Studio: decide diffusion routing before the SWA resolver

[danielhanchen]

Problem

Loading a DiffusionGemma GGUF in Studio could fail with llama-server failed to start. Check that the GGUF file is valid, even though llama-diffusion-cli runs the same model fine.

_read_gguf_metadata assigned self._is_diffusion after calling _resolve_swa_pattern, both inside one try/except. The resolver reaches into transformers/HF, which can raise for an architecture transformers does not know (diffusion-gemma). When it raised, the shared except swallowed it and left _is_diffusion False, so the model was routed to plain llama-server (which cannot serve a block-diffusion GGUF) instead of the diffusion runner. It only reproduced on hosts where the SWA pattern was not already cached or inline in the GGUF.

Fix

• Set _is_diffusion right after the KV parse loop (it only needs arch / canvas_seen), before the resolver, so a resolver error can no longer drop the diffusion routing.
• Skip the SWA resolver for diffusion models: they do not use Studio's SWA pattern (the visual server reads its own metadata), and it can raise for them.

Verification

Monkeypatching _resolve_swa_pattern to raise on a real DiffusionGemma GGUF: _is_diffusion is now True (was False), the resolver is invoked 0 times, and the remaining metadata (context_length, chat_template, reasoning, tools) is still read. No change for non-diffusion models.

[gemini-code-assist]

Code Review

This pull request updates the _read_gguf_metadata method in llama_cpp.py to determine whether a model is a diffusion model earlier in the metadata parsing process. It also skips the sliding window attention (SWA) resolver for diffusion models to prevent potential exceptions on architectures that the resolver does not recognize. There are no review comments to assess, so no feedback is provided.

Important

The consumer version of Gemini Code Assist on GitHub is being sunset. Starting June 18, 2026, new organization installations will be blocked, and all code review activity will officially cease on July 17, 2026.
For more details on the timeline and next steps, please review the Help Documentation.