XLM-R Biomedical African QA Dense Retriever
This model is a fine-tuned SentenceTransformer retriever based on FacebookAI/xlm-roberta-base.
Intended use
Dense retrieval for multilingual biomedical/health QA answer-bank matching.
Recommended document format:
Question: <question>
Answer: <answer>
Training objective
MultipleNegativesRankingLoss using question/document positive pairs.
Loading
from sentence_transformers import SentenceTransformer
model = SentenceTransformer("ahlkal/xlmr-african-qa-retriever")
Use in the main answer-bank pipeline
DENSE_MODEL = "ahlkal/xlmr-african-qa-retriever"
RERANKER_MODEL = "BAAI/bge-reranker-v2-m3"
TOP_K = 60
DENSE_TOP_K = 60
RERANK_K = 45
CE_BATCH_SIZE = 8
Important diagnostic
Compare oracle_rouge1_at_60 against BAAI/bge-m3 and intfloat/multilingual-e5-large.