GitHub · Where software is built

DeepSeek-OCR support is here!
#3555 · shimmyshimmer opened on Nov 4, 2025
1

Labels Milestones New issue

Multi-GPU Support for GRPO Training with Vision-Language Models (VLM)

feature request

#3571

· TranMinhThang-dev opened

on Nov 8, 2025

[Feature] deepseek-coder

feature request

#3570

· binaryblood opened

on Nov 8, 2025

[Feature] Add support to train Hunyuan Image 3.0

feature request

#3562

· nitinh12 opened

on Nov 6, 2025

[Bug] Cannot load qwen3-vl series with lora adapter on vllm.

good first issue

#3560

· poilly54 opened

on Nov 6, 2025

[Bug] Cannot save IBM Hybrid models in 4-bit safetensors format

#3558

· Sweaterdog opened

on Nov 6, 2025

DeepSeek-OCR support is here!

#3555

· shimmyshimmer opened

on Nov 4, 2025

[Bug] No GPU detected when following DGX Spark Manual

#3553

· weoieoeo opened

on Nov 4, 2025

[Bug] bug in GRPO with FSDP2

#3551

· Aurorana opened

on Nov 4, 2025

[Bug] Cannot get validation loss, TypeError: Unsupported types (<class 'unsloth.models._utils.EmptyLogits'>) passed to <code>_pad_across_processes</code>. Only nested list/tuple/dicts of objects that are valid for <code>is_torch_tensor</code> should be passed.

#3549

· Faris-Faiz opened

on Nov 4, 2025

[Issue] FastLlamaModel Class doesn't use revision argument at all

#3544

· sabilmakbar opened

on Nov 3, 2025

[Bug] RuntimeError: Trying to backward through the graph a second time

#3543

· nguyendinhduybigtreetc opened

on Nov 3, 2025

[Bug] OOM with UNSLOTH_VLLM_STANDBY=1 + fast_inference=True (4-bit)

currently fixing

#3542

· l-besiege-l opened

on Nov 1, 2025