import AutoModelForCausalLM, AutoTokenizer
device = "cuda"
tokenizer = AutoTokenizer.from_pretrained("THUDM/glm-4v-9b", trust_remote_code=True)
query = '描述这张图片'
image = Image.open("your image").convert('RGB')
inputs = tokenizer.apply_chat_template([{"role": "user", "image": image, "content": query}],
add_generation_prompt=True, tokenize=True, return_tensors="pt",
return_dict=True) # chat mode
inputs = inputs.to(device)
model = AutoModelForCausalLM.from_pretrained(
"THUDM/glm-4v-9b",
torch_dtype=torch.bfloat16,
low_cpu_mem_usage=True,
trust_remote_code=True
).to(device).eval()
gen_kwargs = {"max_length": 2500, "do_sample": True, "top_k": 1}
with torch.no_grad():
outputs = model.generate(**inputs, **gen_kwargs)
outputs = outputs[:, inputs['input_ids'].shape[1]:]
print(tokenizer.decode(outputs[0]))
```
注意: GLM-4V-9B 暂不支持使用 vLLM 方式调用。
## 完整项目列表
如果你想更进一步了解 GLM-4-9B 系列开源模型，本开源仓库通过以下内容为开发者提供基础的 GLM-4-9B的使用和开发代码
+ [basic_demo](basic_demo/README.md): 在这里包含了
+ 使用 transformers 和 vLLM 后端的交互代码
+ OpenAI API 后端交互代码
+ Batch 推理代码
+ [composite_demo](composite_demo/README.md): 在这里包含了
+ GLM-4-9B-Chat 以及 GLM-4V-9B 开源模型的完整功能演示代码，包含了 All Tools 能力、长文档解读和多模态能力的展示。
+ [fintune_demo](finetune_demo/README.md): 在这里包含了
+ PEFT (LORA, P-Tuning) 微调代码
+ SFT 微调代码
## 友情链接
+ [LLaMA-Factory](https://github.com/hiyouga/LLaMA-Factory): 高效开源微调框架，已支持 GLM-4-9B-Chat 语言模型微调。
+ [SWIFT](https://github.com/modelscope/swift): 魔搭社区的大模型/多模态大模型训练框架，已支持 GLM4-9B-Chat/GLM4v-9B-Chat 模型微调。
+ [Xorbits Inference](https://github.com/xorbitsai/inference): 性能强大且功能全面的分布式推理框架，轻松一键部署你自己的模型或内置的前沿开源模型。
+ [self-llm](https://github.com/datawhalechina/self-llm/tree/master/GLM-4): Datawhale 团队的提供的 GLM-4-9B 系列模型使用教程。
## 协议
+ GLM-4 模型的权重的使用则需要遵循 [模型协议](https://huggingface.co/THUDM/glm-4-9b/blob/main/LICENSE)。
+ 本开源仓库的代码则遵循 [Apache 2.0](LICENSE) 协议。
请您严格遵循开源协议。
## 引用
如果你觉得我们的工作有帮助的话，请考虑引用下列论文。
```
@inproceedings{zeng2022glm,
title={{GLM-130B:} An Open Bilingual Pre-trained Model},
author={Zeng, Aohan and Liu, Xiao and Du, Zhengxiao and Wang, Zihan and Lai, Hanyu and Ding, Ming and Yang, Zhuoyi and Xu, Yifan and Zheng, Wendi and Xia, Xiao and others},
booktitle={The Eleventh International Conference on Learning Representations,
{ICLR} 2023, Kigali, Rwanda, May 1-5, 2023},
year= {2023},
}
```
```
@inproceedings{du2022glm,
title={GLM: General Language Model Pretraining with Autoregressive Blank Infilling},
author={Du, Zhengxiao and Qian, Yujie and Liu, Xiao and Ding, Ming and Qiu, Jiezhong and Yang, Zhilin and Tang, Jie},
booktitle={Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)},
pages={320--335},
year={2022}
}
```
```
@misc{wang2023cogvlm,
title={CogVLM: Visual Expert for Pretrained Language Models},
author={Weihan Wang and Qingsong Lv and Wenmeng Yu and Wenyi Hong and Ji Qi and Yan Wang and Junhui Ji and Zhuoyi Yang and Lei Zhao and Xixuan Song and Jiazheng Xu and Bin Xu and Juanzi Li and Yuxiao Dong and Ming Ding and Jie Tang},
year={2023},
eprint={2311.03079},
archivePrefix={arXiv},
primaryClass={cs.CV}
}
```
basic_demo/README.md 0 → 100644
View file @ 1bfbcff0
# Basic Demo
Read this in [English](README_en.md).
本 demo 中，你将体验到如何使用 GLM-4-9B 开源模型进行基本的任务。
请严格按照文档的步骤进行操作，以避免不必要的错误。
## 设备和依赖检查
### 相关推理测试数据
**本文档的数据均在以下硬件环境测试,实际运行环境需求和运行占用的显存略有不同，请以实际运行环境为准。**
测试硬件信息:
+ OS: Ubuntu 22.04
+ Memory: 512GB
+ Python: 3.10.12 （推荐） / 3.12.3 均已测试
+ CUDA Version: 12.3
+ GPU Driver: 535.104.05
+ GPU: NVIDIA A100-SXM4-80GB * 8
相关推理的压力测试数据如下：
**所有测试均在单张GPU上进行测试,所有显存消耗都按照峰值左右进行测算**
#### GLM-4-9B-Chat
| 精度 | 显存占用 | Prefilling | Decode Speed | Remarks |
|------|-------|------------|---------------|--------------|
| BF16 | 19 GB | 0.2s | 27.8 tokens/s | 输入长度为 1000 |
| BF16 | 21 GB | 0.8s | 31.8 tokens/s | 输入长度为 8000 |
| BF16 | 28 GB | 4.3s | 14.4 tokens/s | 输入长度为 32000 |
| BF16 | 58 GB | 38.1s | 3.4 tokens/s | 输入长度为 128000 |
| 精度 | 显存占用 | Prefilling | Decode Speed | Remarks |
|------|-------|------------|---------------|-------------|
| INT4 | 8 GB | 0.2s | 23.3 tokens/s | 输入长度为 1000 |
| INT4 | 10 GB | 0.8s | 23.4 tokens/s | 输入长度为 8000 |
| INT4 | 17 GB | 4.3s | 14.6 tokens/s | 输入长度为 32000 |
### GLM-4-9B-Chat-1M
| 精度 | 显存占用 | Prefilling | Decode Speed | Remarks |
|------|-------|------------|--------------|--------------|
| BF16 | 75 GB | 98.4s | 2.3 tokens/s | 输入长度为 200000 |
如果您的输入超过200K，我们建议您使用vLLM后端进行多卡推理，以获得更好的性能。
#### GLM-4V-9B
| 精度 | 显存占用 | Prefilling | Decode Speed | Remarks |
|------|-------|------------|---------------|------------|
| BF16 | 28 GB | 0.1s | 33.4 tokens/s | 输入长度为 1000 |
| BF16 | 33 GB | 0.7s | 39.2 tokens/s | 输入长度为 8000 |
| 精度 | 显存占用 | Prefilling | Decode Speed | Remarks |
|------|-------|------------|---------------|------------|
| INT4 | 10 GB | 0.1s | 28.7 tokens/s | 输入长度为 1000 |
| INT4 | 15 GB | 0.8s | 24.2 tokens/s | 输入长度为 8000 |
### 最低硬件要求
如果您希望运行官方提供的最基础代码 (transformers 后端) 您需要：
+ Python >= 3.10
+ 内存不少于 32 GB
如果您希望运行官方提供的本文件夹的所有代码，您还需要：
+ Linux 操作系统 (Debian 系列最佳)
+ 大于 8GB 显存的，支持 CUDA 或者 ROCM 并且支持 `BF16` 推理的 GPU 设备。(`FP16` 精度无法训练，推理有小概率出现问题)
安装依赖
```shell
pip install -r requirements.txt
```
## 基础功能调用
**除非特殊说明，本文件夹所有 demo 并不支持 Function Call 和 All Tools 等进阶用法**
### 使用 transformers 后端代码
+ 使用命令行与 GLM-4-9B 模型进行对话。
```shell
python trans_cli_demo.py # GLM-4-9B-Chat
python trans_cli_vision_demo.py # GLM-4V-9B
```
+ 使用 Gradio 网页端与 GLM-4-9B-Chat 模型进行对话。
```shell
python trans_web_demo.py
```
+ 使用 Batch 推理。
```shell
python cli_batch_request_demo.py
```
### 使用 vLLM 后端代码
+ 使用命令行与 GLM-4-9B-Chat 模型进行对话。
```shell
python vllm_cli_demo.py
```
+ 自行构建服务端，并使用 `OpenAI API` 的请求格式与 GLM-4-9B-Chat 模型进行对话。本 demo 支持 Function Call 和 All Tools功能。
启动服务端：
```shell
python openai_api_server.py
```
客户端请求：
```shell
python openai_api_request.py
```
## 压力测试
用户可以在自己的设备上使用本代码测试模型在 transformers后端的生成速度:
```shell
python trans_stress_test.py
```
basic_demo/README_en.md 0 → 100644
View file @ 1bfbcff0
# Basic Demo
In this demo, you will experience how to use the GLM-4-9B open source model to perform basic tasks.
