SWE-Lego: Pushing the Limits of Supervised Fine-tuning for Software Issue Resolving

• URL: http://arxiv.org/abs/2601.01426v2
• Date: Wed, 07 Jan 2026 08:30:15 GMT
• Title: SWE-Lego: Pushing the Limits of Supervised Fine-tuning for Software Issue Resolving
• Abstract summary: SWE-Lego is a supervised fine-tuning (SFT) recipe designed to achieve state-ofthe-art performance in software engineering (SWE) issue resolving.<n>SWE-Lego comprises three core building blocks, with key findings summarized as follows.
• Score: 44.2496723329337
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: We present SWE-Lego, a supervised fine-tuning (SFT) recipe designed to achieve state-ofthe-art performance in software engineering (SWE) issue resolving. In contrast to prevalent methods that rely on complex training paradigms (e.g., mid-training, SFT, reinforcement learning, and their combinations), we explore how to push the limits of a lightweight SFT-only approach for SWE tasks. SWE-Lego comprises three core building blocks, with key findings summarized as follows: 1) the SWE-Lego dataset, a collection of 32k highquality task instances and 18k validated trajectories, combining real and synthetic data to complement each other in both quality and quantity; 2) a refined SFT procedure with error masking and a difficulty-based curriculum, which demonstrably improves action quality and overall performance. Empirical results show that with these two building bricks alone,the SFT can push SWE-Lego models to state-of-the-art performance among open-source models of comparable size on SWE-bench Verified: SWE-Lego-Qwen3-8B reaches 42.2%, and SWE-Lego-Qwen3-32B attains 52.6%. 3) We further evaluate and improve test-time scaling (TTS) built upon the SFT foundation. Based on a well-trained verifier, SWE-Lego models can be significantly boosted--for example, 42.2% to 49.6% and 52.6% to 58.8% under TTS@16 for the 8B and 32B models, respectively.

