AdarshSingh7647/Eklav-8B-Reranker-CotGen

TEXT GENERATIONPricing:Input $0.468 / Output $1.82Concurrent Unit Cost:1Model Size:8BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Aug 26, 2026Architecture:Transformer Featherless Exclusive Cold

AdarshSingh7647/Eklav-8B-Reranker-CotGen is an 8 billion parameter reranker model based on Qwen/Qwen3-8B, specifically designed for passage reranking tasks. It utilizes a standard full trace CoT SFT (Chain-of-Thought Supervised Fine-Tuning) baseline, trained with the CotGen method to learn reasoning. This model excels at determining passage relevance, achieving an average nDCG@10 of 31.5 on the BRIGHT dataset, and is optimized for use as a pointwise reranker.

Loading preview...

Overview

AdarshSingh7647/Eklav-8B-Reranker-CotGen is an 8 billion parameter model built upon the Qwen/Qwen3-8B base, specifically fine-tuned for passage reranking. It employs a standard full trace Chain-of-Thought (CoT) Supervised Fine-Tuning (SFT) baseline, referred to as CotGen, where the model learns to continue a partial reasoning trace rather than imitating an end-to-end one. This approach aims to improve reasoning capabilities by conditioning the model's own reasoning on a teacher's partial trace.

Key Capabilities

  • Passage Reranking: Designed to determine the relevance of a passage to a given query.
  • Pointwise Reranker: Functions similarly to models like Rank1, generating a reasoning trace that concludes with </think> true or </think> false.
  • Performance: Achieves an average nDCG@10 of 31.5 on the BRIGHT dataset, indicating its effectiveness in ranking relevant passages.

Usage and Recommendations

This model is intended for use as a reranker where relevance is scored from the logits of the final true or false token. It's crucial to use a setup like vLLM with a stop string (e.g., </think> true, </think> false) rather than a fixed token budget for generation. This prevents the model from generating excessively long reasoning traces or degenerating into repetition, ensuring reliable output for reranking tasks.