AceSearcher/AceSearcher-1.5B

TEXT GENERATIONPricing:Input $0.04 / Cached $0.008 / Output $0.08Concurrent Unit Cost:1Model Size:1.5BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Jun 16, 2025License:mitArchitecture:Transformer Open Weights Featherless Exclusive Cold

AceSearcher/AceSearcher-1.5B is a 1.5 billion parameter language model built upon the Qwen-2.5-Instruct-1.5B backbone, featuring a 32768 token context length. Developed by AceSearcher, this model is specifically designed for bootstrapping reasoning and search capabilities in LLMs through reinforced self-play. It excels at complex question decomposition for both QA and fact verification tasks, as well as generating final answers grounded in provided contexts.

Loading preview...

AceSearcher-1.5B Overview

AceSearcher-1.5B is a 1.5 billion parameter model, leveraging the Qwen-2.5-Instruct-1.5B architecture, specifically developed for enhancing reasoning and search functionalities in Large Language Models. Its core innovation lies in its reinforced self-play training methodology, as detailed in the paper "AceSearcher: Bootstrapping Reasoning and Search for LLMs via Reinforced Self-Play".

Key Capabilities

  • Question Decomposition: Proficiently breaks down complex questions into multiple specific sub-questions for both general Question Answering (QA) and Fact Verification tasks. It supports referencing answers from earlier sub-questions.
  • Contextual Question Answering: Answers sub-questions and generates final answers based on provided context passages, with the ability to use its own knowledge if context is insufficient.
  • Fact Verification: Verifies claims for sub-claims and provides a final 'Yes' or 'No' verdict for original claims, grounded in context.
  • Financial Reasoning: Demonstrates specialized decomposition and Python program generation for document-level financial reasoning tasks, utilizing both passages and tabular data.

Good For

  • Developers building systems that require robust question decomposition and multi-step reasoning.
  • Applications needing accurate fact verification against provided textual evidence.
  • Financial analysis tools that process structured and unstructured data for complex queries.
  • Research into advanced reasoning and search mechanisms for LLMs.