Related papers

• SLAI T-Rex: Full-Parameter Post-training of the DeepSeek-V4 Family on Ascend SuperPOD [91.71050339479261]
  This report presents an end-to-end optimization practice on the Ascend NPU SuperPOD.<n>We develop a hierarchical optimization framework spanning model-level parallelism, computation-communication orchestration, and low-level kernel execution.<n>The resulting system achieves 34.22% Model FLOPs Utilization (MFU) with a 2.93x improvement over the open-source baseline recipe.
  arXiv  Detail & Related papers  (2026-07-22T13:49:17Z)
• From SWE-ZERO to SWE-HERO: Execution-free to Execution-based Fine-tuning for Software Engineering Agents [42.7127695634038]
  SWE-ZERO is a two-stage SFT recipe that achieves state-of-the-art results on SWE-bench.<n>We release a dataset of 300k SWE-ZERO and 13k SWE-HERO trajectories distilled from Qwen3-Coder-480B.<n>Despite being trained exclusively on Python, our agents demonstrate robust zero-shot transferability on SWE-bench.
  arXiv  Detail & Related papers  (2026-04-02T00:11:17Z)
• Yet Even Less Is Even Better For Agentic, Reasoning, and Coding LLMs [28.672417618778976]
  Training effective software engineering agents requires large volumes of task-specific trajectories.<n>We propose an end-to-end training framework that achieves superior agentic capabilities with fewer but higher-quality training trajectories.
  arXiv  Detail & Related papers  (2026-04-01T12:33:25Z)
• SWE-Fuse: Empowering Software Agents via Issue-free Trajectory Learning and Entropy-aware RLVR Training [20.30270129690516]
  We propose textbftextitSWE-Fuse, an issue-description-aware training framework that fuses issue-description-guided and issue-free samples for training SWE agents.<n>SWE-Fuse outperforms the best 8B and 32B baselines by 43.0% and 60.2% in solve rate, respectively.
  arXiv  Detail & Related papers  (2026-03-09T03:47:10Z)
• EGSS: Entropy-guided Stepwise Scaling for Reliable Software Engineering [14.718324012970944]
  Agentic Test-Time Scaling (TTS) has delivered state-of-the-art (SOTA) performance on complex software engineering tasks such as code generation and bug fixing.<n>We propose Entropy-Guided Stepwise Scaling (EGSS), a novel TTS framework that balances efficiency and effectiveness through entropy-guided adaptive search and robust test-suite augmentation.
  arXiv  Detail & Related papers  (2026-02-05T03:02:54Z)
• SWE-Master: Unleashing the Potential of Software Engineering Agents via Post-Training [78.37721886775215]
  We present SWE-Master, an open-source framework for building effective software engineering agents.<n>SWE-Master systematically explores the complete agent development pipeline, including teacher-trajectory synthesis and data curation.<n>We evaluate SWE-Master on SWE-bench Verified, a standard benchmark for realistic software engineering tasks.
  arXiv  Detail & Related papers  (2026-02-03T11:38:48Z)
• SWE-RM: Execution-free Feedback For Software Engineering Agents [61.86380395896069]
  Execution-based feedback is widely used in the development of coding agents through test-time scaling (TTS) and reinforcement learning (RL)<n>In contrast, execution-free feedback from reward models can provide more fine-grained signals without depending on unit test cases.<n>We introduce SWE-RM, an accurate and robust reward model adopting a mixture-of-experts architecture with 30B total parameters and 3B activated during inference.
  arXiv  Detail & Related papers  (2025-12-26T08:26:18Z)
• CoT-Saliency: Unified Chain-of-Thought Reasoning for Heterogeneous Saliency Tasks [96.64597365827046]
  We present the first unified framework that jointly handles three operationally heterogeneous saliency tasks.<n>We introduce a Chain-of-Thought (CoT) reasoning process in a Vision-Language Model (VLM) to bridge task heterogeneity.<n>We show our model matches or outperforms specialized SOTA methods and strong closed-source VLMs across all tasks.
  arXiv  Detail & Related papers  (2025-11-01T04:37:01Z)
• Massive Supervised Fine-tuning Experiments Reveal How Data, Layer, and Training Factors Shape LLM Alignment Quality [22.105116314320696]
  Supervised fine-tuning (SFT) is a critical step in aligning large language models with human instructions and values.<n>We trained a wide range of base models on a variety of datasets including code generation, mathematical reasoning, and general-domain tasks.<n>We then identified the dataset properties that matter most and examined the layer-wise modifications introduced by SFT.
  arXiv  Detail & Related papers  (2025-06-17T16:13:15Z)
• SWE-Dev: Building Software Engineering Agents with Training and Inference Scaling [39.53265893083118]
  Large language models (LLMs) have advanced rapidly from conversational problem solving to addressing real-world tasks involving tool use.<n>To address this issue, we present SWE-Dev, an SWE agent built upon open-source LLMs.<n> Experiments on the SWE-bench-Verified benchmark show that the SWE-Dev models can achieve top performance among all open SWE agents.
  arXiv  Detail & Related papers  (2025-06-09T11:03:16Z)
• R2E-Gym: Procedural Environments and Hybrid Verifiers for Scaling Open-Weights SWE Agents [32.06393076572057]
  AgentGym is the largest procedurally-curated executable gym environment for training real-world SWE-agents.<n>It is powered by two main contributions: SYNGEN, a synthetic data curation recipe, and Hybrid Test-time Scaling.<n>Overall, our approach achieves 51% on the SWE-Bench Verified benchmark, reflecting a new state-of-the-art for open-weight SWE-agents.
  arXiv  Detail & Related papers  (2025-04-09T17:55:19Z)
• Training Software Engineering Agents and Verifiers with SWE-Gym [89.55822534364727]
  SWE-Gym is the first environment for training real-world software engineering (SWE) agents.<n>SWE-Gym contains 2,438 real-world Python task instances.<n>We use SWE-Gym to train language model based SWE agents, achieving up to 19% absolute gains in resolve rate.
  arXiv  Detail & Related papers  (2024-12-30T18:15:39Z)
• Deep Facial Synthesis: A New Challenge [75.99659340231078]
  We first introduce a high-quality dataset for FSS, named FS2K, which consists of 2,104 image-sketch pairs. Second, we present the largest-scale FSS study by investigating 139 classical methods. Third, we present a simple baseline for FSS, named FSGAN.
  arXiv  Detail & Related papers  (2021-12-31T13:19:21Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.