Description

A budget-aware replication of Sample-Efficient Learning from Agent Experience, exploring whether coding agents can internalize improvements from prior attempts into LoRA adapters. The project uses a masked training construction to capture in-context learning gains without the need for long prompt contexts at evaluation time, achieving a 25% pass rate on SWE-smith tasks.

Descriptions, tags, and model credits may be AI-generated or inferred from public sources and can be incomplete or wrong. Learn more.