Heng1999/Qwen3-8B-TIR-ASPO

This repository contains the official ASPO-trained Tool-Integrated Reasoning (TIR) model from the paper: Understanding Tool-Integrated Reasoning.

This model is a Qwen3-8B fine-tuned with our novel Advantage Shaping Policy Optimization (ASPO) algorithm. Unlike the baseline DAPO-trained TIR model, this version is specifically optimized to encourage a more dynamic and iterative tool-use style by promoting earlier code invocation.