Introducing AMD support

[danielhanchen]

Hey everyone! This release brings local LLM training and inference to AMD GPUs across Windows, WSL and Linux. Read our AMD Blog + Guide

Starting today, our AMD collaboration, custom Triton kernels, and math algorithms enables you to train and run 500+ models across AMD's Radeon, Instinct, Ryzen and data center GPUs, up to 2× faster with 70% less VRAM and no accuracy loss. Optimized ROCm builds also support GGUF & Safetensors inference.

Train LLMs Locally on AMD

• Train, run RL, chat with and deploy models locally on AMD GPUs.
• More reliable AMD GPU detection and installation across Windows, WSL and Linux.
• Improved ROCm compatibility for AMD MI300X and MI325X GPUs.
• Remote access Unsloth via unsloth studio --secure through free HTTPS via Cloudflare

Run Larger Models on Your Hardware

• Use automatic GPU placement or choose exactly which GPUs and model layers to use.
• Move MoE expert layers into system memory to help larger models fit.
• Split models across multiple GPUs or use Tensor Parallelism.
• Save hardware settings separately for each model and quant.

Faster Chat Restarts and More Reliable Downloads

• Resume long chats without rebuilding the full conversation after an idle model frees its VRAM.
• Stalled Hugging Face XET downloads automatically retry over standard HTTP.
• Existing GGUF files are reused instead of being downloaded again whenever a model loads.

Better Search, Tools and Agents

• Web search can now read PDF papers, manuals and other PDF results.
• Parallel tool calls, reasoning output and tool retries work more reliably.
• A new opt-in MCP endpoint lets compatible AI clients inspect models and training history, start or stop training, load checkpoints, validate recipes and export GGUFs.
  • Enable it with UNSLOTH_STUDIO_ENABLE_MCP=1 and set the required bearer token with UNSLOTH_STUDIO_MCP_TOKEN.

Training and Export Fixes

• Text-only training with multimodal models no longer truncates long examples before packing.
• Fine-tuned Qwen3.5 and Qwen3.6 MTP models now export correctly to GGUF.
• Fixed a Windows permission error that could stop GGUF exports during the final write step.

In Case You Missed It

Our previous Studio release added Dynamic NVFP4 models, deeper personalization, seven new display languages, safer agents and Vulkan GPU acceleration.

Dynamic NVFP4

Unsloth Dynamic NVFP4 keeps accuracy-sensitive layers in FP8 or BF16 while running the rest in W4A4. On NVIDIA Blackwell GPUs, this enables up to 2.5x faster inference, while calibrated FP8 KV caches provide up to 2x longer context.

Read the Dynamic NVFP4 guide and explore our expanded NVFP4 collection, including Qwen3.6, Qwen3.5, Inkling, GLM-4.7 Flash and Gemma 4.

Personalize Your Studio

Choose between Standard, Classic and Minimal color palettes, each with light and dark modes. You can also customize colors, import fonts and adjust font size, contrast, reduced motion and more.

Studio also includes a Voice settings tab for dictation, custom dictionary settings and read-aloud.

New Languages

Unsloth Studio is now available in French, German, Spanish, Hindi, Arabic, Russian and Korean, alongside Chinese (Simplified), Japanese and Portuguese (Brazil). Browser-language auto-detection is now the default.

Safer Agents

A four-level tool-call permission selector-Ask, Approve for me, Off and Full access-provides finer control over agents. Agent workspace isolation and safer installer checks also reduce the risk of unintended changes.

