Ecosyste.ms: Timeline
Browse the timeline of events for every public repo on GitHub. Data updated hourly from GH Archive.
sixsixcoder closed an issue on THUDM/GLM-4
ollama 加载 glm-4-9b-chat 胡言乱语
### System Info / 系統信息 cuda: 12.6 transformer: 4.44.0 OS: win10 python: 3.11.4 ollama: 0.3.8 & 0.2.3 配置: RTX3090 12700kf ### Who can help? / 谁可以帮助到您？ _No response_ ### Information ...
sixsixcoder closed an issue on THUDM/GLM-4
GLM-4V-9B fine-tuning error
### System Info / 系統信息 Traceback (most recent call last): File "/home/sa/swift/swift/cli/sft.py", line 5, in sft_main() File "/home/sa/swift/swift/utils/run_utils.py", line 32, in x_main resul...
sixsixcoder closed an issue on THUDM/GLM-4
Agent流式输出停止符问题
### System Info / 系統信息 调用方式：langchain_openai 模型：zhipuAI官方API的glm4 ### Who can help? / 谁可以帮助到您？ @wwewwt @Sengxian @davidlvxin @codazzy ### Information / 问题信息 - [X] The official exam...
sixsixcoder closed an issue on THUDM/GLM-4
Uncaught exception: Traceback (most recent call last) 运行web端的glm4v模型提示报错
### System Info / 系統信息 Uncaught exception: Traceback (most recent call last): File "D:\Big_model\ChatGLM\GLM-4-main\composite_demo\src\main.py", line 288, in main for response, chat_history in cli...
sixsixcoder closed an issue on THUDM/GLM-4
请教关于算法原理
请问GLM 4v是如何做到高分辨图像适配输入的？与CogVLM的区别？  请问图中High-Resolution Cross-Module在项目代码哪个位置可以体现出来 谢谢！
sixsixcoder closed an issue on THUDM/GLM-4
l微调后推理输出乱码
### System Info / 系統信息 Ubuntu 22.04 Python 3.10.12 CUDA 12.1 torch 2.4.0 transformers 4.44.0 ### Who can help? / 谁可以帮助到您？ _No response_ ### Information / 问题信息 - [ ] The official example sc...
sixsixcoder closed an issue on THUDM/GLM-4
本地glm4-9B模型使用function call 功能的问题
### System Info / 系統信息 python =3.10 运行了官方仓库的openai_api_server.py代码，测试本地模型的function call ### Who can help? / 谁可以帮助到您？ @z ### Information / 问题信息 - [X] The official example scripts / 官方的示例脚本 - ...
sixsixcoder closed an issue on THUDM/GLM-4
为啥system消息会传两次？
我在看basic_demo里面的openai_server.py 里面有一个process_message的函数 他里面的判断都是判断tools_choices是否是"none"，然后我调用openai_request那个文件，他传的都是None 然后在最后一个处理前，msg已经有system和user了，然后最后还有一个判断，insert了system到最开始，就导致实际处理有两...
sixsixcoder pushed 4 commits to main sixsixcoder/sglang
sixsixcoder created a comment on an issue on THUDM/GLM-4
可以参考https://github.com/THUDM/GLM-4/issues/568
sixsixcoder opened a pull request on sgl-project/sglang
Add GLM-4 TextGeneration Model support for SGLang
<!-- Thank you for your contribution! We appreciate it. The following guidelines will help improve your pull request and facilitate feedback. If anything is unclear, don't hesitate to submit your p...
sixsixcoder pushed 2 commits to glm-4 sixsixcoder/sglang
sixsixcoder pushed 1 commit to glm-4 sixsixcoder/sglang
- Simplify batch result resolution (#1735) b121bc0
sixsixcoder created a branch on sixsixcoder/sglang
glm-4 - SGLang is a fast serving framework for large language models and vision language models.
