DiffusionGemma on LM Studio — Status & Working Alternatives
llama.cpp PR #24423 is still in draft. LM Studio can’t bundle DiffusionGemma until the PR lands in llama.cpp mainline. In the meantime — here are 3 ways to run DiffusionGemma today.
PR #24423 still in draft. LM Studio native support blocked. Ecosystem status →
Option 1: Ollama Bridge Workaround (Works Today)
Use a custom Modelfile that wraps the llama.cpp diffusion branch. You get the full DiffusionGemma experience inside Ollama — ollama run diffusiongemma — while waiting for native support.
Best for: developers who want Ollama's familiar CLI and don't mind a one-time bridge setup. ~30 min (mostly downloading + building).
Option 2: llama.cpp — Build from PR Branch
Build llama.cpp from the PR branch — PR #24423 is still in draft, so you need to fetch it explicitly. Full control over build flags (CUDA, Metal, CPU).
Best for: developers comfortable with cmake/make who want maximum performance and don't need a GUI. ~15 min (clone + build + download GGUF).
Option 3: vLLM (Production-Grade, Day-0 Support)
vLLM has supported DiffusionGemma since launch — Google collaborated directly on the integration. If you're serving to users or need an OpenAI-compatible API, this is the production path.
Best for: production deployments, API serving, multi-user setups. ~10 min (pip install + download model, no build needed).
Why Doesn't LM Studio Load DiffusionGemma Yet?
LM Studio bundles its own build of llama.cpp — the engine that powers model loading and inference. PR #24423 is still in draft — not in mainline llama.cpp yet. LM Studio can't bundle until the PR lands. Once it does, DiffusionGemma GGUF files will load natively.
LM Studio updates their bundled llama.cpp from mainline releases. The next LM Studio release that updates its llama.cpp dependency will include diffusion-gemma architecture support. Until then, DiffusionGemma GGUF files will show "unsupported model architecture" in LM Studio — no workaround exists specifically for LM Studio because you can't swap its internal engine.
The three alternatives above all route around this by using the llama.cpp PR branch directly or by using a completely different engine (vLLM).
PR #24423 In Draft — What's Needed for LM Studio
llama.cpp PR #24423 is still in draft. The remaining steps for LM Studio: the PR needs to land in llama.cpp mainline, then LM Studio releases a new version bundling the updated engine.
| What | Status |
|---|---|
| llama.cpp PR #24423 | Draft ⚠️ |
| LM Studio engine update | Pending — no ETA |
| 3 alternatives | Working today |
LM Studio typically adopts new llama.cpp releases within 1–2 versions. Once they ship, this page will have native LM Studio loading instructions. Track full ecosystem →
Got it running locally. Ready for production?
The Production Guide covers Docker Compose, GPU tuning, 20+ error fixes, multi-user serving — 8 chapters that each save you 5+ hours. $39 $29, lifetime updates.
Last updated: July 1, 2026