Files
foxhunt/crates
jgrusewski a8cb75894c feat: add 8 IQL CUDA kernels — TD modulation, adv variance, per-sample support/epsilon, branch advantage, expectile gap
Appends to iql_value_kernel.cu: iql_modulate_td_errors, iql_adv_variance_reduce,
iql_compute_per_sample_support, iql_per_branch_advantage, iql_expectile_gap,
iql_gap_mean_reduce, iql_compute_per_sample_epsilon. All f32, zero atomics, fully
deterministic. Builds clean.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
2026-04-13 14:59:37 +02:00
..