sixsixcoder pushed 41 commits to main sixsixcoder/sglang
- Fix memory leak during abort (#1674) f1088e0
- Fix filter_batch function call (#1681) b6b4094
- Add OLMo model (#1676) a5114b6
- Add a new event loop (#1677) 9116b28
- Fix srt dependency (#1685) d10b933
- [Event] Add online meetup meeting link (#1686) e4b367b
- Launch a thread to overlap CPU and GPU (#1687) dbec2f1
- Returning a per request metric for number of cached_tokens read (#1599) ecb8bad
- add orjson for jsonresponse (#1688) b0facb3
- Update README.md (#1689) d19cc0b
- Add date to logging messages (#1623) (#1679) 2782132
- Update the transformers version in CI (#1690) 02f7f3e
- Use SGLang imports for linear layer (#1696) 5ab20cc
- feat: radix tree code optimize (#1697) b170930
- ORJson. Faster Json serialization (#1694) e5db40d
- Fix the failed unit tests (#1699) 30ee363
- Fix failed ci tests on long prompts; Better error messages for embedding models (#1700) 7feba41
- Fix engine unit test (#1701) dd3809f
- Fix mixed batch for multi modal models (#1702) d17d19e
- Add matched_stop token or str to distinguish between eos or stop str finish_reason generation (#1684) a95d558
- and 21 more ...
sixsixcoder closed an issue on THUDM/GLM-4
8卡lora微调训练，数据集AdvertiseGen，loss从3.2613开始到1000多个step之后一直在2.8 2.9左右震荡，请问这是正常的吗
### Feature request / 功能建议 <img width="886" alt="image" src="https://github.com/user-attachments/assets/d360a32c-30bb-4c68-a7a0-0d02cbc00104"> 在2000个step eval后的结果{'eval_rouge-1': 57.990059834558...
sixsixcoder created a comment on an issue on THUDM/GLM-4
> @tomsonsgs 你好，请问这个问题怎么解决的呢？ > > 已解决，原始代码里<|assistant|>的token也被预测，但这个token无法被训练拟合导致一直有最低的loss存在，改变下数据准备代码即可 请问你是如何修改的代码，可以分享一下经验吗
sixsixcoder created a comment on a pull request on sgl-project/sglang
> You can see the llava for example > > https://github.com/sgl-project/sglang/blob/d19cc0b9c9520328f29dc88f827caa2081458382/python/sglang/srt/models/llava.py#L156-L167 > > > and run the htt...
sixsixcoder reopened an issue on THUDM/GLM-4
l微调后推理输出乱码
### System Info / 系統信息 Ubuntu 22.04 Python 3.10.12 CUDA 12.1 torch 2.4.0 transformers 4.44.0 ### Who can help? / 谁可以帮助到您？ _No response_ ### Information / 问题信息 - [ ] The official example sc...
sixsixcoder reopened an issue on THUDM/GLM-4
8卡lora微调训练，数据集AdvertiseGen，loss从3.2613开始到1000多个step之后一直在2.8 2.9左右震荡，请问这是正常的吗
### Feature request / 功能建议 <img width="886" alt="image" src="https://github.com/user-attachments/assets/d360a32c-30bb-4c68-a7a0-0d02cbc00104"> 在2000个step eval后的结果{'eval_rouge-1': 57.990059834558...
sixsixcoder created a comment on an issue on THUDM/GLM-4
> @sixsixcoder 想问下支持使用glm 4v 9b int4进行推理嘛？ 暂不支持
sixsixcoder created a comment on an issue on THUDM/GLM-4
> > 那么问题应该是你的数据集有问题，具体的你可以去[huggingface/datasets](https://github.com/huggingface/datasets/issues)提问 > > 这里我在719处添加718之前可以进行剧烈的数据集为什么也不可以啊？ 您单独运行这个huggingface加载数据集的代码也会报同样的错，说明这个问题不出在GLM-4的仓库代...
sixsixcoder created a comment on an issue on THUDM/GLM-4
这样的话应该是您的数据集有问题，具体的您可以去[huggingface/datasets](https://github.com/huggingface/datasets/issues)提问
sixsixcoder created a comment on an issue on THUDM/GLM-4
Currently glm-4v-9b only supports bf16 type, and inference takes up about 28G cuda memory
sixsixcoder created a comment on an issue on THUDM/GLM-4
> > 尝试一下这个代码会不会出错 > > ```python > > from datasets import load_dataset > > DATA_DIR = "YOUR/DATA/DIR" > > data_files = {"train": "train.jsonl", "dev": "dev.jsonl"} > > dataset = load_dataset(DA...
sixsixcoder created a comment on an issue on THUDM/GLM-4
bitsandbytes通常在int4类型量化时会用到，如果你不用int4类型推理，应该不会出问题，如果确实需要这个包，你可以参考bitsandbytes官方的一些用法