Updating Unsloth

To update Unsloth or install a fresh Unsloth, use the commands below:

macOS, Linux, WSL:

curl -fsSL https://unsloth.ai/install.sh | sh

Windows:

irm https://unsloth.ai/install.ps1 | iex

What's Changed in Unsloth

• Bump install.sh / install.ps1 pin to unsloth>=2026.7.3 by @danielhanchen in Bump install.sh / install.ps1 pin to unsloth>=2026.7.3 #7155
• fix(dataprep): smart_chunk_text single-chunk path leaks internal tensor type when eos_token_id is None by @chuenchen309 in fix(dataprep): smart_chunk_text single-chunk path leaks internal tensor type when eos_token_id is None #7151
• Fix Settings layout overflow by @shimmyshimmer in Fix Settings layout overflow #7167
• Keep nested dropdown menus on screen by @shimmyshimmer in Keep nested dropdown menus on screen #7168
• Fix DeepSeek reasoning test shim by @shimmyshimmer in Fix DeepSeek reasoning test shim #7169
• Harden desktop release token permissions by @wasimysaid in Harden desktop release token permissions #7172
• Studio: clickable sidebar settings cog, long name truncation, Canvas menu opt-in by @shimmyshimmer in Studio: clickable sidebar settings cog, long name truncation, Canvas menu opt-in #7171
• CI: opt tool-calling smoke tests out of the chat tool approval gate by @danielhanchen in CI: opt tool-calling smoke tests out of the chat tool approval gate #7162
• Fix Inkling reasoning-effort coercion for duck-typed engine stand-ins by @danielhanchen in Fix Inkling reasoning-effort coercion for duck-typed engine stand-ins #7158
• fix config cards clipping content at narrow window widths by @NilayYadav in fix config cards clipping content at narrow window widths #7146
• Studio: use one shared Hugging Face token across Settings and training by @NilayYadav in Studio: use one shared Hugging Face token across Settings and training #7152
• Studio: don't drop parallel tool calls after an internal no-op by @oobabooga in Studio: don't drop parallel tool calls after an internal no-op #7157
• Studio: extract text from PDF web results by @oobabooga in Studio: extract text from PDF web results #7154
• don't kill live llama-servers when a new Studio instance starts by @NilayYadav in don't kill live llama-servers when a new Studio instance starts #7182
• fix(save): unsloth_push_to_hub_gguf(save_method="lora") raises NameError by @chuenchen309 in fix(save): unsloth_push_to_hub_gguf(save_method="lora") raises NameError #7193
• Feat/model picker per model config by @Sneakr in Feat/model picker per model config #6647
• fix(studio): recover stalled Hub downloads over HTTP by @rodboev in fix(studio): recover stalled Hub downloads over HTTP #6858
• fix(mlx): relax Metal context-store timeout by default by @Lyxot in fix(mlx): relax Metal context-store timeout by default #7141
• Studio: don't re-download updated GGUFs on load by @oobabooga in Studio: don't re-download updated GGUFs on load #7209
• Fix Studio reasoning channel rendering by @Lyxot in Fix Studio reasoning channel rendering #7121
• Fix Studio safetensors tool reprompt reasoning leakage by @Lyxot in Fix Studio safetensors tool reprompt reasoning leakage #7134
• Studio: don't apply nest_asyncio on plain CLI starts (breaks asyncio on Python 3.14+) by @NilayYadav in Studio: don't apply nest_asyncio on plain CLI starts (breaks asyncio on Python 3.14+) #7186
• Propagate fp8 block_size before the early return in get_lora_parameters_bias by @vineethsaivs in Propagate fp8 block_size before the early return in get_lora_parameters_bias #7189
• feat(studio): expose an opt-in MCP control plane by @RitwijParmar in feat(studio): expose an opt-in MCP control plane #7191
• fix(tokenizer): check for tokenizer.model after saving it, not before by @chuenchen309 in fix(tokenizer): check for tokenizer.model after saving it, not before #7194
• Stabilize Studio regression tests by @shimmyshimmer in Stabilize Studio regression tests #7192
• fix(dataprep): skip .jsonl lines that are valid JSON but not objects by @chuenchen309 in fix(dataprep): skip .jsonl lines that are valid JSON but not objects #7195
• fix(studio): equal padding in the dataset source segmented control by @shimmyshimmer in fix(studio): equal padding in the dataset source segmented control #7230
• fix canvas staying open after starting a New Chat by @NilayYadav in fix canvas staying open after starting a New Chat #7229
• Allow API key for Ollama connections by @shimmyshimmer in Allow API key for Ollama connections #7173
• Studio: keep the Bypass permissions pill label when composer pills collapse by @shimmyshimmer in Studio: keep the Bypass permissions pill label when composer pills collapse #7231
• fix(studio): honor MLX adapter state in compare mode by @Lyxot in fix(studio): honor MLX adapter state in compare mode #7196
• Refresh the fastapi C2-loop scan baseline entry for the current release by @danielhanchen in Refresh the fastapi C2-loop scan baseline entry for the current release #7223
