`bridge.perf_recipes.muse_glimmer.h100.muse_glimmer`#

H100 performance recipes for Muse Glimmer.

Module Contents#

Functions#

`_muse_glimmer_30b_pretrain_32gpu_h100_config`	Build the common Muse Glimmer dense-decoder H100 benchmark.
`muse_glimmer_30b_pretrain_32gpu_h100_bf16_config`	Muse Glimmer dense decoder pretrain: 32× H100, BF16, TP=4 PP=4 CP=2.
`muse_glimmer_30b_pretrain_32gpu_h100_fp8cs_config`	Muse Glimmer dense decoder pretrain: 32× H100, FP8 current scaling.

Data#

API#

bridge.perf_recipes.muse_glimmer.h100.muse_glimmer._muse_glimmer_30b_pretrain_32gpu_h100_config() → megatron.bridge.training.config.ConfigContainer#: Build the common Muse Glimmer dense-decoder H100 benchmark.

bridge.perf_recipes.muse_glimmer.h100.muse_glimmer.muse_glimmer_30b_pretrain_32gpu_h100_bf16_config() → megatron.bridge.training.config.ConfigContainer#: Muse Glimmer dense decoder pretrain: 32× H100, BF16, TP=4 PP=4 CP=2.

bridge.perf_recipes.muse_glimmer.h100.muse_glimmer.muse_glimmer_30b_pretrain_32gpu_h100_fp8cs_config() → megatron.bridge.training.config.ConfigContainer#: Muse Glimmer dense decoder pretrain: 32× H100, FP8 current scaling.

bridge.perf_recipes.muse_glimmer.h100.muse_glimmer.__all__#: [‘muse_glimmer_30b_pretrain_32gpu_h100_bf16_config’, ‘muse_glimmer_30b_pretrain_32gpu_h100_fp8cs_con…