Downloads · 30 days
0
Isaac74/qwen3-0.6b-lightweight-semantic-mathlib-search-adapter
qwen3-0.6b-lightweight-semantic-mathlib-search-adapter is a sentence similarity model from Isaac74. Use it when you need a score for how close two texts are. It is set up for peft. The card lists the license as apache-2.0.
What this is: A small, reproducible demo adapter for Qwen/Qwen3-Embedding-0.6B that enables lightweight, LLM-free semantic search over mathlib (Lean 4). Supports both natural language and formula-based search (can be…
Downloads · 30 days
0
Access
Public
Updated Aug 9, 2025
Repo size
2.7 GB
Likes
0
Public
Click a slice to open those files.
.idx2.6 GB · 95%
From the Hugging Face model README
What this is:
A small, reproducible demo adapter for Qwen/Qwen3-Embedding-0.6B that enables lightweight, LLM-free semantic search over mathlib (Lean 4).
Supports both natural language and formula-based search (can be used in combination) for finding Lean theorems or mathematical statements.
Does not support LaTeX syntax search yet.
Data is built from ~180k Lean theorems extracted via LeanDojo for Lean 4.20.1 (there may be minor gaps vs. the full mathlib).
Part of the AITP 2025 submission: “Towards Lightweight and LLM-Free Semantic Search for mathlib4.”
adapter/adapter_config.json & adapter/adapter_model.safetensors — LoRA weights for the encoderfaiss/<*.idx> — HNSW FAISS index (built for cosine/inner-product search)faiss/*public_index.jsonl — public mapping { idx, full_name, statement }https://github.com/IsaacLi74/Lightweight-and-LLM-Free-Semantic-Search-for-mathlib4