lm_head Gradient Norm

How it works

Comparisons

Evidence quality

Highlighted claims

• Among evaluated signals, only the lm_head gradient norm produced a sharp spike synchronized with collapse onset. — When to Stop Reusing: Dynamic Gradient Gating for Sample-Efficient RLVR
• The lm_head gradient norm lower-bounds empirical Pearson chi-squared divergence at the batch level. — When to Stop Reusing: Dynamic Gradient Gating for Sample-Efficient RLVR
• The lm_head gradient receives the raw error signal directly, while intermediate layer gradients are filtered through a Jacobian. — When to Stop Reusing: Dynamic Gradient Gating for Sample-Efficient RLVR
• A surge in lm_head gradient norm is interpreted as a certified early-warning indicator rather than a heuristic proxy. — When to Stop Reusing: Dynamic Gradient Gating for Sample-Efficient RLVR