AI Training & Fine-tuning

2 summarised stories about AI Training & Fine-tuning, each linking back to the original source. Browse all topics →

Thursday, 16 July 2026

Function-Aware Fill-in-the-Middle as Mid-Training for Coding Agent Foundation Models

arXiv cs.CL 18 hours ago

Researchers developed function-aware fill-in-the-middle mid-training, a technique that masks functions in code based on program dependency analysis to help coding agents better integrate external tool outputs. Mid-training Qwen models on 2.6 billion tokens from 968 GitHub repositories improved performance on SWE-Bench-Verified by +2.8 to +3.2 points depending on model size, with larger gains of +3.7 to +5.4 on SWE-Bench-Lite. The approach preserves general coding ability while improving agent performance, as the function-call inductive bias transfers across different benchmarks and post-training methods.

AI Agents AI Benchmarking AI Code Generation AI Training & Fine-tuning

HELP: Human-Efficient Large-Scale Robot Post-Training with Rollout Segmentation

arXiv cs.AI 18 hours ago

Researchers developed HELP, a robot post-training pipeline where two human operators supervise twelve robots simultaneously using role specialization and an automatic rollout segmentation system to identify useful training data. The system achieved 80%-95% success rates across four manipulation tasks and improved task throughput by 1.7x to 4.2x compared to the base model. This approach enables more efficient human-robot collaboration by reducing operator workload and focusing training on meaningful robot behaviors and failures.

AI Efficiency AI Training & Fine-tuning Human-in-the-Loop Robotics