This is a from-scratch reproduction and extension of the AgentDropout paper's experiments, run against real LLM APIs (OpenRouter) across three models, six benchmarks, and six methods. It's a full ...
{type:'call.start',from:'loop',to:'history',method:'append',input:'用户：统计 data.txt 行数'}, {type:'call.end',from:'history',to:'loop',method:'append',output ...
Some results have been hidden because they may be inaccessible to you
Show inaccessible results