• Studio: enforce 60s minimum on idle auto-unload TTL (0 stays off) by @NilayYadav in Studio: enforce 60s minimum on idle auto-unload TTL (0 stays off) #7185
• Replace standalone Studio wording with Unsloth by @shimmyshimmer in Replace standalone Studio wording with Unsloth #7221
• Studio: hide the RAG embedder and llama.cpp probe from the hub cached inventory by @shimmyshimmer in Studio: hide the RAG embedder and llama.cpp probe from the hub cached inventory #7018
• fix(registry): don't register deepseek models at import time by @chuenchen309 in fix(registry): don't register deepseek models at import time #7227
• studio: show the active run's saved config in the Training Progress popover by @hakanbaysal in studio: show the active run's saved config in the Training Progress popover #7217
• test(version-compat): keep GRPO fake-run logits finite on CPU by @danielhanchen in test(version-compat): keep GRPO fake-run logits finite on CPU #7247
• Studio: GPU memory configuration for GGUF models by @oobabooga in Studio: GPU memory configuration for GGUF models #6414
• Give opencode real timeout headroom in Local Agent Guides CI by @danielhanchen in Give opencode real timeout headroom in Local Agent Guides CI #7235
• Installer: allow torch 2.11.x on the CUDA install path (fresh install + studio) by @danielhanchen in Installer: allow torch 2.11.x on the CUDA install path (fresh install + studio) #6959
• fix(chat_templates): bind loop_messages when default_system_message is None by @chuenchen309 in fix(chat_templates): bind loop_messages when default_system_message is None #7199
• Installer: preserve the previous torch release across every flavor and vendor on re-runs by @danielhanchen in Installer: preserve the previous torch release across every flavor and vendor on re-runs #7250
• Studio: fix stale GGUF load-marker ordering test after llama_extra_args relocation by @danielhanchen in Studio: fix stale GGUF load-marker ordering test after llama_extra_args relocation #7252
• Route OpenCode yolo aliases to native auto mode by @Imagineer99 in Route OpenCode yolo aliases to native auto mode #7187
• Pin the Hermes remote installer and harden consent by @Imagineer99 in Pin the Hermes remote installer and harden consent #7179
• Translate PWD for WSL-launched Windows agents by @Imagineer99 in Translate PWD for WSL-launched Windows agents #7111
• persist llama.cpp KV cache across idle auto-unload (slot save/restore) by @NilayYadav in persist llama.cpp KV cache across idle auto-unload (slot save/restore) #7204
• Fix text-only VLM CPT packing truncation by @alkinun in Fix text-only VLM CPT packing truncation #7211
• feat(studio): add UNSLOTH_LLAMA_CPP_BACKEND env var to force CPU fallback [Bug] Unsloth Studio built llama.cpp with Vulkan instead of CPU only #7213 by @VectorCipher in feat(studio): add UNSLOTH_LLAMA_CPP_BACKEND env var to force CPU fallback #7213 #7228
• Experimental: correct varlen sample packing for hybrid linear-attention models by @danielhanchen in Experimental: correct varlen sample packing for hybrid linear-attention models #7249
• install: let UNSLOTH_TORCH_INDEX_FAMILY / _URL override CUDA wheel detection by @danielhanchen in install: let UNSLOTH_TORCH_INDEX_FAMILY / _URL override CUDA wheel detection #6692
• Studio: Data settings tab, uploaded files manager, quant pinning, and chat image preview fix by @shimmyshimmer in Studio: Data settings tab, uploaded files manager, quant pinning, and chat image preview fix #7029
• Studio: fix per-GPU VRAM reporting on Windows ROCm by @danielhanchen in Studio: fix per-GPU VRAM reporting on Windows ROCm #7238
• Studio: do not show Run for embedding-only non-GGUF models in the Model Hub by @danielhanchen in Studio: do not show Run for embedding-only non-GGUF models in the Model Hub #7245
• Studio: make Stop and stall deadlines interrupt a wedged stream portably by @danielhanchen in Studio: make Stop and stall deadlines interrupt a wedged stream portably #7236
• Studio: clarify tool permission controls by @shimmyshimmer in Studio: clarify tool permission controls #7181
• Update README with latest features and Unsloth Start by @Imagineer99 in Update README with latest features and Unsloth Start #7258
• Studio: lock the last enabled GPU switch instead of silently ignoring it by @oobabooga in Studio: lock the last enabled GPU switch instead of silently ignoring it #7259
• Fix unsloth start on Windows: agent install, PATH resolution, and local model selection by @danielhanchen in Fix unsloth start on Windows: agent install, PATH resolution, and local model selection #7257
• Fix On Device model picker startup ordering by @Imagineer99 in Fix On Device model picker startup ordering #6994
• Studio: validate Hugging Face tokens before use by @oobabooga in Studio: validate Hugging Face tokens before use #7261
• Studio (Windows): keep prompt caching on full GPU offload by @danielhanchen in Studio (Windows): keep prompt caching on full GPU offload #7260

