Pinned Loading

1. mteb-br mteb-br Public

   MTEB Portuguese — Massive Text Embedding Benchmark for Brazilian Portuguese

   Python 9

2. isocronas isocronas Public

   JavaScript 2

3. embedding-vocab-trimmer embedding-vocab-trimmer Public

   How to trim an embedding model's vocabulary for a target language — training-free, no GPU. Example: EmbeddingGemma-300M → PT-BR at half the params.

   Python 26

4. colibri-embed colibri-embed Public

   🐦 Colibri — a ~157M Brazilian-Portuguese embedding model (MTEB(por) 0.6501). Vocab trim → multi-teacher distillation (Qwen3-4B+8B) → model soup.

   Python 2

5. gemma3moe-builder gemma3moe-builder Public

   Sparse-upcycle a dense Gemma-3-270m into a Mixture-of-Experts (gemma3moe) and serve it on llama.cpp (CPU). Upcycle, train, inspect routing, GGUF.

   Python 56

6. cadif cadif Public

   CADIF: a Composition Assay for De-confounding Instruction-Following

   Python 1