GraphForge: Evidence-Graph Workspace Synthesis Lifts GDPVal +65.7 With Only 2,169 Trajectories
ustc-community · hf · 2026-10-02
GraphForge synthesizes verifiable working-agent tasks grounded in real files via evidence graphs: tasks and rubrics are derived from file relations, each criterion anchored to files needed to verify it, with rollout-based executability checks and a revision agent. Fine-tuning Qwen3.6-27B on 2,169 trajectories brings GDPVal to 1445.7 (+65.7) under OpenHands, and Workspace-Bench-Lite and SpreadsheetBench II to +7.7 and +13.7 under Claude Code. Rejection fine-tuning using evidence-anchored rubrics adds further gains. Data and models are released.
More from coding & agent
- AI in finance shows a power law: top 1% users generate 35x more files with agents — peterjliu · 2026-11-19
- Linux LTS patch flood: 3318 patches for 6.6 alone, 'no way humans are reviewing this' — pkqzy888 · 2026-11-19
- Caliper Harness: open-source coding harness disciplines small local models with plans, checks, reviews — Saraozte01 · 2026-11-19
- AI systems lack executive function, not scale: four manager-layer strategies to fix it — Designer-Material633 · 2026-11-19
- Google employee replaces doomscrolling with 3 fully automated personal AI newsletters — DynamicWebPaige · 2026-11-19
- Designer spins up an agent to build state-airline brand gallery mid-flight delay — beh_zod · 2026-11-19