Files
foxhunt/config/gpu/h100.toml
jgrusewski c74a687ea8 feat: position-gated episodes + 5000-bar limit — close 45x training/val gap
Episode done flag: timer-based -> position-gated (trade complete = done).
V(flat)=0 is correct terminal anchor. Soft reset keeps equity on
trade completion; hard reset only on data-end or capital breach.
H100: 100 bars -> 5000 bars, gpu_n_episodes -> 1024.
ExperienceProfile gains optional gpu_n_episodes field.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
2026-04-16 21:56:44 +02:00

15 lines
279 B
TOML

# H100 PCIe/SXM (80GB VRAM) -- full production
[training]
batch_size = 8192
num_atoms = 51
buffer_size = 500000
hidden_dim_base = 256
replay_buffer_vram_fraction = 0.70
[experience]
gpu_timesteps_per_episode = 5000
gpu_n_episodes = 1024
[cuda]
cuda_stack_bytes = 65536 # 64KB