bridge.perf_recipes.llama.r100.llama3

R100 performance recipes for Llama 3.

Module Contents

Functions

llama3_8b_pretrain_8gpu_r100_bf16_config	Llama3 8B pretrain: 8× R100, BF16.
llama3_8b_pretrain_8gpu_r100_fp8cs_config	Llama3 8B pretrain: 8× R100, FP8 current-scaling.
llama3_8b_pretrain_8gpu_r100_fp8mx_config	Llama3 8B pretrain: 8× R100, MXFP8.
llama3_8b_pretrain_8gpu_r100_nvfp4_config	Llama3 8B pretrain: 8× R100, NVFP4.

API

bridge.perf_recipes.llama.r100.llama3.llama3_8b_pretrain_8gpu_r100_bf16_config() → megatron.bridge.perf_recipes.llama.common.ConfigContainer

Llama3 8B pretrain: 8× R100, BF16.

bridge.perf_recipes.llama.r100.llama3.llama3_8b_pretrain_8gpu_r100_fp8cs_config() → megatron.bridge.perf_recipes.llama.common.ConfigContainer

Llama3 8B pretrain: 8× R100, FP8 current-scaling.

bridge.perf_recipes.llama.r100.llama3.llama3_8b_pretrain_8gpu_r100_fp8mx_config() → megatron.bridge.perf_recipes.llama.common.ConfigContainer

Llama3 8B pretrain: 8× R100, MXFP8.

bridge.perf_recipes.llama.r100.llama3.llama3_8b_pretrain_8gpu_r100_nvfp4_config() → megatron.bridge.perf_recipes.llama.common.ConfigContainer

Llama3 8B pretrain: 8× R100, NVFP4.