v0.1.481-beta

✨ New Features

• Unsloth Studio can now export to NVFP4, FP8, and imatrix GGUFs after training.
• Unsloth Studio can act as a llama-swap API system.
• Added Japanese and Brazilian Portuguese language support to the Unsloth UI.
• MLX support is now included.
• Safetensors tool calling and healing support is now available.
• GRPO training is now 1.3x faster.
• HTTP fallback added for stalled downloads.
• Improved offline mode functionality.
• MoE training is now 3 to 5x faster.
• Support added for DeepSeek-V4-Flash with Thinking toggles and improved chat template.
• API requests can opt into model auto switching between local GGUFs.
• API now defaults to safe model swapping: unknown model names keep the current model and avoid surprise downloads.
• /v1/models endpoint now returns clean model IDs instead of local .gguf file paths.
• Idle auto-unload feature added to free VRAM after inactivity.
• API clients can control tool-call healing per request and opt into extra nudge retry for malformed tool markup.
• Unsloth now allows selecting multiple export formats simultaneously.
• Support for portable FP8/INT8 exports, GGUF LoRA, and source-matched exports.
• `save_pretrained_merged` now supports compressed FP8/FP4 exports.
• Unsloth export UI includes imatrix GGUF and compressed export options.
• Exporting multiple checkpoints now avoids more repeated base-model downloads.
• FP8, INT8, and GGUF-LoRA exports now respect `trust_remote_code`.
• GGUF export handles models with missing quantization settings more reliably.
• RAG attachments can now use whole-document context.
• File chat now reads more PDFs and Word documents correctly, including right-to-left text, Indic text, and DOCX tables.
• Local RAG checks are less likely to fail due to proxy settings.
• Embedding model customization added, featuring Hugging Face search and a reorganized settings tab.
• GRPO training now supports sequence packing by default for old/ref log-probability calculations.
• GRPO now avoids repeating the same shared prompt across grouped completions.
• GRPO logit scaling now works correctly with DDP-wrapped models.
• Full fine-tuning now uses the correct precision on V100 and other GPUs without bf16 support.
• RL trainers no longer reject valid full fine-tuning precision settings.
• MoE LoRA detection now targets the correct expert MLP layers.
• Grouped MoE can now turn on automatically for loaded and PEFT models.
• FP8 quantization now handles odd tensor shapes and scale formats more reliably.
• FP8 dequant kernels now use safer offset handling.
• Llama 3 RoPE scaling is preserved with Transformers v5.