From f1be4619e0cf9943764188eb2523c14809f8cb58 Mon Sep 17 00:00:00 2001 From: jgrusewski Date: Fri, 17 Apr 2026 09:23:51 +0200 Subject: [PATCH] =?UTF-8?q?fix:=20H100=20VRAM=20fraction=200.70=E2=86=920.?= =?UTF-8?q?55=20=E2=80=94=20OOM=20with=20state=5Fdim=3D96=20+=2086=20tenso?= =?UTF-8?q?rs?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit batch_size auto-scaled to 16384 consuming 78GB, leaving only 3GB for experience collector → OOM. Reducing replay buffer fraction gives more headroom. state_dim grew 88→96 (plan features), weight tensors grew 68→86 (ISV + plan head). Co-Authored-By: Claude Opus 4.6 (1M context) --- config/gpu/h100.toml | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/config/gpu/h100.toml b/config/gpu/h100.toml index 60b2210dd..caa17e1ac 100644 --- a/config/gpu/h100.toml +++ b/config/gpu/h100.toml @@ -4,7 +4,7 @@ batch_size = 8192 num_atoms = 51 buffer_size = 500000 hidden_dim_base = 256 -replay_buffer_vram_fraction = 0.70 +replay_buffer_vram_fraction = 0.55 # Reduced from 0.70: state_dim grew 88→96, 86 weight tensors [experience] gpu_timesteps_per_episode = 5000