End-to-end Question Answering on AGORA (test)
74.7RelevancyBase ColBERT / DPO Mistral
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Base ColBERT / DPO MistralRetriever=Base ColBERTv2, Generator=Mistral-7B-Instruct aligned with DPO2026.03 | 74.7 | 60.1 | |
| Base ColBERT / Base MistralRetriever=Base ColBERTv2, Generator=Base Mistral-7B-Instruct2026.03 | 74.6 | 60.5 | |
| CL ColBERT / Base MistralRetriever=Fine-tuned ColBERTv2 (Contrastive Learning), Generator=Base Mistral-7B-Instruct2026.03 | 74.4 | 58 | |
| GPT-5.4model_family=GPT-5.42026.03 | 74.4 | 77.7 | |
| Base Mistral (w/o RAG)RAG=false, Generator=Mistral-7B-Instruct2026.03 | 73.3 | 58.1 | |
| CL ColBERT / DPO MistralRetriever=Fine-tuned ColBERTv2 (Contrastive Learning), Generator=Mistral-7B-Instruct aligned with DPO2026.03 | 70 | 55.9 |