Unsloth implementation

#1
by Kernel - opened

Is it faster than original model? What was unsloth optimization here

Unsloth AI org

We had to change the modeling files to make it compatible with the latest transformers!
Notebook: https://colab.research.google.com/github/unslothai/notebooks/blob/main/nb/Deepseek_OCR_2_(3B).ipynb

installing the latest vllm nightly as per the tutorial (Version: 0.15.0rc2.dev54+g6bf3b46d7) throws an error because the model arch isn't registered. Value error, Model architectures ['DeepseekOCR2ForCausalLM'] are not supported for now. Is there a way around this?

Unsloth AI org

installing the latest vllm nightly as per the tutorial (Version: 0.15.0rc2.dev54+g6bf3b46d7) throws an error because the model arch isn't registered. Value error, Model architectures ['DeepseekOCR2ForCausalLM'] are not supported for now. Is there a way around this?

vLLM doesn't support it as of late apologies

Sign up or log in to comment