foxhunt

Files

jgrusewski 88eb7aa241 feat(sp14-β): allocate rollout-sized aux buffers + AuxHeadsForwardOps in collector

Step 2 of β migration: 5 new buffers sized to alloc_episodes (vs
trainer's batch_size). AuxHeadsForwardOps instance is collector-
owned and stream-bound to collector stream via
AuxHeadsForwardOps::new(&stream). Param tensors shared with trainer
via existing f32_weight_ptrs_from_base path.

Buffer sizing: exp_aux_nb_hidden_buf [alloc_episodes × 32],
exp_aux_nb_logits/softmax_buf [alloc_episodes × 2],
exp_aux_nb_label_buf [alloc_episodes] i32, exp_aux_dir_acc_buf [6]
mapped-pinned (post-B1.1a 6-float layout matches trainer).

Compile clean; sp14_oracle_tests 2/2 non-GPU pass (7 GPU tests
ignored on RTX 3050 Ti host).

No aux forward yet — buffers allocated, ready for wire.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

2026-05-07 21:46:22 +02:00

infra

feat(infra): GPU training launcher with local/cloud routing

2026-02-23 11:19:58 +01:00

plans

feat(dqn): enable Branching DQN with 45 factored actions (5×3×3)

2026-03-11 22:00:13 +01:00

superpowers

plan(sp15): fix 3 critical compile-breakers from third review

2026-05-06 10:32:10 +02:00

dqn-backward-nan-audit.md

fix(dqn): SP3 Mech 10 — ISV-driven h_s2 activation clamp + slot 49 diag