glide-the

你好，在2.0的贡献我们没有计划维护，目前主分支已经是3.0的代码，感谢你对本贡献，期待你的下次参与

I agree

https://github.com/THUDM/CogVideo/pull/268 Exporting Huggingface Diffusers lora LoRA Weights from SAT Checkpoints diffuser's draft pr can try this operation https://github.com/huggingface/diffusers/pull/9412

@KihongK I don't see any fine-tuning of sft work logger error .

There are still compatibility issues between the base and chat models. **Please do not merge this PR for now.**

The code now supports **GLM variant models**, including **LLaMA-style** and **GPT-2-style** vocabularies. Tested inference compatibility for the following base models: - https://huggingface.co/THUDM/glm-4-9b - https://huggingface.co/THUDM/glm-4-9b-hf ### Current Status Inference works with...

> Excellent! I've been using GLM-4-32B and the tool calling format is non-standard (GLM-4 sample code maps to and from their tool calling format which is new-line delimited rather than...

@johnpyp @ngxson Hi ,I added this code to implement the glm func tools behavior support for function call behavior, add in the EOG detection logic for src/llama-vocab.cpp#L1976-L1977 pr: https://github.com/ggml-org/llama.cpp/pull/13339

Here are a few key differences in the diffusers framework used in our publicly released SAT fine-tuning code: **LoRA weights have a rank parameter**, with the 2B transformer model defaulting...

The SAT framework is somewhat different from the Diffusers framework, and the trained weights need to be inferred separately [sat infer lora ](https://github.com/THUDM/CogVideo/blob/main/sat/README.md) If diffuser inference is needed, you need...