From 7a225863997546355ce71bec4e639820f9e095b3 Mon Sep 17 00:00:00 2001 From: localai-org-maint-bot Date: Mon, 10 Aug 2026 10:17:44 +0200 Subject: [PATCH] feat(gallery): add Shieldstral 1.0 3B variants (#11441) MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Add Q4_K_M and Q8_0 llama.cpp builds of Mistral AI’s multimodal safety classifier, including the shared Pixtral projector. Assisted-by: Codex:gpt-5 Co-authored-by: localai-org-maint-bot <306269227+localai-org-maint-bot@users.noreply.github.com> --- gallery/index.yaml | 76 ++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 76 insertions(+) diff --git a/gallery/index.yaml b/gallery/index.yaml index ad45363bf..4d9a90e47 100644 --- a/gallery/index.yaml +++ b/gallery/index.yaml @@ -31904,6 +31904,82 @@ - filename: llama-cpp/mmproj/mmproj-mistralai_Ministral-3-8B-Reasoning-2512-f32.gguf sha256: 92252621cb957949379ff81ee14b15887d37eade3845a6e937e571b98c2c84c2 uri: huggingface://unsloth/Ministral-3-8B-Reasoning-2512-GGUF/mmproj-F32.gguf +- &shieldstral-1-0-3b + name: "shieldstral-1.0-3b" + variants: + - model: shieldstral-1.0-3b-q8 + url: "github:mudler/LocalAI/gallery/virtual.yaml@master" + urls: + - https://huggingface.co/mistralai/Shieldstral-1.0-3B + - https://huggingface.co/noctrex/Shieldstral-1.0-3B-GGUF + description: | + Shieldstral 1.0 3B is Mistral AI's compact, policy-adaptive multimodal safety + classifier. It evaluates text, images, or combined inputs against a + natural-language safety policy and answers yes or no. The model supports + twelve languages and a recommended context length of up to 32K tokens. + This entry uses the Q4_K_M GGUF quantization and includes the Pixtral vision + projector. + license: apache-2.0 + tags: + - llm + - gguf + - safety + - moderation + - classifier + - vision + - multimodal + - multilingual + - gpu + - cpu + overrides: + backend: llama-cpp + known_usecases: + - chat + mmproj: llama-cpp/mmproj/Shieldstral-1.0-3B/mmproj-F16.gguf + options: + - use_jinja:true + parameters: + context_size: 32768 + model: llama-cpp/models/Shieldstral-1.0-3B/Shieldstral-1.0-3B-Q4_K_M.gguf + temperature: 0 + max_tokens: 1 + template: + use_tokenizer_template: true + files: + - filename: llama-cpp/models/Shieldstral-1.0-3B/Shieldstral-1.0-3B-Q4_K_M.gguf + sha256: df76fe59cf4083202fdf693527892774acd435ee8e9023c5a3aaa0c2b35d49ec + uri: huggingface://noctrex/Shieldstral-1.0-3B-GGUF/Shieldstral-1.0-3B-Q4_K_M.gguf + - filename: llama-cpp/mmproj/Shieldstral-1.0-3B/mmproj-F16.gguf + sha256: 765ad52bfeec11a8f58d81ef2738a8a72184770666d990505d69276e788858e1 + uri: huggingface://noctrex/Shieldstral-1.0-3B-GGUF/mmproj-F16.gguf +- !!merge <<: *shieldstral-1-0-3b + name: "shieldstral-1.0-3b-q8" + variants: [] + description: | + Shieldstral 1.0 3B is Mistral AI's compact, policy-adaptive multimodal safety + classifier. This higher-quality variant uses the Q8_0 GGUF quantization and + includes the Pixtral vision projector. + overrides: + backend: llama-cpp + known_usecases: + - chat + mmproj: llama-cpp/mmproj/Shieldstral-1.0-3B/mmproj-F16.gguf + options: + - use_jinja:true + parameters: + context_size: 32768 + model: llama-cpp/models/Shieldstral-1.0-3B-q8/Shieldstral-1.0-3B-Q8_0.gguf + temperature: 0 + max_tokens: 1 + template: + use_tokenizer_template: true + files: + - filename: llama-cpp/models/Shieldstral-1.0-3B-q8/Shieldstral-1.0-3B-Q8_0.gguf + sha256: 35b755bed2d473fb3d88f7d1d7b83203bd9f0c1b8bff42624e6fd9231d89d3c4 + uri: huggingface://noctrex/Shieldstral-1.0-3B-GGUF/Shieldstral-1.0-3B-Q8_0.gguf + - filename: llama-cpp/mmproj/Shieldstral-1.0-3B/mmproj-F16.gguf + sha256: 765ad52bfeec11a8f58d81ef2738a8a72184770666d990505d69276e788858e1 + uri: huggingface://noctrex/Shieldstral-1.0-3B-GGUF/mmproj-F16.gguf - name: mistralai_ministral-3-3b-instruct-2512-multimodal url: github:mudler/LocalAI/gallery/mistral-0.3.yaml@master urls: