From 88307b8e3bd7edf5e8bc7cafeeb3fb7f838a5c05 Mon Sep 17 00:00:00 2001 From: jgrusewski Date: Fri, 17 Apr 2026 11:58:08 +0200 Subject: [PATCH] =?UTF-8?q?test:=20remove=20plan=5Fnoise=5Finject=20from?= =?UTF-8?q?=20graph=5Fmega=20=E2=80=94=20isolating=20hang=20cause?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit plan_noise_inject is the ONLY new kernel inside graph_mega vs the working 11b1a1ca9. Everything else is identical (submit_forward_ops_main, submit_aux_ops, capture_graph_mega). If this fixes the hang, plan_noise_inject is incompatible with graph_mega on sm_90. Co-Authored-By: Claude Opus 4.6 (1M context) --- crates/ml/src/cuda_pipeline/gpu_dqn_trainer.rs | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/crates/ml/src/cuda_pipeline/gpu_dqn_trainer.rs b/crates/ml/src/cuda_pipeline/gpu_dqn_trainer.rs index 1a164e286..7f5c7429f 100644 --- a/crates/ml/src/cuda_pipeline/gpu_dqn_trainer.rs +++ b/crates/ml/src/cuda_pipeline/gpu_dqn_trainer.rs @@ -7884,7 +7884,7 @@ impl GpuDqnTrainer { self.launch_isv_feature_gate(batch_size)?; self.launch_recursive_confidence_forward(batch_size)?; self.launch_trade_plan_forward(batch_size)?; - self.launch_plan_noise_inject(batch_size)?; + // plan_noise_inject runs outside graph (pre-replay) — see fused_training.rs } // ── 2+3. Loss + gradient (blended MSE + C51 via c51_alpha ramp) ─