bridge.perf_recipes.deepseek.vr200.deepseek_v3#

VR200 performance recipes for DeepSeek V3.

Module Contents#

Functions#

deepseek_v3_pretrain_128gpu_vr200_bf16_config

DeepSeek V3 pretrain: 128× VR200, BF16.

deepseek_v3_pretrain_128gpu_vr200_fp8cs_config

DeepSeek V3 pretrain: 128× VR200, FP8 current-scaling.

deepseek_v3_pretrain_128gpu_vr200_fp8mx_config

DeepSeek V3 pretrain: 128× VR200, MXFP8.

deepseek_v3_pretrain_128gpu_vr200_nvfp4_config

DeepSeek V3 pretrain: 128× VR200, NVFP4.

deepseek_v3_pretrain_256gpu_vr200_bf16_config

DeepSeek V3 pretrain: 256× VR200, BF16 (alias of GB200).

deepseek_v3_pretrain_256gpu_vr200_fp8cs_config

DeepSeek V3 pretrain: 256× VR200, FP8-CS (alias of GB200).

deepseek_v3_pretrain_256gpu_vr200_fp8mx_config

DeepSeek V3 pretrain: 256× VR200, FP8-MX (alias of GB200).

deepseek_v3_pretrain_256gpu_vr200_nvfp4_config

DeepSeek V3 pretrain: 256× VR200, NVFP4 (alias of GB200).

API#

bridge.perf_recipes.deepseek.vr200.deepseek_v3.deepseek_v3_pretrain_128gpu_vr200_bf16_config() megatron.bridge.perf_recipes.deepseek.common.ConfigContainer#

DeepSeek V3 pretrain: 128× VR200, BF16.

bridge.perf_recipes.deepseek.vr200.deepseek_v3.deepseek_v3_pretrain_128gpu_vr200_fp8cs_config() megatron.bridge.perf_recipes.deepseek.common.ConfigContainer#

DeepSeek V3 pretrain: 128× VR200, FP8 current-scaling.

bridge.perf_recipes.deepseek.vr200.deepseek_v3.deepseek_v3_pretrain_128gpu_vr200_fp8mx_config() megatron.bridge.perf_recipes.deepseek.common.ConfigContainer#

DeepSeek V3 pretrain: 128× VR200, MXFP8.

bridge.perf_recipes.deepseek.vr200.deepseek_v3.deepseek_v3_pretrain_128gpu_vr200_nvfp4_config() megatron.bridge.perf_recipes.deepseek.common.ConfigContainer#

DeepSeek V3 pretrain: 128× VR200, NVFP4.

bridge.perf_recipes.deepseek.vr200.deepseek_v3.deepseek_v3_pretrain_256gpu_vr200_bf16_config() megatron.bridge.perf_recipes.deepseek.common.ConfigContainer#

DeepSeek V3 pretrain: 256× VR200, BF16 (alias of GB200).

bridge.perf_recipes.deepseek.vr200.deepseek_v3.deepseek_v3_pretrain_256gpu_vr200_fp8cs_config() megatron.bridge.perf_recipes.deepseek.common.ConfigContainer#

DeepSeek V3 pretrain: 256× VR200, FP8-CS (alias of GB200).

bridge.perf_recipes.deepseek.vr200.deepseek_v3.deepseek_v3_pretrain_256gpu_vr200_fp8mx_config() megatron.bridge.perf_recipes.deepseek.common.ConfigContainer#

DeepSeek V3 pretrain: 256× VR200, FP8-MX (alias of GB200).

bridge.perf_recipes.deepseek.vr200.deepseek_v3.deepseek_v3_pretrain_256gpu_vr200_nvfp4_config() megatron.bridge.perf_recipes.deepseek.common.ConfigContainer#

DeepSeek V3 pretrain: 256× VR200, NVFP4 (alias of GB200).