feat(detect): unlimited-ocr / deepseek-ocr → vision_ocr family - #86
Merged
Conversation
…r family Unlimited-OCR is the DeepSeek-OCR architecture; its GGUF general.architecture is "unlimited-ocr" (llama.cpp mtmd support in ggml-org/llama.cpp#24969, merged 2026-06-24). Add unlimited-ocr / deepseek-ocr / deepseek2-ocr → vision_ocr so the inspect verdict reports "supported" (deploy now) instead of needs_family. Requires a llama-server built after that merge — a serving image rebuild. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
"supported" only means a family contract exists — the node's llama-server must also be new enough to LOAD the arch, a separate gate. resolve_family now carries a runtime_note for archs whose serving support landed recently (unlimited-ocr / deepseek-ocr → "needs a llama-server built after llama.cpp#24969; rebuild the serving image if it won't load"), inspect_repo surfaces it, and the HF search panel shows it as an amber caveat under the verdict. A user isn't misled into a deploy that fails at load on an old build. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Owner
Author
|
Added the runtime-support honesty on top of the arch mapping (commit 4a1e308):
So a 14 tests pass. |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What
Maps the
unlimited-ocr/deepseek-ocr/deepseek2-ocrarchitectures to thevision_ocrfamily in the arch→family registry, so/hf/inspectreturns supported (deploy now) for Unlimited-OCR instead ofneeds_family.Why
UnlimitedOCRForCausalLMIS the DeepSeek-OCR architecture (SAM-ViT-B + CLIP-L/14 vision tower → projector → DeepSeek-V2 MoE decoder). Its GGUFgeneral.architectureisunlimited-ocr. llama.cpp added mtmd support in ggml-org/llama.cpp#24969 (merged 2026-06-24).Runtime gate (separate from this change)
Serving it needs a
llama-serverbuilt after that merge — the Dockerfile clones llama.cpp HEAD unpinned, so a serving image rebuild picks it up. The definitive check is empirical: deploy and read the deployment's runtime log.Tests
tests/test_arch_detect.py— all three arch spellings resolve tovision_ocr(supported). 13 passed.🤖 Generated with Claude Code