Please follow the steps in the document strictly to avoid unnecessary errors.
## Device and dependency check
### Related inference test data
**The data in this document are tested in the following hardware environment. The actual operating environment
requirements and the GPU memory occupied by the operation are slightly different. Please refer to the actual operating
environment.**
Test hardware information:
+ OS: Ubuntu 22.04
+ Memory: 512GB
+ Python: 3.10.12 (recommend) / 3.12.3 have been tested
+ CUDA Version: 12.3
+ GPU Driver: 535.104.05
+ GPU: NVIDIA A100-SXM4-80GB * 8
The stress test data of relevant inference are as follows:
**All tests are performed on a single GPU, and all GPU memory consumption is calculated based on the peak value**
#
### GLM-4-9B-Chat
| Dtype | GPU Memory | Prefilling | Decode Speed | Remarks |
|-------|------------|------------|---------------|------------------------|
| BF16 | 19 GB | 0.2s | 27.8 tokens/s | Input length is 1000 |
| BF16 | 21 GB | 0.8s | 31.8 tokens/s | Input length is 8000 |
| BF16 | 28 GB | 4.3s | 14.4 tokens/s | Input length is 32000 |
| BF16 | 58 GB | 38.1s | 3.4 tokens/s | Input length is 128000 |
| Dtype | GPU Memory | Prefilling | Decode Speed | Remarks |
|-------|------------|------------|---------------|-----------------------|
| INT4 | 8 GB | 0.2s | 23.3 tokens/s | Input length is 1000 |
| INT4 | 10 GB | 0.8s | 23.4 tokens/s | Input length is 8000 |
| INT4 | 17 GB | 4.3s | 14.6 tokens/s | Input length is 32000 |
### GLM-4-9B-Chat-1M
| Dtype | GPU Memory | Prefilling | Decode Speed | Remarks |
|-------|------------|------------|------------------|------------------------|
| BF16 | 74497MiB | 98.4s | 2.3653 tokens/s | Input length is 200000 |
If your input exceeds 200K, we recommend that you use the vLLM backend with multi gpus for inference to get better
performance.
#### GLM-4V-9B
| Dtype | GPU Memory | Prefilling | Decode Speed | Remarks |
|-------|------------|------------|---------------|----------------------|
| BF16 | 28 GB | 0.1s | 33.4 tokens/s | Input length is 1000 |
| BF16 | 33 GB | 0.7s | 39.2 tokens/s | Input length is 8000 |
| Dtype | GPU Memory | Prefilling | Decode Speed | Remarks |
|-------|------------|------------|---------------|----------------------|
| INT4 | 10 GB | 0.1s | 28.7 tokens/s | Input length is 1000 |
| INT4 | 15 GB | 0.8s | 24.2 tokens/s | Input length is 8000 |
### Minimum hardware requirements
If you want to run the most basic code provided by the official (transformers backend) you need:
+ Python >= 3.10
+ Memory of at least 32 GB
If you want to run all the codes in this folder provided by the official, you also need:
+ Linux operating system (Debian series is best)
+ GPU device with more than 8GB GPU memory, supporting CUDA or ROCM and supporting `BF16` reasoning (`FP16` precision
cannot be finetuned, and there is a small probability of problems in infering)
Install dependencies
```shell
pip install -r requirements.txt
```
## Basic function calls
**Unless otherwise specified, all demos in this folder do not support advanced usage such as Function Call and All Tools
**
### Use transformers backend code
+ Use the command line to communicate with the GLM-4-9B model.
```shell
python trans_cli_demo.py # GLM-4-9B-Chat
python trans_cli_vision_demo.py # GLM-4V-9B
```
+ Use the Gradio web client to communicate with the GLM-4-9B-Chat model.
```shell
python trans_web_demo.py
```
+ Use Batch inference.
```shell
python cli_batch_request_demo.py
```
### Use vLLM backend code
+ Use the command line to communicate with the GLM-4-9B-Chat model.
```shell
python vllm_cli_demo.py
```
+ Build the server by yourself and use the request format of `OpenAI API` to communicate with the glm-4-9b model. This
demo supports Function Call and All Tools functions.
Start the server:
```shell
python openai_api_server.py
```
Client request:
```shell
python openai_api_request.py
```
## Stress test
Users can use this code to test the generation speed of the model on the transformers backend on their own devices:
```shell
python trans_stress_test.py
```
\ No newline at end of file
basic_demo/openai_api_request.py 0 → 100644
View file @ 1bfbcff0
"""
This script creates a OpenAI Request demo for the glm-4-9b model, just Use OpenAI API to interact with the model.
"""
from openai import OpenAI
base_url = "http://127.0.0.1:8000/v1/"
client = OpenAI(api_key="EMPTY", base_url=base_url)
def function_chat(use_stream=False):
messages = [
{
"role": "user", "content": "What's the Celsius temperature in San Francisco?"
},
# Give Observations
# {
# "role": "assistant",
# "content": None,
# "function_call": None,
# "tool_calls": [
# {
# "id": "call_1717912616815",
# "function": {
# "name": "get_current_weather",
# "arguments": "{\"location\": \"San Francisco, CA\", \"format\": \"celsius\"}"
# },
# "type": "function"
# }
# ]
# },
# {
# "tool_call_id": "call_1717912616815",
# "role": "tool",
# "name": "get_current_weather",
# "content": "23°C",
# }
]
tools = [
{
"type": "function",
"function": {
"name": "get_current_weather",
"description": "Get the current weather",
"parameters": {
"type": "object",
"properties": {
"location": {
"type": "string",
"description": "The city and state, e.g. San Francisco, CA",
},
"format": {
"type": "string",
"enum": ["celsius", "fahrenheit"],
"description": "The temperature unit to use. Infer this from the users location.",
},
},
"required": ["location", "format"],
},
}
},
]
# All Tools: CogView
# messages = [{"role": "user", "content": "帮我画一张天空的画画吧"}]
# tools = [{"type": "cogview"}]
# All Tools: Searching
# messages = [{"role": "user", "content": "今天黄金的价格"}]
# tools = [{"type": "simple_browser"}]
response = client.chat.completions.create(
model="glm-4",
messages=messages,
tools=tools,
stream=use_stream,
max_tokens=256,
temperature=0.9,
presence_penalty=1.2,
top_p=0.1,
tool_choice="auto"
)
if response:
if use_stream:
for chunk in response:
print(chunk)
else:
print(response)
else:
print("Error:", response.status_code)
def simple_chat(use_stream=False):
messages = [
{
"role": "system",
"content": "请在你输出的时候都带上“喵喵喵”三个字，放在开头。",
},
{
"role": "user",
"content": "你是谁"
}
]
response = client.chat.completions.create(
model="glm-4",
messages=messages,
stream=use_stream,
max_tokens=256,
temperature=0.4,
presence_penalty=1.2,
top_p=0.8,
)
if response:
if use_stream:
for chunk in response:
print(chunk)
else:
print(response)
else:
print("Error:", response.status_code)
if __name__ == "__main__":
# simple_chat(use_stream=False)
function_chat(use_stream=False)
basic_demo/openai_api_server.py 0 → 100644
View file @ 1bfbcff0
import time
from asyncio.log import logger
import re
import uvicorn
import gc
import json
import torch
from vllm import SamplingParams, AsyncEngineArgs, AsyncLLMEngine
from fastapi import FastAPI, HTTPException, Response
from fastapi.middleware.cors import CORSMiddleware
from contextlib import asynccontextmanager
from typing import List, Literal, Optional, Union
from pydantic import BaseModel, Field
from transformers import AutoTokenizer, LogitsProcessor
from sse_starlette.sse import EventSourceResponse
EventSourceResponse.DEFAULT_PING_INTERVAL = 1000
MODEL_PATH = 'THUDM/glm-4-9b-chat'