New Contributors

• @chuenchen309 made their first contribution in fix(dataprep): smart_chunk_text single-chunk path leaks internal tensor type when eos_token_id is None #7151
• @VectorCipher made their first contribution in feat(studio): add UNSLOTH_LLAMA_CPP_BACKEND env var to force CPU fallback #7213 #7228

Full Changelog: v0.1.49-beta...v0.1.50-beta

What's Changed in Unsloth-Zoo

• mlx: fixes for bnb dequant, report_to, early stopping, and NEFTune by @oobabooga in mlx: fixes for bnb dequant, report_to, early stopping, and NEFTune unsloth-zoo#851
• MLX: clear error when mlx-lm/mlx-vlm is too old for a QK-norm arch (gemma4/qwen3_5) by @danielhanchen in MLX: clear error when mlx-lm/mlx-vlm is too old for a QK-norm arch (gemma4/qwen3_5) unsloth-zoo#845
• MLX: fix gemma-4-e2b (gemma4) audio-tower conv double transpose on load by @danielhanchen in MLX: fix gemma-4-e2b (gemma4) audio-tower conv double transpose on load unsloth-zoo#879
• Add Mixtral MoE experts patch for Split LoRA (fix 4bit QLoRA forward crash) by @danielhanchen in Add Mixtral MoE experts patch for Split LoRA (fix 4bit QLoRA forward crash) unsloth-zoo#881
• Forward declared tools to the DiffusionGemma visual server by @oobabooga in Forward declared tools to the DiffusionGemma visual server unsloth-zoo#864
• Add mixtral_moe to temporary_patches import test list by @danielhanchen in Add mixtral_moe to temporary_patches import test list unsloth-zoo#884
• feat(mlx): add DDP training support by @Lyxot in feat(mlx): add DDP training support unsloth-zoo#841
• feat(mlx): support distributed inference loading by @Lyxot in feat(mlx): support distributed inference loading unsloth-zoo#854
• fix(mlx): preserve inference norm dtypes by @Lyxot in fix(mlx): preserve inference norm dtypes unsloth-zoo#844
• fix(mlx): normalize legacy tokenizer sidecars for VLM loads by @Lyxot in fix(mlx): normalize legacy tokenizer sidecars for VLM loads unsloth-zoo#846
• fix(mlx): support text label parity datasets by @Lyxot in fix(mlx): support text label parity datasets unsloth-zoo#793
• MLX: fix gated-delta (Qwen3.5) training and add fused backward kernels by @Lyxot in MLX: fix gated-delta (Qwen3.5) training and add fused backward kernels unsloth-zoo#754
• (GRPO) Fix SAPO on version TRL > 0.27.2 and add Luspo from TRL >= 0.28.0 by @marcandrelarochelle in (GRPO) Fix SAPO on version TRL > 0.27.2 and add Luspo from TRL >= 0.28.0 unsloth-zoo#869
• (GRPO) Add importance sampling modes (TRL >= 0.26.0), Switch to vllm_importance_sampling_clip (support for TRL >= 1.6.0) and Fix VESPO not using the importance sampling ratio by @marcandrelarochelle in (GRPO) Add importance sampling modes (TRL >= 0.26.0), Switch to vllm_importance_sampling_clip (support for TRL >= 1.6.0) and Fix VESPO not using the importance sampling ratio unsloth-zoo#870
• saving: fold per-expert gpt-oss 4bit expert LoRA into merged_16bit by @danielhanchen in saving: fold per-expert gpt-oss 4bit expert LoRA into merged_16bit unsloth-zoo#885
