XingYing-stack/TIPS-Qwen3-4B-Instruct-2507-Math

TEXT GENERATIONPricing:Input $0.4 / Cached $0.08 / Output $0.8Concurrent Unit Cost:1Model Size:4BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Aug 27, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

The XingYing-stack/TIPS-Qwen3-4B-Instruct-2507-Math is a 4 billion parameter generative reward model, initialized from Qwen/Qwen3-4B-Instruct-2507. It is specifically trained with outcome-only GRPO to reason over mathematical solutions, producing step-level and outcome labels. This model is designed for reward modeling and process verification in mathematical contexts, rather than general-purpose chat applications.

Loading preview...

Model Overview

The TIPS-Qwen3-4B-Instruct-2507-Math is a specialized 4 billion parameter generative reward model. It is built upon the Qwen/Qwen3-4B-Instruct-2507 base model and has undergone further training using an outcome-only Generative Reward Policy Optimization (GRPO) approach.

Key Capabilities

  • Mathematical Reasoning: Designed to analyze and reason over mathematical solutions.
  • Generative Reward Modeling: Produces both step-level and final outcome labels for mathematical problems.
  • Process Verification: Intended for verifying the correctness of solution steps and overall outcomes in mathematical tasks.

Intended Use

This model is specifically purposed for:

  • Reward modeling in mathematical domains.
  • Verifying the process and results of mathematical solutions.

It is not recommended for general-purpose conversational AI or chat applications. Users should refer to the TIPS repository for appropriate prompt templates and evaluation scripts. Training data is available at XingYing-stack/TIPS-Training-Data.