MAX_MODEL_LENGTH = 8192
@asynccontextmanager
async def lifespan(app: FastAPI):
yield
if torch.cuda.is_available():
torch.cuda.empty_cache()
torch.cuda.ipc_collect()
app = FastAPI(lifespan=lifespan)
app.add_middleware(
CORSMiddleware,
allow_origins=["*"],
allow_credentials=True,
allow_methods=["*"],
allow_headers=["*"],
)
class ModelCard(BaseModel):
id: str
object: str = "model"
created: int = Field(default_factory=lambda: int(time.time()))
owned_by: str = "owner"
root: Optional[str] = None
parent: Optional[str] = None
permission: Optional[list] = None
class ModelList(BaseModel):
object: str = "list"
data: List[ModelCard] = ["glm-4"]
class FunctionCall(BaseModel):
name: str
arguments: str
class FunctionCallResponse(BaseModel):
name: Optional[str] = None
arguments: Optional[str] = None
class UsageInfo(BaseModel):
prompt_tokens: int = 0
total_tokens: int = 0
completion_tokens: Optional[int] = 0
class ChatCompletionMessageToolCall(BaseModel):
id: str
function: FunctionCall
type: Literal["function"]
class ChatMessage(BaseModel):
role: Literal["user", "assistant", "system", "tool"]
content: Optional[str] = None
function_call: Optional[FunctionCallResponse] = None
tool_calls: Optional[List[ChatCompletionMessageToolCall]] = None
class DeltaMessage(BaseModel):
role: Optional[Literal["user", "assistant", "system"]] = None
content: Optional[str] = None
function_call: Optional[FunctionCallResponse] = None
class ChatCompletionResponseChoice(BaseModel):
index: int
message: ChatMessage
finish_reason: Literal["stop", "length", "tool_calls"]
class ChatCompletionResponseStreamChoice(BaseModel):
delta: DeltaMessage
finish_reason: Optional[Literal["stop", "length", "tool_calls"]]
index: int
class ChatCompletionResponse(BaseModel):
model: str
id: str
object: Literal["chat.completion", "chat.completion.chunk"]
choices: List[Union[ChatCompletionResponseChoice, ChatCompletionResponseStreamChoice]]
created: Optional[int] = Field(default_factory=lambda: int(time.time()))
usage: Optional[UsageInfo] = None
class ChatCompletionRequest(BaseModel):
model: str
messages: List[ChatMessage]
temperature: Optional[float] = 0.8
top_p: Optional[float] = 0.8
max_tokens: Optional[int] = None
stream: Optional[bool] = False
tools: Optional[Union[dict, List[dict]]] = None
tool_choice: Optional[Union[str, dict]] = "None"
repetition_penalty: Optional[float] = 1.1
class InvalidScoreLogitsProcessor(LogitsProcessor):
def __call__(
self, input_ids: torch.LongTensor, scores: torch.FloatTensor
) -> torch.FloatTensor:
if torch.isnan(scores).any() or torch.isinf(scores).any():
scores.zero_()
scores[..., 5] = 5e4
return scores
def process_response(output: str, use_tool: bool = False) -> Union[str, dict]:
lines = output.strip().split("\n")
arguments_json = None
special_tools = ["cogview", "simple_browser"]
tool_call_pattern = re.compile(r'^[a-zA-Z_][a-zA-Z0-9_]*$')
if len(lines) >= 2 and tool_call_pattern.match(lines[0]):
function_name = lines[0].strip()
arguments = "\n".join(lines[1:]).strip()
try:
arguments_json = json.loads(arguments)
is_tool_call = True
except json.JSONDecodeError:
is_tool_call = function_name in special_tools
if is_tool_call and use_tool:
content = {
"name": function_name,
"arguments": json.dumps(arguments_json if isinstance(arguments_json, dict) else arguments, ensure_ascii=False)
}
if function_name == "simple_browser":
search_pattern = re.compile(r'search\("(.+?)"\s*,\s*recency_days\s*=\s*(\d+)\)')
match = search_pattern.match(arguments)
if match:
content["arguments"] = json.dumps({
"query": match.group(1),
"recency_days": int(match.group(2))
}, ensure_ascii=False)
elif function_name == "cogview":
content["arguments"] = json.dumps({
"prompt": arguments
}, ensure_ascii=False)
return content
return output.strip()
@torch.inference_mode()
async def generate_stream_glm4(params):
messages = params["messages"]
tools = params["tools"]
tool_choice = params["tool_choice"]
temperature = float(params.get("temperature", 1.0))
repetition_penalty = float(params.get("repetition_penalty", 1.0))
top_p = float(params.get("top_p", 1.0))
max_new_tokens = int(params.get("max_tokens", 8192))
messages = process_messages(messages, tools=tools, tool_choice=tool_choice)
inputs = tokenizer.apply_chat_template(messages, add_generation_prompt=True, tokenize=False)
params_dict = {
"n": 1,
"best_of": 1,
"presence_penalty": 1.0,
"frequency_penalty": 0.0,
"temperature": temperature,
"top_p": top_p,
"top_k": -1,
"repetition_penalty": repetition_penalty,
"use_beam_search": False,
"length_penalty": 1,
"early_stopping": False,
"stop_token_ids": [151329, 151336, 151338],
"ignore_eos": False,
"max_tokens": max_new_tokens,
"logprobs": None,
"prompt_logprobs": None,
"skip_special_tokens": True,
}
sampling_params = SamplingParams(**params_dict)
async for output in engine.generate(inputs=inputs, sampling_params=sampling_params, request_id=f"{time.time()}"):
output_len = len(output.outputs[0].token_ids)
input_len = len(output.prompt_token_ids)
ret = {
"text": output.outputs[0].text,
"usage": {
"prompt_tokens": input_len,
"completion_tokens": output_len,
"total_tokens": output_len + input_len
},
"finish_reason": output.outputs[0].finish_reason,
}
yield ret
gc.collect()
torch.cuda.empty_cache()
def process_messages(messages, tools=None, tool_choice="none"):
_messages = messages
processed_messages = []
msg_has_sys = False
def filter_tools(tool_choice, tools):
function_name = tool_choice.get('function', {}).get('name', None)
if not function_name:
return []
filtered_tools = [
tool for tool in tools
if tool.get('function', {}).get('name') == function_name
]
return filtered_tools
if tool_choice != "none":
if isinstance(tool_choice, dict):
tools = filter_tools(tool_choice, tools)
if tools:
processed_messages.append(
{
"role": "system",
"content": None,
"tools": tools
}
)
msg_has_sys = True
if isinstance(tool_choice, dict) and tools:
processed_messages.append(
{
"role": "assistant",
"metadata": tool_choice["function"]["name"],
"content": ""
}
)
for m in _messages:
role, content, func_call = m.role, m.content, m.function_call
tool_calls = getattr(m, 'tool_calls', None)
if role == "function":
processed_messages.append(
{
"role": "observation",
"content": content
}
)
elif role == "tool":
processed_messages.append(
{
"role": "observation",
"content": content,
"function_call": True
}
)
elif role == "assistant":
if tool_calls:
for tool_call in tool_calls:
processed_messages.append(
{
"role": "assistant",
"metadata": tool_call.function.name,
"content": tool_call.function.arguments
}
)
else:
for response in content.split("\n"):
if "\n" in response:
metadata, sub_content = response.split("\n", maxsplit=1)
else:
metadata, sub_content = "", response
processed_messages.append(
{
"role": role,
"metadata": metadata,
"content": sub_content.strip()
}
)
else:
if role == "system" and msg_has_sys:
msg_has_sys = False
continue
processed_messages.append({"role": role, "content": content})
if not tools or tool_choice == "none":
for m in _messages:
if m.role == 'system':
processed_messages.insert(0, {"role": m.role, "content": m.content})
break
return processed_messages
@app.get("/health")
async def health() -> Response:
"""Health check."""
