unslothai/unsloth ships Docker, multi-user accounts, AMD support, and training updates

Releases·Tool Release

Unsloth ships large performance release: Vulkan on AMD, MLX speedups, PyTorch 2.11

Unsloth added local-run support for Qwen3.8-Flash-Next and GLM-5.3-Flash with GGUF weights on Hugging Face, then shipped follow-ups enabling multi-token prediction for up to 2x faster inference and a major performance/bugfix release. The latest release switches AMD to Vulkan by default for a 20% prefill/decoding boost over ROCm and includes 200+ bug fixes. Practitioners gain faster local GGUF inference, one-click Hermes support, and MLX fine-tuning on Apple Silicon.

Unsloth added local-run support for Qwen3.8-Flash-Next and GLM-5.3-Flash with GGUF weights on Hugging Face, then shipped follow-ups enabling multi-token prediction for up to 2x faster inference and a major performance/bugfix release. The latest release switches AMD to Vulkan by default for a 20% prefill/decoding boost over ROCm and includes 200+ bug fixes. Practitioners gain faster local GGUF inference, one-click Hermes support, and MLX fine-tuning on Apple Silicon.

Practitioners shared local-inference benchmarks for Qwen3.8 27B on AMD consumer GPUs: one ran Ollama/ROCm vs llama.cpp Vulkan on an RX 7900 XTX, and another weighs upgrading from dual RTX 2060 12GB to an RX 6800 + 6800 XT pair. The posts serve as practical reference points for AMD multi-GPU llama.cpp performance.

Releases·Tool Releasefeatured

Unsloth announces official AMD GPU support for local LLM fine-tuning and inference

Unsloth now officially supports AMD hardware for local inference, fine-tuning, reinforcement learning, and deployment. The open-source tool works on Windows, Linux, and WSL with Radeon RX 9000/7000 series, Instinct MI350/MI300 GPUs, Strix Halo/Ryzen AI Max systems, and AMD CPUs for GPU-free inference. It claims up to 70% less VRAM for training and 80% less for reinforcement learning.

Unsloth now officially supports AMD hardware for local inference, fine-tuning, reinforcement learning, and deployment. The open-source tool works on Windows, Linux, and WSL with Radeon RX 9000/7000 series, Instinct MI350/MI300 GPUs, Strix Halo/Ryzen AI Max systems, and AMD CPUs for GPU-free inference. It claims up to 70% less VRAM for training and 80% less for reinforcement learning.

Releases·Tool Release

Unsloth Desktop adds command palette, packed 4-bit LoRA training, faster Laya decisions

Unsloth ships a Desktop update adding a command palette, shareable GGUF run settings, and clearer errors. The release trains LoRA directly on packed NVFP4, INT4, and MXFP4 checkpoints, cutting peak memory roughly in half, and speeds Laya decision models up to 4.1x.

Unsloth ships a Desktop update adding a command palette, shareable GGUF run settings, and clearer errors. The release trains LoRA directly on packed NVFP4, INT4, and MXFP4 checkpoints, cutting peak memory roughly in half, and speeds Laya decision models up to 4.1x.

Unsloth released Unsloth Desktop, the first desktop app to run and train AI models locally, available on Windows, macOS, and Linux. The open-source app supports MLX, GGUF, diffusion models, and connects Claude Code and Codex to local LLMs, with claims of 2x faster training and 70% less VRAM usage.