bridge.perf_recipes.llama.b200.llama31#

B200 performance recipes for Llama 3.1.

Module Contents#

Functions#

llama31_405b_pretrain_128gpu_b200_bf16_config

Llama3.1 405B pretrain: 128× B200, BF16, TP=4 PP=8 CP=2.

llama31_405b_pretrain_128gpu_b200_fp8cs_config

Llama3.1 405B pretrain: 128× B200, FP8 current-scaling, TP=4 PP=8 CP=2.

llama31_405b_pretrain_128gpu_b200_fp8mx_config

Llama3.1 405B pretrain: 128× B200, MXFP8, TP=4 PP=8 CP=2.

llama31_405b_pretrain_128gpu_b200_nvfp4_config

Llama3.1 405B pretrain: 128× B200, NVFP4, TP=4 PP=16.

llama31_405b_pretrain_256gpu_b200_nvfp4_config

Llama3.1 405B pretrain: 256x B200, NVFP4.

Data#

API#

bridge.perf_recipes.llama.b200.llama31.llama31_405b_pretrain_128gpu_b200_bf16_config() megatron.bridge.perf_recipes.llama.common.ConfigContainer#

Llama3.1 405B pretrain: 128× B200, BF16, TP=4 PP=8 CP=2.

bridge.perf_recipes.llama.b200.llama31.llama31_405b_pretrain_128gpu_b200_fp8cs_config() megatron.bridge.perf_recipes.llama.common.ConfigContainer#

Llama3.1 405B pretrain: 128× B200, FP8 current-scaling, TP=4 PP=8 CP=2.

bridge.perf_recipes.llama.b200.llama31.llama31_405b_pretrain_128gpu_b200_fp8mx_config() megatron.bridge.perf_recipes.llama.common.ConfigContainer#

Llama3.1 405B pretrain: 128× B200, MXFP8, TP=4 PP=8 CP=2.

bridge.perf_recipes.llama.b200.llama31.llama31_405b_pretrain_128gpu_b200_nvfp4_config() megatron.bridge.perf_recipes.llama.common.ConfigContainer#

Llama3.1 405B pretrain: 128× B200, NVFP4, TP=4 PP=16.

bridge.perf_recipes.llama.b200.llama31.llama31_405b_pretrain_256gpu_b200_bf16_config#

None

bridge.perf_recipes.llama.b200.llama31.llama31_405b_pretrain_256gpu_b200_fp8cs_config#

None

bridge.perf_recipes.llama.b200.llama31.llama31_405b_pretrain_256gpu_b200_fp8mx_config#

None

bridge.perf_recipes.llama.b200.llama31.llama31_405b_pretrain_256gpu_b200_nvfp4_config() megatron.bridge.perf_recipes.llama.common.ConfigContainer#

Llama3.1 405B pretrain: 256x B200, NVFP4.