return Response(status_code=200)
@app.get("/v1/models", response_model=ModelList)
async def list_models():
model_card = ModelCard(id="glm-4")
return ModelList(data=[model_card])
@app.post("/v1/chat/completions", response_model=ChatCompletionResponse)
async def create_chat_completion(request: ChatCompletionRequest):
if len(request.messages) < 1 or request.messages[-1].role == "assistant":
raise HTTPException(status_code=400, detail="Invalid request")
gen_params = dict(
messages=request.messages,
temperature=request.temperature,
top_p=request.top_p,
max_tokens=request.max_tokens or 1024,
echo=False,
stream=request.stream,
repetition_penalty=request.repetition_penalty,
tools=request.tools,
tool_choice=request.tool_choice,
)
logger.debug(f"==== request ====\n{gen_params}")
if request.stream:
predict_stream_generator = predict_stream(request.model, gen_params)
output = await anext(predict_stream_generator)
if output:
return EventSourceResponse(predict_stream_generator, media_type="text/event-stream")
logger.debug(f"First result output：\n{output}")
function_call = None
if output and request.tools:
try:
function_call = process_response(output, use_tool=True)
except:
logger.warning("Failed to parse tool call")
if isinstance(function_call, dict):
function_call = FunctionCallResponse(**function_call)
generate = parse_output_text(request.model, output, function_call=function_call)
return EventSourceResponse(generate, media_type="text/event-stream")
else:
return EventSourceResponse(predict_stream_generator, media_type="text/event-stream")
response = ""
async for response in generate_stream_glm4(gen_params):
pass
if response["text"].startswith("\n"):
response["text"] = response["text"][1:]
response["text"] = response["text"].strip()
usage = UsageInfo()
function_call, finish_reason = None, "stop"
tool_calls = None
if request.tools:
try:
function_call = process_response(response["text"], use_tool=True)
except Exception as e:
logger.warning(f"Failed to parse tool call: {e}")
if isinstance(function_call, dict):
finish_reason = "tool_calls"
function_call_response = FunctionCallResponse(**function_call)
function_call_instance = FunctionCall(
name=function_call_response.name,
arguments=function_call_response.arguments
)
tool_calls = [
ChatCompletionMessageToolCall(
id=f"call_{int(time.time() * 1000)}",
function=function_call_instance,
type="function")]
message = ChatMessage(
role="assistant",
content=None if tool_calls else response["text"],
function_call=None,
tool_calls=tool_calls,
)
logger.debug(f"==== message ====\n{message}")
choice_data = ChatCompletionResponseChoice(
index=0,
message=message,
finish_reason=finish_reason,
)
task_usage = UsageInfo.model_validate(response["usage"])
for usage_key, usage_value in task_usage.model_dump().items():
setattr(usage, usage_key, getattr(usage, usage_key) + usage_value)
return ChatCompletionResponse(
model=request.model,
id="",
choices=[choice_data],
object="chat.completion",
usage=usage
)
async def predict_stream(model_id, gen_params):
output = ""
is_function_call = False
has_send_first_chunk = False
function_name = None
async for new_response in generate_stream_glm4(gen_params):
decoded_unicode = new_response["text"]
delta_text = decoded_unicode[len(output):]
output = decoded_unicode
lines = output.strip().split("\n")
if not is_function_call and len(lines) >= 2 and re.match(r'^[a-zA-Z_][a-zA-Z0-9_]*$', lines[0]):
is_function_call = True
function_name = lines[0].strip()
if is_function_call:
for char in delta_text:
function_call = {"name": function_name, "arguments": char}
message = DeltaMessage(
content=None,
role="assistant",
function_call=function_call
)
choice_data = ChatCompletionResponseStreamChoice(
index=0,
delta=message,
finish_reason=None
)
chunk = ChatCompletionResponse(
model=model_id,
id="",
choices=[choice_data],
created=int(time.time()),
object="chat.completion.chunk"
)
yield chunk.model_dump_json(exclude_unset=True)
else:
if len(output) > 7:
finish_reason = new_response.get("finish_reason", None)
if not has_send_first_chunk:
message = DeltaMessage(
content="",
role="assistant",
function_call=None,
)
choice_data = ChatCompletionResponseStreamChoice(
index=0,
delta=message,
finish_reason=finish_reason
)
chunk = ChatCompletionResponse(
model=model_id,
id="",
choices=[choice_data],
created=int(time.time()),
object="chat.completion.chunk"
)
yield chunk.model_dump_json(exclude_unset=True)
send_msg = delta_text if has_send_first_chunk else output
has_send_first_chunk = True
message = DeltaMessage(
content=send_msg,
role="assistant",
function_call=None,
)
choice_data = ChatCompletionResponseStreamChoice(
index=0,
delta=message,
finish_reason=finish_reason
)
chunk = ChatCompletionResponse(
model=model_id,
id="",
choices=[choice_data],
created=int(time.time()),
object="chat.completion.chunk"
)
yield chunk.model_dump_json(exclude_unset=True)
if is_function_call:
yield json.dumps({"text": output})
else:
yield '[DONE]'
async def parse_output_text(model_id: str, value: str, function_call: FunctionCallResponse = None):
delta = DeltaMessage(role="assistant", content=value)
if function_call is not None:
delta.function_call = function_call
choice_data = ChatCompletionResponseStreamChoice(
index=0,
delta=delta,
finish_reason=None
)
chunk = ChatCompletionResponse(model=model_id, id="", choices=[choice_data], object="chat.completion.chunk")
yield "{}".format(chunk.model_dump_json(exclude_unset=True))
yield '[DONE]'
if __name__ == "__main__":
tokenizer = AutoTokenizer.from_pretrained(MODEL_PATH, trust_remote_code=True)
engine_args = AsyncEngineArgs(
model=MODEL_PATH,
tokenizer=MODEL_PATH,
tensor_parallel_size=1,
dtype="bfloat16",
trust_remote_code=True,
# 占用显存的比例，请根据你的显卡显存大小设置合适的值，例如，如果你的显卡有80G，您只想使用24G，请按照24/80=0.3设置
gpu_memory_utilization=0.9,
enforce_eager=True,
worker_use_ray=False,
engine_use_ray=False,
disable_log_requests=True,
max_model_len=MAX_MODEL_LENGTH,
)
engine = AsyncLLMEngine.from_engine_args(engine_args)
uvicorn.run(app, host='0.0.0.0', port=8000, workers=1)
basic_demo/requirements.txt 0 → 100644
View file @ 1bfbcff0
# use vllm
# vllm>=0.4.3
torch>=2.3.0
torchvision>=0.18.0
transformers==4.40.0
huggingface-hub>=0.23.1
sentencepiece>=0.2.0
pydantic>=2.7.1
timm>=0.9.16
tiktoken>=0.7.0
accelerate>=0.30.1
sentence_transformers>=2.7.0
# web demo
gradio>=4.33.0
# openai demo
openai>=1.31.1
einops>=0.7.0
sse-starlette>=2.1.0
# INT4
bitsandbytes>=0.43.1
# PEFT model, not need if you don't use PEFT finetune model.
# peft>=0.11.0
\ No newline at end of file
basic_demo/trans_batch_demo.py 0 → 100644
View file @ 1bfbcff0
"""
Here is an example of using batch request glm-4-9b,
here you need to build the conversation format yourself and then call the batch function to make batch requests.
Please note that in this demo, the memory consumption is significantly higher.
"""
from typing import Optional, Union
from transformers import AutoModel, AutoTokenizer, LogitsProcessorList
MODEL_PATH = 'THUDM/glm-4-9b-chat'
tokenizer = AutoTokenizer.from_pretrained(
MODEL_PATH,
trust_remote_code=True,
encode_special_tokens=True)
model = AutoModel.from_pretrained(MODEL_PATH, trust_remote_code=True, device_map="auto").eval()
def process_model_outputs(inputs, outputs, tokenizer):
responses = []
for input_ids, output_ids in zip(inputs.input_ids, outputs):
response = tokenizer.decode(output_ids[len(input_ids):], skip_special_tokens=True).strip()
responses.append(response)
return responses
def batch(
model,
tokenizer,
messages: Union[str, list[str]],
max_input_tokens: int = 8192,
max_new_tokens: int = 8192,
num_beams: int = 1,
do_sample: bool = True,
top_p: float = 0.8,
temperature: float = 0.8,
logits_processor: Optional[LogitsProcessorList] = LogitsProcessorList(),
):
messages = [messages] if isinstance(messages, str) else messages
batched_inputs = tokenizer(messages, return_tensors="pt", padding="max_length", truncation=True,
max_length=max_input_tokens).to(model.device)
gen_kwargs = {
"max_new_tokens": max_new_tokens,
"num_beams": num_beams,
"do_sample": do_sample,
"top_p": top_p,
"temperature": temperature,
"logits_processor": logits_processor,
"eos_token_id": model.config.eos_token_id
}
batched_outputs = model.generate(**batched_inputs, **gen_kwargs)
batched_response = process_model_outputs(batched_inputs, batched_outputs, tokenizer)
return batched_response
if __name__ == "__main__":
batch_message = [
[
{"role": "user", "content": "我的爸爸和妈妈结婚为什么不能带我去"},
{"role": "assistant", "content": "因为他们结婚时你还没有出生"},
{"role": "user", "content": "我刚才的提问是"}
],
[
{"role": "user", "content": "你好，你是谁"}
]
]
batch_inputs = []
max_input_tokens = 1024
for i, messages in enumerate(batch_message):
new_batch_input = tokenizer.apply_chat_template(messages, add_generation_prompt=True, tokenize=False)
max_input_tokens = max(max_input_tokens, len(new_batch_input))
batch_inputs.append(new_batch_input)
gen_kwargs = {
"max_input_tokens": max_input_tokens,
"max_new_tokens": 8192,
"do_sample": True,
"top_p": 0.8,
"temperature": 0.8,
"num_beams": 1,
}
batch_responses = batch(model, tokenizer, batch_inputs, **gen_kwargs)
for response in batch_responses:
print("=" * 10)
print(response)
basic_demo/trans_cli_demo.py 0 → 100644
View file @ 1bfbcff0
"""
This script creates a CLI demo with transformers backend for the glm-4-9b model,
allowing users to interact with the model through a command-line interface.
