diff --git a/gallery/index.yaml b/gallery/index.yaml index 7388eb7e0580..b9aaebd6f7e4 100644 --- a/gallery/index.yaml +++ b/gallery/index.yaml @@ -22169,3 +22169,28 @@ - filename: Zirel-2.i1-Q4_K_S.gguf sha256: 9856e987f5f59c874a8fe26ffb2a2c5b7c60b85186131048536b3f1d91a235a6 uri: huggingface://mradermacher/Zirel-2-i1-GGUF/Zirel-2.i1-Q4_K_S.gguf +- !!merge <<: *granite4 + name: "granite-4.0-h-small-mxfp4_moe" + urls: + - https://huggingface.co/noctrex/Granite-4.0-H-Small-MXFP4_MOE-GGUF + description: | + **Granite-4.0-H-Small** + *32B-parameter, long-context, MoE-based instruction-tuned language model by IBM* + + Granite-4.0-H-Small is a powerful, instruction-finetuned, 32-billion-parameter large language model built on a decoder-only Mixture-of-Experts (MoE) architecture. Designed for enterprise and advanced applications, it excels in instruction following, tool calling, multilingual dialogue, code generation, and reasoning tasks. + + Key Features: + - **Architecture**: MoE transformer with GQA, Mamba2, SwiGLU activation, and shared embeddings. + - **Context Length**: Up to 128K tokens for long-context understanding. + - **Capabilities**: Text generation, summarization, question answering, RAG, code completion, function calling, and multilingual support (14+ languages). + - **Performance**: Top-tier results on benchmarks like MMLU, GSM8K, HumanEval, and IFEval. + - **Safety & Alignment**: Enhanced via supervised fine-tuning and reinforcement learning; includes safety prompts and robust alignment. + + Built upon the *Granite-4.0-H-Small-Base*, this model is ideal for building intelligent AI assistants, enterprise automation tools, and research applications. + + **Original Model**: [ibm-granite/granite-4.0-h-small](https://huggingface.co/ibm-granite/granite-4.0-h-small) + **License**: Apache 2.0 + **Use Case**: Enterprise AI, intelligent agents, code and content generation, multilingual applications. + overrides: + parameters: + model: noctrex/Granite-4.0-H-Small-MXFP4_MOE-GGUF