`bridge.perf_recipes.nemotronh.h100.nemotronh`#

H100 performance recipes for NemotronH and Nemotron 3.

Module Contents#

Functions#

`nemotronh_56b_pretrain_64gpu_h100_fp8cs_config`	NemotronH 56B pretrain: 64× H100, FP8 current-scaling.
`nemotron_3_nano_pretrain_16gpu_h100_bf16_config`	Nemotron 3 Nano pretrain: 16× H100, BF16, recompute MoE+layernorm.
`nemotron_3_nano_pretrain_16gpu_h100_fp8cs_config`	Nemotron 3 Nano pretrain: 16× H100, FP8 current-scaling, recompute.

API#

bridge.perf_recipes.nemotronh.h100.nemotronh.nemotronh_56b_pretrain_64gpu_h100_fp8cs_config() → megatron.bridge.perf_recipes.nemotronh.common.ConfigContainer#: NemotronH 56B pretrain: 64× H100, FP8 current-scaling.

bridge.perf_recipes.nemotronh.h100.nemotronh.nemotron_3_nano_pretrain_16gpu_h100_bf16_config() → megatron.bridge.perf_recipes.nemotronh.common.ConfigContainer#: Nemotron 3 Nano pretrain: 16× H100, BF16, recompute MoE+layernorm.

bridge.perf_recipes.nemotronh.h100.nemotronh.nemotron_3_nano_pretrain_16gpu_h100_fp8cs_config() → megatron.bridge.perf_recipes.nemotronh.common.ConfigContainer#: Nemotron 3 Nano pretrain: 16× H100, FP8 current-scaling, recompute.