Usage:
- Run the script to start the CLI demo.
- Interact with the model by typing questions and receiving responses.
Note: The script includes a modification to handle markdown to plain text conversion,
ensuring that the CLI interface displays formatted text correctly.
"""
import os
import torch
from threading import Thread
from transformers import AutoTokenizer, StoppingCriteria, StoppingCriteriaList, TextIteratorStreamer, AutoModel
MODEL_PATH = os.environ.get('MODEL_PATH', 'THUDM/glm-4-9b-chat')
## If use peft model.
# def load_model_and_tokenizer(model_dir, trust_remote_code: bool = True):
# if (model_dir / 'adapter_config.json').exists():
# model = AutoModel.from_pretrained(
# model_dir, trust_remote_code=trust_remote_code, device_map='auto'
# )
# tokenizer_dir = model.peft_config['default'].base_model_name_or_path
# else:
# model = AutoModel.from_pretrained(
# model_dir, trust_remote_code=trust_remote_code, device_map='auto'
# )
# tokenizer_dir = model_dir
# tokenizer = AutoTokenizer.from_pretrained(
# tokenizer_dir, trust_remote_code=trust_remote_code, use_fast=False
# )
# return model, tokenizer
tokenizer = AutoTokenizer.from_pretrained(
MODEL_PATH,
trust_remote_code=True,
encode_special_tokens=True
)
model = AutoModel.from_pretrained(
MODEL_PATH,
trust_remote_code=True,
device_map="auto").eval()
class StopOnTokens(StoppingCriteria):
def __call__(self, input_ids: torch.LongTensor, scores: torch.FloatTensor, **kwargs) -> bool:
stop_ids = model.config.eos_token_id
for stop_id in stop_ids:
if input_ids[0][-1] == stop_id:
return True
return False
if __name__ == "__main__":
history = []
max_length = 8192
top_p = 0.8
temperature = 0.6
stop = StopOnTokens()
print("Welcome to the GLM-4-9B CLI chat. Type your messages below.")
while True:
user_input = input("\nYou: ")
if user_input.lower() in ["exit", "quit"]:
break
history.append([user_input, ""])
messages = []
for idx, (user_msg, model_msg) in enumerate(history):
if idx == len(history) - 1 and not model_msg:
messages.append({"role": "user", "content": user_msg})
break
if user_msg:
messages.append({"role": "user", "content": user_msg})
if model_msg:
messages.append({"role": "assistant", "content": model_msg})
model_inputs = tokenizer.apply_chat_template(
messages,
add_generation_prompt=True,
tokenize=True,
return_tensors="pt"
).to(model.device)
streamer = TextIteratorStreamer(
tokenizer=tokenizer,
timeout=60,
skip_prompt=True,
skip_special_tokens=True
)
generate_kwargs = {
"input_ids": model_inputs,
"streamer": streamer,
"max_new_tokens": max_length,
"do_sample": True,
"top_p": top_p,
"temperature": temperature,
"stopping_criteria": StoppingCriteriaList([stop]),
"repetition_penalty": 1.2,
"eos_token_id": model.config.eos_token_id,
}
t = Thread(target=model.generate, kwargs=generate_kwargs)
t.start()
print("GLM-4:", end="", flush=True)
for new_token in streamer:
if new_token:
print(new_token, end="", flush=True)
history[-1][1] += new_token
history[-1][1] = history[-1][1].strip()
basic_demo/trans_stress_test.py 0 → 100644
View file @ 1bfbcff0
import argparse
import time
from transformers import AutoModelForCausalLM, AutoTokenizer, TextIteratorStreamer, BitsAndBytesConfig
import torch
from threading import Thread
MODEL_PATH = 'THUDM/glm-4-9b-chat'
def stress_test(token_len, n, num_gpu):
device = torch.device(f"cuda:{num_gpu - 1}" if torch.cuda.is_available() and num_gpu > 0 else "cpu")
tokenizer = AutoTokenizer.from_pretrained(
MODEL_PATH,
trust_remote_code=True,
padding_side="left"
)
model = AutoModelForCausalLM.from_pretrained(
MODEL_PATH,
trust_remote_code=True,
torch_dtype=torch.bfloat16
).to(device).eval()
# Use INT4 weight infer
# model = AutoModelForCausalLM.from_pretrained(
# MODEL_PATH,
# trust_remote_code=True,
# quantization_config=BitsAndBytesConfig(load_in_4bit=True),
# low_cpu_mem_usage=True,
# ).eval()
times = []
decode_times = []
print("Warming up...")
vocab_size = tokenizer.vocab_size
warmup_token_len = 20
random_token_ids = torch.randint(3, vocab_size - 200, (warmup_token_len - 5,), dtype=torch.long)
start_tokens = [151331, 151333, 151336, 198]
end_tokens = [151337]
input_ids = torch.tensor(start_tokens + random_token_ids.tolist() + end_tokens, dtype=torch.long).unsqueeze(0).to(
device)
attention_mask = torch.ones_like(input_ids, dtype=torch.bfloat16).to(device)
position_ids = torch.arange(len(input_ids[0]), dtype=torch.bfloat16).unsqueeze(0).to(device)
warmup_inputs = {
'input_ids': input_ids,
'attention_mask': attention_mask,
'position_ids': position_ids
}
with torch.no_grad():
_ = model.generate(
input_ids=warmup_inputs['input_ids'],
attention_mask=warmup_inputs['attention_mask'],
max_new_tokens=2048,
do_sample=False,
repetition_penalty=1.0,
eos_token_id=[151329, 151336, 151338]
)
print("Warming up complete. Starting stress test...")
for i in range(n):
random_token_ids = torch.randint(3, vocab_size - 200, (token_len - 5,), dtype=torch.long)
input_ids = torch.tensor(start_tokens + random_token_ids.tolist() + end_tokens, dtype=torch.long).unsqueeze(
0).to(device)
attention_mask = torch.ones_like(input_ids, dtype=torch.bfloat16).to(device)
position_ids = torch.arange(len(input_ids[0]), dtype=torch.bfloat16).unsqueeze(0).to(device)
test_inputs = {
'input_ids': input_ids,
'attention_mask': attention_mask,
'position_ids': position_ids
}
streamer = TextIteratorStreamer(
tokenizer=tokenizer,
timeout=36000,
skip_prompt=True,
skip_special_tokens=True
)
generate_kwargs = {
"input_ids": test_inputs['input_ids'],
"attention_mask": test_inputs['attention_mask'],
"max_new_tokens": 512,
"do_sample": False,
"repetition_penalty": 1.0,
"eos_token_id": [151329, 151336, 151338],
"streamer": streamer
}
start_time = time.time()
t = Thread(target=model.generate, kwargs=generate_kwargs)
t.start()
first_token_time = None
all_token_times = []
for token in streamer:
current_time = time.time()
if first_token_time is None:
first_token_time = current_time
times.append(first_token_time - start_time)
all_token_times.append(current_time)
t.join()
end_time = time.time()
avg_decode_time_per_token = len(all_token_times) / (end_time - first_token_time) if all_token_times else 0
decode_times.append(avg_decode_time_per_token)
print(
f"Iteration {i + 1}/{n} - Prefilling Time: {times[-1]:.4f} seconds - Average Decode Time: {avg_decode_time_per_token:.4f} tokens/second")
torch.cuda.empty_cache()
avg_first_token_time = sum(times) / n
avg_decode_time = sum(decode_times) / n
print(f"\nAverage First Token Time over {n} iterations: {avg_first_token_time:.4f} seconds")
print(f"Average Decode Time per Token over {n} iterations: {avg_decode_time:.4f} tokens/second")
return times, avg_first_token_time, decode_times, avg_decode_time
def main():
parser = argparse.ArgumentParser(description="Stress test for model inference")
parser.add_argument('--token_len', type=int, default=1000, help='Number of tokens for each test')
parser.add_argument('--n', type=int, default=3, help='Number of iterations for the stress test')
parser.add_argument('--num_gpu', type=int, default=1, help='Number of GPUs to use for inference')
args = parser.parse_args()
token_len = args.token_len
n = args.n
num_gpu = args.num_gpu
stress_test(token_len, n, num_gpu)
if __name__ == "__main__":
main()
basic_demo/trans_web_demo.py 0 → 100644
View file @ 1bfbcff0
"""
This script creates an interactive web demo for the GLM-4-9B model using Gradio,
a Python library for building quick and easy UI components for machine learning models.
