antalvdb/olifant-speculative-ibases
Updated
Language models built on TiMBL instead of neural networks, and their use as CPU draft models for speculative decoding.
Note Olifant as a draft model for speculative decoding: four trained models, a generation method for Transformers and a llama.cpp adapter (INLG 2026).
Note A small Olifant model that loads through the Transformers API.
Note Larger conversational Olifant models in three sizes.
OLIFANT memory-based LLM explainability demo
Note Demo: see which training examples support each prediction.
Generate text using a secure API with confidence highlights
Note Demo: text generation with Olifant.
Note The paper that introduces Olifant.
Note Towards Green Text Generation: Memory-Based Speculative Decoding (INLG 2026)