Skip to content
Kitbay

train-sentence-transformers

Apache-2.0Updated on Nov 11, 202611,334GitHub starsIncludes scripts

A Hugging Face skill for training or fine-tuning models with the sentence-transformers library: dense or static embedding models, cross-encoder rerankers, SPLADE-style sparse encoders and ColBERT-style multi-vector encoders.

What the skill teaches

A Hugging Face skill for training or fine-tuning models with the sentence-transformers library: dense or static embedding models, cross-encoder rerankers, SPLADE-style sparse encoders and ColBERT-style multi-vector encoders. The SKILL.md acts as a router: it helps the agent pick the model type, lists the reference files to read in full, sets defaults and constraints, and tells the agent to start from one of the bundled example training scripts rather than writing one from scratch. Bundled scripts cover distillation, LoRA, Matryoshka, multilingual and multi-dataset training, plus hard-negative mining.

Capabilities

  • Fine-tune embedding models with sentence-transformers
  • Train cross-encoder rerankers
  • Train sparse and multi-vector encoders
  • Mine hard negatives for training data