It's designed to showcase the capabilities of the GLM-4-9B model in a user-friendly interface,
allowing users to interact with the model through a chat-like interface.
"""
import os
from pathlib import Path
from threading import Thread
from typing import Union
import gradio as gr
import torch
from peft import AutoPeftModelForCausalLM, PeftModelForCausalLM
from transformers import (
AutoModelForCausalLM,
AutoTokenizer,
PreTrainedModel,
PreTrainedTokenizer,
PreTrainedTokenizerFast,
StoppingCriteria,
StoppingCriteriaList,
TextIteratorStreamer
)
ModelType = Union[PreTrainedModel, PeftModelForCausalLM]
TokenizerType = Union[PreTrainedTokenizer, PreTrainedTokenizerFast]
MODEL_PATH = os.environ.get('MODEL_PATH', 'THUDM/glm-4-9b-chat')
TOKENIZER_PATH = os.environ.get("TOKENIZER_PATH", MODEL_PATH)
def _resolve_path(path: Union[str, Path]) -> Path:
return Path(path).expanduser().resolve()
def load_model_and_tokenizer(
model_dir: Union[str, Path], trust_remote_code: bool = True
) -> tuple[ModelType, TokenizerType]:
model_dir = _resolve_path(model_dir)
if (model_dir / 'adapter_config.json').exists():
model = AutoPeftModelForCausalLM.from_pretrained(
model_dir, trust_remote_code=trust_remote_code, device_map='auto'
)
tokenizer_dir = model.peft_config['default'].base_model_name_or_path
else:
model = AutoModelForCausalLM.from_pretrained(
model_dir, trust_remote_code=trust_remote_code, device_map='auto'
)
tokenizer_dir = model_dir
tokenizer = AutoTokenizer.from_pretrained(
tokenizer_dir, trust_remote_code=trust_remote_code, use_fast=False
)
return model, tokenizer
model, tokenizer = load_model_and_tokenizer(MODEL_PATH, trust_remote_code=True)
class StopOnTokens(StoppingCriteria):
def __call__(self, input_ids: torch.LongTensor, scores: torch.FloatTensor, **kwargs) -> bool:
stop_ids = model.config.eos_token_id
for stop_id in stop_ids:
if input_ids[0][-1] == stop_id:
return True
return False
def parse_text(text):
lines = text.split("\n")
lines = [line for line in lines if line != ""]
count = 0
for i, line in enumerate(lines):
if "```" in line:
count += 1
items = line.split('`')
if count % 2 == 1:
lines[i] = f'<pre><code class="language-{items[-1]}">'
else:
lines[i] = f'<br></code></pre>'
else:
if i > 0:
if count % 2 == 1:
line = line.replace("`", "\`")
line = line.replace("<", "&lt;")
line = line.replace(">", "&gt;")
line = line.replace(" ", "&nbsp;")
line = line.replace("*", "&ast;")
line = line.replace("_", "&lowbar;")
line = line.replace("-", "&#45;")
line = line.replace(".", "&#46;")
line = line.replace("!", "&#33;")
line = line.replace("(", "&#40;")
line = line.replace(")", "&#41;")
line = line.replace("$", "&#36;")
lines[i] = "<br>" + line
text = "".join(lines)
return text
def predict(history, prompt, max_length, top_p, temperature):
stop = StopOnTokens()
messages = []
if prompt:
messages.append({"role": "system", "content": prompt})
for idx, (user_msg, model_msg) in enumerate(history):
if prompt and idx == 0:
continue
if idx == len(history) - 1 and not model_msg:
messages.append({"role": "user", "content": user_msg})
break
if user_msg:
messages.append({"role": "user", "content": user_msg})
if model_msg:
messages.append({"role": "assistant", "content": model_msg})
model_inputs = tokenizer.apply_chat_template(messages,
add_generation_prompt=True,
tokenize=True,
return_tensors="pt").to(next(model.parameters()).device)
streamer = TextIteratorStreamer(tokenizer, timeout=60, skip_prompt=True, skip_special_tokens=True)
generate_kwargs = {
"input_ids": model_inputs,
"streamer": streamer,
"max_new_tokens": max_length,
"do_sample": True,
"top_p": top_p,
"temperature": temperature,
"stopping_criteria": StoppingCriteriaList([stop]),
"repetition_penalty": 1.2,
"eos_token_id": model.config.eos_token_id,
}
t = Thread(target=model.generate, kwargs=generate_kwargs)
t.start()
for new_token in streamer:
if new_token:
history[-1][1] += new_token
yield history
with gr.Blocks() as demo:
gr.HTML("""<h1 align="center">GLM-4-9B Gradio Simple Chat Demo</h1>""")
chatbot = gr.Chatbot()
with gr.Row():
with gr.Column(scale=3):
with gr.Column(scale=12):
user_input = gr.Textbox(show_label=False, placeholder="Input...", lines=10, container=False)
with gr.Column(min_width=32, scale=1):
submitBtn = gr.Button("Submit")
with gr.Column(scale=1):
prompt_input = gr.Textbox(show_label=False, placeholder="Prompt", lines=10, container=False)
pBtn = gr.Button("Set Prompt")
with gr.Column(scale=1):
emptyBtn = gr.Button("Clear History")
max_length = gr.Slider(0, 32768, value=8192, step=1.0, label="Maximum length", interactive=True)
top_p = gr.Slider(0, 1, value=0.8, step=0.01, label="Top P", interactive=True)
temperature = gr.Slider(0.01, 1, value=0.6, step=0.01, label="Temperature", interactive=True)
def user(query, history):
return "", history + [[parse_text(query), ""]]
def set_prompt(prompt_text):
return [[parse_text(prompt_text), "成功设置prompt"]]
pBtn.click(set_prompt, inputs=[prompt_input], outputs=chatbot)
submitBtn.click(user, [user_input, chatbot], [user_input, chatbot], queue=False).then(
predict, [chatbot, prompt_input, max_length, top_p, temperature], chatbot
)
emptyBtn.click(lambda: (None, None), None, [chatbot, prompt_input], queue=False)
demo.queue()
demo.launch(server_name="127.0.0.1", server_port=8000, inbrowser=True, share=True)
basic_demo/vllm_cli_demo.py 0 → 100644
View file @ 1bfbcff0
"""
This script creates a CLI demo with vllm backand for the glm-4-9b model,
allowing users to interact with the model through a command-line interface.
Usage:
- Run the script to start the CLI demo.
- Interact with the model by typing questions and receiving responses.
Note: The script includes a modification to handle markdown to plain text conversion,
ensuring that the CLI interface displays formatted text correctly.
