executing-plans
executing-plans
Use when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available
スキルを探す
生産性・共同作業 · SKILL.md
cfm_training_objective
cfm_training_objectiveCompute Conditional Flow Matching losses and bounded optimizer updates for a small parametric vector field.
Independent GitHub snapshot · Automated classification · Original documentation governs
スキルの説明
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.