Change8

v0.1.512-beta

📦 unslothView on GitHub →
11 features🐛 14 fixes🔧 15 symbols

Summary

This release introduces local support for Kimi K3 and DeepSeek v4 Flash models via Unsloth Dynamic GGUFs, alongside parallel chat capabilities and a new Deep Research mode. It also brings significant improvements to AMD and Intel GPU support, DoRA training, and various installer, MLX, export, and inference fixes.

Migration Steps

  1. Update Unsloth using the provided installation scripts for macOS, Linux, WSL, or Windows.

✨ New Features

  • Kimi K3 and DeepSeek v4 Flash models can now run locally with Unsloth Dynamic GGUFs.
  • Unsloth can keep multiple chats generating in parallel, with isolated progress indicators and stop controls for each conversation.
  • New Deep Research mode allows local models to plan, read, and cite sources for research tasks.
  • Improved AMD and Intel GPU support, including better detection and ROCm inference/training reliability.
  • DoRA training is now available alongside LoRA and full fine-tuning.
  • Large exports can utilize all visible GPUs to overcome GPU 0 memory limitations.
  • MLX support enhanced with streaming datasets, continued pretraining, callbacks, and better VLM and LoRA support.
  • Unsloth and saving utilities now function without bitsandbytes.
  • The `unsloth start --as-subagent` command allows delegation of tasks to a local Unsloth model from agents like Codex, Claude Code, and Pi.
  • XET download now auto-fallbacks to HTTP if stalled, with reduced memory usage for more stable and efficient downloads.
  • Intel XPU enables local chat and training on Intel Arc and Data Center GPUs.

🐛 Bug Fixes

  • Fixed incorrect `flash_attention_2` model output.
  • Fixed `.json` datasets and Qwen3.5/3.6 MoE and GRPO notebook setup.
  • Fixed 4-bit NaNs, library conflicts, and long RDNA startup stalls on AMD GPUs.
  • Fixed Studio desktop reliability issues.
  • Fixed resume training crash recovery and MLX checkpoints.
  • Fixed loading split GGUFs from the local HF cache in Studio.
  • Fixed local CLI streamed generation error handling.
  • Fixed Studio: don't let a malformed HF token empty the model picker's Recommended list.
  • Fixed Studio: make tab navigation feel immediate.
  • Fixed Studio: show HF token tick only after validation.
  • Fixed (rocm): prepend system ROCm libs on native Linux to avoid bundled HIP crash.
  • Fixed studio: extend the _grouped_mm null-kernel guard to Linux ROCm RDNA4 (gfx1201).
  • Fixed (install.ps1): use ordinal IndexOf when stripping index URL credentials.
  • Fixed install.sh: route Strix to the AMD arch index on rocm7.2 (#7264) + PCI detection hint.

Affected Symbols