• Fix gpt_oss mask factory signature introspection on transformers 5.x by @danielhanchen in Fix gpt_oss mask factory signature introspection on transformers 5.x unsloth-zoo#880
• Fix CPU test drift on transformers 5.5.0 and the MLX bitsandbytes stub by @danielhanchen in Fix CPU test drift on transformers 5.5.0 and the MLX bitsandbytes stub unsloth-zoo#886
• Guard MLX distributed eval against zero-token batches by @danielhanchen in Guard MLX distributed eval against zero-token batches unsloth-zoo#888
• Fix chat template marker auto-detection for Zephyr's context-dependent tokenization by @danielhanchen in Fix chat template marker auto-detection for Zephyr's context-dependent tokenization unsloth-zoo#899
• Persist torch.compile artifacts across processes with Mega-cache by @danielhanchen in Persist torch.compile artifacts across processes with Mega-cache unsloth-zoo#901
• Fix ModuleNotFoundError crash in fla_vendor's is_flash_linear_attention_available probe by @amareshhebbar in Fix ModuleNotFoundError crash in fla_vendor's is_flash_linear_attention_available probe unsloth-zoo#897
• Fix false DownloadStallError for non-standard sharded weight filenames by @olsh in Fix false DownloadStallError for non-standard sharded weight filenames unsloth-zoo#900
• Fix false DownloadStallError for diffusers components with non-standard sharded weight filenames by @danielhanchen in Fix false DownloadStallError for diffusers components with non-standard sharded weight filenames unsloth-zoo#902
• tokenizer_utils: match IGNORED_TOKENIZER_NAMES case-insensitively in fix_untrained_tokens by @oobabooga in tokenizer_utils: match IGNORED_TOKENIZER_NAMES case-insensitively in fix_untrained_tokens unsloth-zoo#823
• Fix GGUF export crash for fine-tuned MTP models (Qwen3.5) by @oobabooga in Fix GGUF export crash for fine-tuned MTP models (Qwen3.5) unsloth-zoo#904
• fix: ROCm compatibility — guard SM-arch paths, FlashInfer, and mem_get_info by @zhihuidu-amd in fix: ROCm compatibility — guard SM-arch paths, FlashInfer, and mem_get_info unsloth-zoo#910
• Fix Windows GGUF export PermissionError [Errno 13]: anchor converter --outfile to input_folder by @LeoBorcherding in Fix Windows GGUF export PermissionError [Errno 13]: anchor converter --outfile to input_folder unsloth-zoo#907
• Fix silent expert-weight transposition in MoE preprocess_weight by @BardiaKoopah in Fix silent expert-weight transposition in MoE preprocess_weight unsloth-zoo#913
• Replace standalone Studio wording with Unsloth by @shimmyshimmer in Replace standalone Studio wording with Unsloth unsloth-zoo#914
• Fix Gemma 4 audio training crash by @BardiaKoopah in Fix Gemma 4 audio training crash unsloth-zoo#917
• Fix Gemma 4 MLX GGUF tensor namespace restoration by @Lyxot in Fix Gemma 4 MLX GGUF tensor namespace restoration unsloth-zoo#905

---

This discussion was created from the release Introducing AMD support.