Issues

Issues

• [Feature] Update Gemma 4 model family to Unsloth Dynamic UD3.1

  unslothai/unsloth#17573 · opened on Nov 12, 2026

• [Bug] DiffusionGemma load by Studio ignores selected context and parallelism settings

  unslothai/unsloth#17558 · opened on Nov 11, 2026
  1

• [Feature] kromoss support in studio

  unslothai/unsloth#17552 · opened on Nov 11, 2026
  1

• [Bug] ROCm 7.14 / MI300X (gfx942): stride assertion in _scaled_dot_product_flash_attention_backward – MATH fallback required

  unslothai/unsloth#17548 · opened on Nov 11, 2026
  1

• [Bug] Studio: YuE2 can sing with Instrumental enabled

  unslothai/unsloth#17473 · opened on Nov 11, 2026

• [Bug] unsloth docker image tags are incorrect

  unslothai/unsloth#17470 · opened on Nov 11, 2026

• [Feature] ship an rpm

  unslothai/unsloth#17445 · opened on Nov 10, 2026

• [Feature] Infer max token length default per model based on context length instead of a 32K default in custom connections

  unslothai/unsloth#17444 · opened on Nov 10, 2026

• [Feature] Make audio model selection more visible above “Load model into”

  unslothai/unsloth#17441 · opened on Nov 10, 2026

• [Feature] Add an optional speaker accent hint to transcription

  unslothai/unsloth#17440 · opened on Nov 10, 2026
  1

• [Feature] Make Model Hub search category-aware and capability tags clickable

  unslothai/unsloth#17439 · opened on Nov 10, 2026

• Nested folders in projects

  unslothai/unsloth#17431 · opened on Nov 10, 2026

• Qwen3.8-Flash-Next UD 3.1 GGUFs vs current MTP heads: draft acceptance drops ~74% -> 56-66% at production temperature (Ryzen AI Max+ 395) — are 3.1-matched heads planned?

  unslothai/unsloth#17402 · opened on Nov 10, 2026
  3

• [Feature] Add Ukrainian (uk) language coverage to Unsloth Dynamic GGUF imatrix calibration datasets

  unslothai/unsloth#17371 · opened on Nov 10, 2026
  1

• [Feature] Add UI toggle to disable API monitor logs

  unslothai/unsloth#17366 · opened on Nov 10, 2026

• [Feature] Please add Batches for New Chat

  unslothai/unsloth#17333 · opened on Nov 10, 2026
  1

• [Feature] Studio: persist "Full access" permission mode (session-only today)

  unslothai/unsloth#17297 · opened on Nov 9, 2026
  2

• Settings: composer settings are split across the Appearance and Chat tabs, with no single owner

  unslothai/unsloth#17163 · opened on Nov 8, 2026

• [Feature] Allow sourcing /opt/intel/oneapi/setvars.sh before running llama-server to enable SYCL

  unslothai/unsloth#17048 · opened on Nov 7, 2026
  2

• [Feature] Add a fullscreen mode on the desktop software.

  unslothai/unsloth#17016 · opened on Nov 7, 2026

• Studio: repeated full prompt processing across turns with Qwen hybrid model after tool use

  unslothai/unsloth#17015 · opened on Nov 7, 2026
  4

• [Feature] Add a "Keep all checkpoints" option for when an Image LoRA completes

  unslothai/unsloth#16980 · opened on Nov 7, 2026
  2

• [Feature] StrataFlux: Memory-Hierarchy-Native Low-Bit LLM Inference

  unslothai/unsloth#16967 · opened on Nov 6, 2026

• [Feature] Quantization-Aware Distillation (QAD) for PQ2 Models

  unslothai/unsloth#16916 · opened on Nov 6, 2026

• [Bug] Sync remaining account-level preferences to server-side storage

  unslothai/unsloth#16912 · opened on Nov 6, 2026
  3