sxiong/DeepControl-Qwen2.5-7B

TEXT GENERATIONPricing:Input $0.4 / Cached $0.08 / Output $0.8Concurrent Unit Cost:1Model Size:7.6BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Aug 16, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

DeepControl-Qwen2.5-7B is a 7.6 billion parameter deep search agent checkpoint based on the Qwen2.5-7B-Instruct architecture, developed by Siheng Xiong. This model is specifically fine-tuned for search-augmented LLM reasoning, as detailed in the paper "Adaptive Information Control for Search-Augmented LLM Reasoning." It is designed to operate within a DeepControl system, leveraging external retrieval from corpora like Wikipedia 2018 for enhanced reasoning tasks.

Loading preview...

DeepControl-Qwen2.5-7B: A Search-Augmented Reasoning Agent

DeepControl-Qwen2.5-7B is a specialized 7.6 billion parameter language model derived from Qwen2.5-7B-Instruct. Developed by Siheng Xiong and his colleagues, this model functions as a deep search agent, specifically designed to enhance LLM reasoning through adaptive information control and search augmentation. Its development is detailed in the research paper "Adaptive Information Control for Search-Augmented LLM Reasoning" (EMNLP 2026).

Key Capabilities and Features

  • Search-Augmented Reasoning: Optimized to work as part of a larger system that integrates external search and retrieval mechanisms.
  • DeepControl System Integration: Requires the full DeepControl system, including a specific system prompt, action parser, and a retrieval corpus (e.g., Wikipedia 2018), to function as intended.
  • Optimized for Retrieval Tasks: Configured for specific inference settings such as BF16 dtype, a context length of 8,192 tokens, and controlled generation parameters (temperature 0, max 512 new tokens per action, max 8 agent turns).

When to Use This Model

This model is ideal for researchers and developers working on:

  • Advanced Search Agents: Building LLM-powered agents that require robust external information retrieval and adaptive control over search queries.
  • Research in Search-Augmented LLMs: Experimenting with the principles outlined in the associated research paper for improving reasoning with external knowledge.
  • Complex Question Answering: Deploying systems that need to synthesize information from large external corpora to answer intricate questions.