`bridge.perf_recipes.llama.vr200.llama31`#

VR200 performance recipes for Llama 3.1.

Module Contents#

Functions#

`llama31_405b_pretrain_256gpu_vr200_bf16_config`	Llama3.1 405B pretrain: 256x VR200, BF16 (alias of GB300).
`llama31_405b_pretrain_256gpu_vr200_fp8mx_config`	Llama3.1 405B pretrain: 256x VR200, FP8-MX (alias of GB300).
`llama31_405b_pretrain_256gpu_vr200_nvfp4_config`	Llama3.1 405B pretrain: 256x VR200, NVFP4 (alias of GB300).

API#

bridge.perf_recipes.llama.vr200.llama31.llama31_405b_pretrain_256gpu_vr200_bf16_config() → megatron.bridge.perf_recipes.llama.common.ConfigContainer#: Llama3.1 405B pretrain: 256x VR200, BF16 (alias of GB300).

bridge.perf_recipes.llama.vr200.llama31.llama31_405b_pretrain_256gpu_vr200_fp8mx_config() → megatron.bridge.perf_recipes.llama.common.ConfigContainer#: Llama3.1 405B pretrain: 256x VR200, FP8-MX (alias of GB300).

bridge.perf_recipes.llama.vr200.llama31.llama31_405b_pretrain_256gpu_vr200_nvfp4_config() → megatron.bridge.perf_recipes.llama.common.ConfigContainer#: Llama3.1 405B pretrain: 256x VR200, NVFP4 (alias of GB300).