计划执行与检查节点
executing-plans
按既定实施计划完成工作，保持过程检查和反馈。
探索技能
效率与协作 · SKILL.md
cfm_training_objective
cfm_training_objective【效率与协作】Compute Conditional Flow Matching losses and bounded optimizer updates for a small parametric vector field.
独立 GitHub 快照 · 自动分类 · 以原作者文档为准
技能说明
GitHubname: cfm_training_objective description: Compute Conditional Flow Matching losses and bounded optimizer updates for a small parametric vector field.
CFM Training Objective
Use this skill to compute E||v_theta(t,x_t)-u_t||^2 and log real optimizer updates. Inputs are path batches and predictor parameters; outputs are scalar loss and a trace with loss_before, loss_after, params_before, and params_after. Run python tests/test_cfm_training.py. The included nonlinear-feature linear model is for bounded recovery, not full image training.