"""
import time
import asyncio
from transformers import AutoTokenizer
from vllm import SamplingParams, AsyncEngineArgs, AsyncLLMEngine
from typing import List, Dict
MODEL_PATH = 'THUDM/glm-4-9b'
def load_model_and_tokenizer(model_dir: str):
engine_args = AsyncEngineArgs(
model=model_dir,
tokenizer=model_dir,
tensor_parallel_size=1,
dtype="bfloat16",
trust_remote_code=True,
gpu_memory_utilization=0.3,
enforce_eager=True,
worker_use_ray=True,
engine_use_ray=False,
disable_log_requests=True
# 如果遇见 OOM 现象，建议开启下述参数
# enable_chunked_prefill=True,
# max_num_batched_tokens=8192
)
tokenizer = AutoTokenizer.from_pretrained(
model_dir,
trust_remote_code=True,
encode_special_tokens=True
)
engine = AsyncLLMEngine.from_engine_args(engine_args)
return engine, tokenizer
engine, tokenizer = load_model_and_tokenizer(MODEL_PATH)
async def vllm_gen(messages: List[Dict[str, str]], top_p: float, temperature: float, max_dec_len: int):
inputs = tokenizer.apply_chat_template(
messages,
add_generation_prompt=True,
tokenize=False
)
params_dict = {
"n": 1,
"best_of": 1,
"presence_penalty": 1.0,
"frequency_penalty": 0.0,
"temperature": temperature,
"top_p": top_p,
"top_k": -1,
"use_beam_search": False,
"length_penalty": 1,
"early_stopping": False,
"stop_token_ids": [151329, 151336, 151338],
"ignore_eos": False,
"max_tokens": max_dec_len,
"logprobs": None,
"prompt_logprobs": None,
"skip_special_tokens": True,
}
sampling_params = SamplingParams(**params_dict)
async for output in engine.generate(inputs=inputs, sampling_params=sampling_params, request_id=f"{time.time()}"):
yield output.outputs[0].text
async def chat():
history = []
max_length = 8192
top_p = 0.8
temperature = 0.6
print("Welcome to the GLM-4-9B CLI chat. Type your messages below.")
while True:
user_input = input("\nYou: ")
if user_input.lower() in ["exit", "quit"]:
break
history.append([user_input, ""])
messages = []
for idx, (user_msg, model_msg) in enumerate(history):
if idx == len(history) - 1 and not model_msg:
messages.append({"role": "user", "content": user_msg})
break
if user_msg:
messages.append({"role": "user", "content": user_msg})
if model_msg:
messages.append({"role": "assistant", "content": model_msg})
print("\nGLM-4: ", end="")
current_length = 0
output = ""
async for output in vllm_gen(messages, top_p, temperature, max_length):
print(output[current_length:], end="", flush=True)
current_length = len(output)
history[-1][1] = output
if __name__ == "__main__":
asyncio.run(chat())
composite_demo/.gitignore 0 → 100644
View file @ 1bfbcff0
*venv
*.DS_Store
*model
*.idea/
# Created by https://www.toptal.com/developers/gitignore/api/python
# Edit at https://www.toptal.com/developers/gitignore?templates=python
### Python ###
# Byte-compiled / optimized / DLL files
__pycache__/
*.py[cod]
*$py.class
# C extensions
*.so
# Distribution / packaging
.Python
build/
develop-eggs/
dist/
downloads/
eggs/
.eggs/
lib/
lib64/
parts/
sdist/
var/
wheels/
share/python-wheels/
*.egg-info/
.installed.cfg
*.egg
MANIFEST
# PyInstaller
# Usually these files are written by a python script from a template
# before PyInstaller builds the exe, so as to inject date/other infos into it.
*.manifest
*.spec
# Installer logs
pip-log.txt
pip-delete-this-directory.txt
# Unit test / coverage reports
htmlcov/
.tox/
.nox/
.coverage
.coverage.*
.cache
nosetests.xml
coverage.xml
*.cover
*.py,cover
.hypothesis/
.pytest_cache/
cover/
# Translations
*.mo
*.pot
# Django stuff:
*.log
local_settings.py
db.sqlite3
db.sqlite3-journal
# Flask stuff:
instance/
.webassets-cache
# Scrapy stuff:
.scrapy
# Sphinx documentation
docs/_build/
# PyBuilder
.pybuilder/
target/
# Jupyter Notebook
.ipynb_checkpoints
# IPython
profile_default/
ipython_config.py
# pyenv
# For a library or package, you might want to ignore these files since the code is
# intended to run in multiple environments; otherwise, check them in:
# .python-version
# pipenv
# According to pypa/pipenv#598, it is recommended to include Pipfile.lock in version control.
# However, in case of collaboration, if having platform-specific dependencies or dependencies
# having no cross-platform support, pipenv may install dependencies that don't work, or not
# install all needed dependencies.
#Pipfile.lock
# poetry
# Similar to Pipfile.lock, it is generally recommended to include poetry.lock in version control.
# This is especially recommended for binary packages to ensure reproducibility, and is more
# commonly ignored for libraries.
# https://python-poetry.org/docs/basic-usage/#commit-your-poetrylock-file-to-version-control
#poetry.lock
# pdm
# Similar to Pipfile.lock, it is generally recommended to include pdm.lock in version control.
#pdm.lock
# pdm stores project-wide configurations in .pdm.toml, but it is recommended to not include it
# in version control.
# https://pdm.fming.dev/#use-with-ide
.pdm.toml
# PEP 582; used by e.g. github.com/David-OConnor/pyflow and github.com/pdm-project/pdm
__pypackages__/
# Celery stuff
celerybeat-schedule
celerybeat.pid
# SageMath parsed files
*.sage.py
# Environments
.env
.venv
env/
venv/
ENV/
env.bak/
venv.bak/
# Spyder project settings
.spyderproject
.spyproject
# Rope project settings
.ropeproject
# mkdocs documentation
/site
# mypy
.mypy_cache/
.dmypy.json
dmypy.json
# Pyre type checker
.pyre/
# pytype static type analyzer
.pytype/
# Cython debug symbols
cython_debug/
# PyCharm
# JetBrains specific template is maintained in a separate JetBrains.gitignore that can
# be found at https://github.com/github/gitignore/blob/main/Global/JetBrains.gitignore
# and can be added to the global gitignore or merged into this file. For a more nuclear
# option (not recommended) you can uncomment the following to ignore the entire idea folder.
#.idea/
### Python Patch ###
# Poetry local configuration file - https://python-poetry.org/docs/configuration/#local-configuration
poetry.toml
# ruff
.ruff_cache/
# LSP config files
pyrightconfig.json
# End of https://www.toptal.com/developers/gitignore/api/python
composite_demo/README.md 0 → 100644
View file @ 1bfbcff0
# GLM-4-9B Web Demo
Read this in [English](README_en.md)
![Demo webpage](assets/demo.png)
## 安装
我们建议通过 [Conda](https://docs.conda.io/en/latest/) 进行环境管理。
执行以下命令新建一个 conda 环境并安装所需依赖：
```bash
conda create -n glm-4-demo python=3.12
conda activate glm-4-demo
pip install -r requirements.txt
```
请注意，本项目需要 Python 3.10 或更高版本。
此外，使用 Code Interpreter 还需要安装 Jupyter 内核：
```bash
ipython kernel install --name glm-4-demo --user
```
您可以修改 `~/.local/share/jupyter/kernels/glm-4-demo/kernel.json` 来改变 Jupyter 内核的配置，包括内核的启动参数等。例如，若您希望在使用 All Tools 的 Python 代码执行能力时使用 Matplotlib 画图，可以在 `argv` 数组中添加 `"--matplotlib=inline"`。
若要使用浏览器和搜索功能，还需要启动浏览器后端。首先，根据 [Node.js](https://nodejs.org/en/download/package-manager)
官网的指示安装 Node.js，然后安装包管理器 [PNPM](https://pnpm.io) 之后安装浏览器服务的依赖：
```bash
cd browser
npm install -g pnpm
pnpm install
```
## 运行
1. 修改 `browser/src/config.ts` 中的 `BING_SEARCH_API_KEY` 配置浏览器服务需要使用的 Bing 搜索 API Key：
```diff
export default {
BROWSER_TIMEOUT: 10000,
BING_SEARCH_API_URL: 'https://api.bing.microsoft.com/v7.0',
BING_SEARCH_API_KEY: '<PUT_YOUR_BING_SEARCH_KEY_HERE>',
HOST: 'localhost',
PORT: 3000,
};
```
2. 文生图功能需要调用 CogView API。修改 `src/tools/config.py`
，提供文生图功能需要使用的 [智谱 AI 开放平台](https://open.bigmodel.cn) API Key：
```diff
BROWSER_SERVER_URL = 'http://localhost:3000'
