# RepoForge: Training a SOTA Fast-thinking SWE Agent with an End-to-End Data Curation Pipeline Synergizing SFT and RL at Scale

arXiv: https://arxiv.org/abs/2508.01550 · HTML: https://mlsanity.com/paper/2508.01550/

## Key facts

- Cited by 8 papers in the map (data through 14 Oct 2026).
- Placed in Reasoning for agents, planning and code, one of 17 directions in LLM reasoning.
- Builds on 9 of its 17 references that are in the map.
- First posted on arXiv 3 Aug 2025.

## Placed in

- LLM reasoning / [Reasoning for agents, planning and code](https://mlsanity.com/lr/reasoning-for-agents-planning-and-code/)
- LLM agents and tool use / [Software engineering coding agents](https://mlsanity.com/agents/software-engineering-coding-agents/)

## Builds on (most cited first, 9 of 9)

- [HybridFlow: A Flexible and Efficient RLHF Framework](https://mlsanity.com/paper/2409.19256/) (2024)
- [Search-R1: Training LLMs to Reason and Leverage Search Engines with Reinforcement Learning](https://mlsanity.com/paper/2503.09516/) (2025)
- [Training Software Engineering Agents and Verifiers with SWE-Gym](https://mlsanity.com/paper/2412.21139/) (2024)
- [ToRL: Scaling Tool-Integrated RL](https://mlsanity.com/paper/2503.23383/) (2025)
- [R2E-Gym: Procedural Environments and Hybrid Verifiers for Scaling Open-Weights SWE Agents](https://mlsanity.com/paper/2504.07164/) (2025)
- [MMSearch-R1: Incentivizing LMMs to Search](https://mlsanity.com/paper/2506.20670/) (2025)
- [SWE-bench Goes Live!](https://mlsanity.com/paper/2505.23419/) (2025)
- [Skywork-SWE: Unveiling Data Scaling Laws for Software Engineering in LLMs](https://mlsanity.com/paper/2506.19290/) (2025)
- [DatasetAgent: A Novel Multi-Agent System for Auto-Constructing Datasets from Real-World Images](https://mlsanity.com/paper/2507.08648/) (2025)

## Built on by (most cited first, 8 of 8)

- [SWE-rebench V2: Language-Agnostic SWE Task Collection at Scale](https://mlsanity.com/paper/2602.23866/) (2026)
- [Agentic Software Issue Resolution with Large Language Models: A Survey](https://mlsanity.com/paper/2512.22256/) (2025)
- [ExecCritic: Learn to Test, Test to Improve for Coding Agents](https://mlsanity.com/paper/2609.09133/) (2026)
- [EnvX: Agentize Everything with Agentic AI](https://mlsanity.com/paper/2509.08088/) (2025)
- [PhoenixRepair: Rethinking Repair Strategy Exploration in Software Agents](https://mlsanity.com/paper/2607.18859/) (2026)
- [Beyond Correctness: Enhancing Architectural Reasoning in Code LLMs via Scalable Labeling with Agentic Judgment](https://mlsanity.com/paper/2606.14948/) (2026)
- [Teaching Agents to Code Reliably](https://mlsanity.com/paper/2610.03984/) (2026)
- [SWE-Stream: Scaling Executable Training Data for SWE Agents via Progressive Construction and Elastic Streaming](https://mlsanity.com/paper/2610.20193/) (2026)

## Abstract (arXiv)

Training software engineering (SWE) LLMs is bottlenecked by expensive infrastructure, inefficient evaluation pipelines, scarce training data, and costly quality control. We present RepoForge, an autonomous, end-to-end pipeline that generates, evaluates, and trains SWE agents at scale. Our key contributions include: (1) RepoForge-8B-Agent, achieving 17.4% on SWE-Bench-Verified~citepswebenchverified2024, establishing new state-of-the-art for ≤8B non-thinking LLMs; (2) 7,304 executable environments auto-generated from real GitHub commits with zero manual intervention; (3) 14× storage reduction (1.4GB rightarrow 102MB per instance) via intelligent dependency management and image pruning; (4) >70% faster evaluation using a Ray-powered~citepray2018 distributed RepoForge harness; (5) 19,000× cheaper labeling through our automated SPICE~citepspice2024 difficulty assessment technique. By unifying storage-efficient sandboxing, Ray-powered evaluation harness, automated data generation, SPICE-based labeling, and bubble-free RL scaffold, we demonstrate that even ≤8B models can reach new state-of-the-art performance on demanding benchmarks like SWE-Bench-Verified. Our approach addresses critical bottlenecks in SWE agent training: high storage costs of container-based evaluation, inefficient sequential reward pipelines, limited availability of high-quality training data, expensive manual labeling, and multi-turn RL pipeline bottlenecks.