DermL2V Adapter

DermL2V is a semantic encoder specialized for unstructured dermatology text. It maps free-form dermatology narratives and questions to dense embeddings for semantic search, retrieval, and similarity-based comparison.

This repository provides the DermL2V PEFT adapter and the Attribute-Aware Relevance Aggregation module, the model's pooling mechanism for constructing text embeddings. The underlying Llama base model is not included and must be obtained separately.

Required model chain

meta-llama/Meta-Llama-3.1-8B-Instruct
  -> McGill-NLP/LLM2Vec-Meta-Llama-31-8B-Instruct-mntp
  -> McGill-NLP/LLM2Vec-Meta-Llama-31-8B-Instruct-mntp-supervised
  -> DermL2V_adapter

The base model is subject to the Llama 3.1 Community License. Users must obtain access to the gated base model and comply with its license and use policy.

The standard adapter_model.safetensors filename is intentional: PEFT uses this filename for automatic adapter loading.