IPYKERNEL = 'glm-4-demo'
ZHIPU_AI_KEY = '<PUT_YOUR_ZHIPU_AI_KEY_HERE>'
COGVIEW_MODEL = 'cogview-3'
```
3. 启动浏览器后端，在单独的 shell 中：
```bash
cd browser
pnpm start
```
4. 运行以下命令在本地加载模型并启动 demo：
```bash
streamlit run src/main.py
```
之后即可从命令行中看到 demo 的地址，点击即可访问。初次访问需要下载并加载模型，可能需要花费一定时间。
如果已经在本地下载了模型，可以通过 `export *_MODEL_PATH=/path/to/model` 来指定从本地加载模型。可以指定的模型包括：
- `CHAT_MODEL_PATH`: 用于 All Tools 模式与文档解读模式，默认为 `THUDM/glm-4-9b-chat`。
- `VLM_MODEL_PATH`: 用于 VLM 模式，默认为 `THUDM/glm-4v-9b`。
Chat 模型支持使用 [vLLM](https://github.com/vllm-project/vllm) 推理。若要使用，请安装 vLLM 并设置环境变量 `USE_VLLM=1`。
如果需要自定义 Jupyter 内核，可以通过 `export IPYKERNEL=<kernel_name>` 来指定。
## 使用
GLM-4 Demo 拥有三种模式：
- All Tools: 具有完整工具调用能力的对话模式，原生支持网页浏览、代码执行、图片生成，并支持自定义工具。
- 文档解读: 支持上传文档进行文档解读与对话。
- 多模态: 支持上传图像进行图像理解与对话。
### All Tools
本模式兼容 ChatGLM3-6B 的工具注册流程。
+ 代码能力，绘图能力，联网能力已经自动集成，用户只需按照要求配置对应的Key。
+ 本模式下不支持系统提示词，模型会自动构建提示词。
对话模式下，用户可以直接在侧边栏修改 top_p, temperature 等参数来调整模型的行为。
与模型对话时，模型将会自主决定进行工具调用。
![Tool calling](assets/tool.png)
由于原始结果可能较长，默认情况下工具调用结果被隐藏，可以通过展开折叠框查看原始的工具调用结果。
模型拥有进行网页搜索和 Python 代码执行的能力。同时，模型也可以连续调用多个工具。例如：
![Consecutive tool calling, 1](assets/web_plot_1.png)
此时模型通过调用浏览器工具进行搜索获取到了需要的数据，之后将会调用 Python 工具执行代码，利用 Matplotlib 绘图：
![Consecutive tool calling, 2](assets/web_plot_2.png)
如果提供了智谱开放平台 API Key，模型也可以调用 CogView 进行图像生成：
![Image generation](assets/cogview.png)
#### 自定义工具
可以通过在 `tool_registry.py` 中注册新的工具来增强模型的能力。只需要使用 `@register_tool`
装饰函数即可完成注册。对于工具声明，函数名称即为工具的名称，函数 docstring
即为工具的说明；对于工具的参数，使用 `Annotated[typ: type, description: str, required: bool]` 标注参数的类型、描述和是否必须。
例如，`get_weather` 工具的注册如下：
```python
@register_tool
def get_weather(
city_name: Annotated[str, 'The name of the city to be queried', True],
) -> str:
"""
Get the weather for `city_name` in the following week
"""
...
```
![The model uses tool to query the weather of Bangkok.](assets/weather.png)
### 文档解读
用户可以上传文档，使用 GLM-4-9B的长文本能力，对文本进行理解。可以解析 pptx，docx，pdf等文件。
+ 本模式下不支持工具调用和系统提示词。
+ 如果文本很长，可能导致模型需要的显存较高，请确认你的硬件配置。
![Doc reader demo](assets/doc_reader.png)
### 多模态
多模态模式下，用户可以利用 GLM-4V 的多模态理解能力，上传图像并与 GLM-4V 进行多轮对话：
用户可以上传图片，使用 GLM-4-9B的图像理解能力，对图片进行理解。
+ 本模式必须使用 glm-4v-9b 模型。
+ 本模式下不支持工具调用和系统提示词。
+ 模型仅能对一张图片进行理解和联系对话，如需更换图片，需要开启一个新的对话。
+ 图像支持的分辨率为 1120 x 1120
![VLM demo](assets/vlm.png)
composite_demo/README_en.md 0 → 100644
View file @ 1bfbcff0
# GLM-4-9B Web Demo
![Demo webpage](assets/demo.png)
## Installation
We recommend using [Conda](https://docs.conda.io/en/latest/) for environment management.
Execute the following commands to create a conda environment and install the required dependencies:
```bash
conda create -n glm-4-demo python=3.12
conda activate glm-4-demo
pip install -r requirements.txt
```
Please note that this project requires Python 3.10 or higher.
In addition, you need to install the Jupyter kernel to use the Code Interpreter:
```bash
ipython kernel install --name glm-4-demo --user
```
You can modify `~/.local/share/jupyter/kernels/glm-4-demo/kernel.json` to change the configuration of the Jupyter
kernel, including the kernel startup parameters. For example, if you want to use Matplotlib to draw when using the
Python code execution capability of All Tools, you can add `"--matplotlib=inline"` to the `argv` array.
To use the browser and search functions, you also need to start the browser backend. First, install Node.js according to
the instructions on the [Node.js](https://nodejs.org/en/download/package-manager)
official website, then install the package manager [PNPM](https://pnpm.io) and then install the browser service
dependencies:
```bash
cd browser
npm install -g pnpm
pnpm install
```
## Run
1. Modify `BING_SEARCH_API_KEY` in `browser/src/config.ts` to configure the Bing Search API Key that the browser service
needs to use:
```diff
export default {
BROWSER_TIMEOUT: 10000,
BING_SEARCH_API_URL: 'https://api.bing.microsoft.com/v7.0',
BING_SEARCH_API_KEY: '<PUT_YOUR_BING_SEARCH_KEY_HERE>',
HOST: 'localhost',
PORT: 3000,
};
```
2. The Wenshengtu function needs to call the CogView API. Modify `src/tools/config.py`
, provide the [Zhipu AI Open Platform](https://open.bigmodel.cn) API Key required for the Wenshengtu function:
```diff
BROWSER_SERVER_URL = 'http://localhost:3000'
IPYKERNEL = 'glm4-demo'
ZHIPU_AI_KEY = '<PUT_YOUR_ZHIPU_AI_KEY_HERE>'
COGVIEW_MODEL = 'cogview-3'
```
3. Start the browser backend in a separate shell:
```bash
cd browser
pnpm start
```
4. Run the following commands to load the model locally and start the demo:
```bash
streamlit run src/main.py
```
Then you can see the demo address from the command line and click it to access it. The first access requires downloading
and loading the model, which may take some time.
If you have downloaded the model locally, you can specify to load the model from the local
by `export *_MODEL_PATH=/path/to/model`. The models that can be specified include:
- `CHAT_MODEL_PATH`: used for All Tools mode and document interpretation mode, the default is `THUDM/glm-4-9b-chat`.
- `VLM_MODEL_PATH`: used for VLM mode, the default is `THUDM/glm-4v-9b`.
The Chat model supports reasoning using [vLLM](https://github.com/vllm-project/vllm). To use it, please install vLLM and
set the environment variable `USE_VLLM=1`.
If you need to customize the Jupyter kernel, you can specify it by `export IPYKERNEL=<kernel_name>`.
## Usage
GLM4 Demo has three modes:
- All Tools mode
- VLM mode
- Text interpretation mode
### All Tools mode
You can enhance the model's capabilities by registering new tools in `tool_registry.py`. Just use `@register_tool`
decorated function to complete the registration. For tool declarations, the function name is the name of the tool, and
the function docstring
is the description of the tool; for tool parameters, use `Annotated[typ: type, description: str, required: bool]` to
annotate the parameter type, description, and whether it is required.
For example, the registration of the `get_weather` tool is as follows:
```python
@register_tool
def get_weather(
city_name: Annotated[str, 'The name of the city to be queried', True],
) -> str:
"""
Get the weather for `city_name` in the following week
"""
...
```
This mode is compatible with the tool registration process of ChatGLM3-6B.
+ Code capability, drawing capability, and networking capability have been automatically integrated. Users only need to
configure the corresponding Key as required.
+ System prompt words are not supported in this mode. The model will automatically build prompt words.
## Text interpretation mode
Users can upload documents and use the long text capability of GLM-4-9B to understand the text. It can parse pptx, docx,
pdf and other files.
+ Tool calls and system prompt words are not supported in this mode.
+ If the text is very long, the model may require a high amount of GPU memory. Please confirm your hardware
configuration.
## Image Understanding Mode
Users can upload images and use the image understanding capabilities of GLM-4-9B to understand the images.
+ This mode must use the glm-4v-9b model.
+ Tool calls and system prompts are not supported in this mode.
+ The model can only understand and communicate with one image. If you need to change the image, you need to open a new
conversation.
+ The supported image resolution is 1120 x 1120
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment