From 287ef12bf04aad6884e4652ca34b1a81ea600568 Mon Sep 17 00:00:00 2001 From: localai-org-maint-bot Date: Sun, 30 Aug 2026 09:01:05 +0200 Subject: [PATCH] feat(gallery): add Granite 4.2 variants (#11779) Add the 3B, 8B, and 30B safetensors checkpoints as one vLLM variant family so LocalAI can select the largest build that fits. Configure the parsers and sampling defaults recommended for Granite reasoning and tool calls. Assisted-by: Codex:gpt-5 Co-authored-by: localai-org-maint-bot <306269227+localai-org-maint-bot@users.noreply.github.com> --- gallery/index.yaml | 106 +++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 106 insertions(+) diff --git a/gallery/index.yaml b/gallery/index.yaml index cab4117e0..202ce3d14 100644 --- a/gallery/index.yaml +++ b/gallery/index.yaml @@ -16341,6 +16341,112 @@ - filename: antares-1b-q8_0.gguf sha256: f0b147c42d9056d59eb3f0ab51d4d57a2259bf1b4e37ae986ef96a6c2e0ac6b7 uri: huggingface://mitkox/antares-1b-Q8_0-GGUF/antares-1b-q8_0.gguf +- &granite-4-2-3b-vllm + name: "granite-4.2-3b:vllm" + variants: + - model: granite-4.2-8b:vllm + - model: granite-4.2-30b:vllm + url: "github:mudler/LocalAI/gallery/vllm.yaml@master" + urls: + - https://huggingface.co/ibm-granite/granite-4.2-3b + description: | + Granite 4.2 3B is IBM's compact dense reasoning model for code generation, + tool calling, agentic workflows, multilingual chat, and long-context tasks. + This entry serves the bfloat16 safetensors with vLLM and supports a + 128K-token context. It is the smallest fallback in a family that also offers + the higher-capacity 8B and 30B checkpoints as variants. + license: apache-2.0 + icon: https://cdn-avatars.huggingface.co/v1/production/uploads/639bcaa2445b133a4e942436/CEW-OjXkRkDNmTxSu8Egh.png + tags: + - llm + - vllm + - gpu + - granite + - granite-4.2 + - 3b + - reasoning + - code + - function-calling + - agentic + - multilingual + - long-context + - instruction-tuned + last_checked: "2026-08-29" + overrides: + context_size: 131072 + options: + - reasoning_parser:granite + - tool_parser:qwen3_coder + parameters: + model: ibm-granite/granite-4.2-3b + temperature: 1 + top_p: 0.95 +- !!merge <<: *granite-4-2-3b-vllm + name: "granite-4.2-8b:vllm" + variants: [] + urls: + - https://huggingface.co/ibm-granite/granite-4.2-8b + description: | + Granite 4.2 8B is IBM's mid-sized dense reasoning model for code generation, + tool calling, agentic workflows, multilingual chat, and long-context tasks. + This entry serves the higher-capacity bfloat16 safetensors with vLLM and + supports a 128K-token context. + tags: + - llm + - vllm + - gpu + - granite + - granite-4.2 + - 8b + - reasoning + - code + - function-calling + - agentic + - multilingual + - long-context + - instruction-tuned + overrides: + context_size: 131072 + options: + - reasoning_parser:granite + - tool_parser:qwen3_coder + parameters: + model: ibm-granite/granite-4.2-8b + temperature: 1 + top_p: 0.95 +- !!merge <<: *granite-4-2-3b-vllm + name: "granite-4.2-30b:vllm" + variants: [] + urls: + - https://huggingface.co/ibm-granite/granite-4.2-30b + description: | + Granite 4.2 30B is IBM's largest dense Granite 4.2 reasoning model for code + generation, tool calling, agentic workflows, multilingual chat, and + long-context tasks. This entry serves the bfloat16 safetensors with vLLM and + supports a 128K-token context. + tags: + - llm + - vllm + - gpu + - granite + - granite-4.2 + - 30b + - reasoning + - code + - function-calling + - agentic + - multilingual + - long-context + - instruction-tuned + overrides: + context_size: 131072 + options: + - reasoning_parser:granite + - tool_parser:qwen3_coder + parameters: + model: ibm-granite/granite-4.2-30b + temperature: 1 + top_p: 0.95 - name: ibm-granite_granite-4.0-h-small url: github:mudler/LocalAI/gallery/granite4.yaml@master urls: