AdarshSingh7647/Eklav-0.6B-Reranker-CotGen
Eklav-0.6B-Reranker-CotGen by AdarshSingh7647 is an 0.8 billion parameter model based on Qwen3-0.6B, specifically trained for passage reranking tasks using a standard full trace CoT SFT (CotGen) method. This model generates a reasoning trace before outputting a true/false relevance judgment, with relevance scored from final token logits. It is designed for pointwise reranking, achieving an average nDCG@10 of 16.1 on the BRIGHT dataset.
Loading preview...
Eklav-0.6B-Reranker-CotGen: A CoT-based Reranker
Eklav-0.6B-Reranker-CotGen is a 0.8 billion parameter model developed by AdarshSingh7647, built upon the Qwen3-0.6B base architecture. This model is specifically designed for passage reranking, utilizing a standard full trace CoT (Chain-of-Thought) Supervised Fine-Tuning (SFT) method, referred to as CotGen, as its training baseline.
Key Capabilities and Features
- Passage Reranking: Optimized for determining the relevance of a passage to a given query.
- Reasoning Trace Generation: The model generates an internal reasoning trace before producing a final relevance judgment.
- Pointwise Reranking: Operates as a pointwise reranker, similar to Rank1, where relevance is determined by the logits of the final
trueorfalsetoken, rather than parsing generated text. - Performance: Achieves an average nDCG@10 of 16.1 on the BRIGHT dataset.
- Robust Generation: Recommended for use with vLLM and stop strings (e.g.,
</think> true,</think> false) to prevent degeneration into repetition, as the model can generate long reasoning traces.
Use Cases
This model is ideal for applications requiring efficient and accurate document or passage retrieval systems, where reranking search results based on relevance to a query is critical. Its CoT-based approach allows for a more structured decision-making process in relevance scoring.