pub fn compute_kd_loss(
student_out: &CausalLmOutput,
teacher_out: &CausalLmOutput,
l2l_loss_scale: f64,
) -> Result<KdLossBreakdown>Expand description
Compute combined Knowledge Distillation loss (KD + L2L MSE) from student and teacher outputs.
Slices &hidden_states[1..] to skip embedding outputs, matching upstream outputs.hidden_states[1:].
ยงErrors
Returns an error if hidden_states are missing or loss calculation fails.