GitHub · Where software is built

8000 GitHub · Where software is built

DeepSeek-OCR support is here!
#3555 · shimmyshimmer opened on Nov 4, 2025
6

Labels Milestones New issue

[Bug] save_pretrained_gguf fails due to llama.cpp/unsloth_convert_hf_to_gguf.py not found on jupyter in Docker

#3635

· kishida opened

on Nov 24, 2025

[Bug] save_pretrained_merged("merged_16bit") re-downloads FP16 weights into output .cache instead of using HF cache

#3633

· shimmyshimmer opened

on Nov 24, 2025

[Feature] Support for loading and accelerating encoder-based HuggingFace models

feature request

fixed - pending confirmation

#3632

· Abishek0070 opened

on Nov 24, 2025

UnslothVisionDataCollator() with streaming datasets.

#3630

· poilly54 opened

on Nov 24, 2025

[Bug] Models already trained - getting stuck at training run

#3627

· Maxtheconquerer opened

on Nov 21, 2025

[Bug] num_generations seems to be ignored in GRPO rollout

#3626

· yuxizhe opened

on Nov 21, 2025

[Bug] Some tokenizers broken when using <code>FastLanguageModel</code>

fixed - pending confirmation

#3624

· BramVanroy opened

on Nov 20, 2025

[Bug] Llama-4 loading error: AttributeError: SequentialLlama4TextExperts has no attribute down_proj

#3622

· THEOLIN opened

on Nov 20, 2025

[Bug] Qwen2.5-VL + streaming HF dataset: RuntimeError: shape '[0, 4, -1]' is invalid in visual encoder

#3617

· K-Hooshanfar opened

on Nov 19, 2025

[Feature] Support for Multi Turn GRPO

feature request

#3615

· dipta007 opened

on Nov 19, 2025

[Bug] Running GRPO-trained Gemma 3 models with inference

#3613

· Maxtheconquerer opened

on Nov 18, 2025

[Feature] Add support Longcat-flash compressed model

feature request

#3608

· eezhang123 opened

on Nov 18, 2025

0