bridge.perf_recipes.qwen.vr200.qwen3_moe#
VR200 performance recipes for Qwen3 MoE.
Module Contents#
Functions#
Qwen3 235B A22B pretrain: 256× VR200, BF16 (alias of GB200). |
|
Qwen3 235B A22B pretrain: 256× VR200, FP8-MX. |
|
Qwen3 235B A22B pretrain: 256× VR200, NVFP4 (alias of GB200). |
|
Qwen3 30B-A3B pretrain: 8× VR200, BF16 (alias of GB200). |
|
Qwen3 30B-A3B pretrain: 8× VR200, FP8-MX. |
API#
- bridge.perf_recipes.qwen.vr200.qwen3_moe.qwen3_235b_a22b_pretrain_256gpu_vr200_bf16_config() megatron.bridge.perf_recipes.qwen.common.ConfigContainer#
Qwen3 235B A22B pretrain: 256× VR200, BF16 (alias of GB200).
- bridge.perf_recipes.qwen.vr200.qwen3_moe.qwen3_235b_a22b_pretrain_256gpu_vr200_fp8mx_config() megatron.bridge.perf_recipes.qwen.common.ConfigContainer#
Qwen3 235B A22B pretrain: 256× VR200, FP8-MX.
- bridge.perf_recipes.qwen.vr200.qwen3_moe.qwen3_235b_a22b_pretrain_256gpu_vr200_nvfp4_config() megatron.bridge.perf_recipes.qwen.common.ConfigContainer#
Qwen3 235B A22B pretrain: 256× VR200, NVFP4 (alias of GB200).
- bridge.perf_recipes.qwen.vr200.qwen3_moe.qwen3_30b_a3b_pretrain_8gpu_vr200_bf16_config() megatron.bridge.perf_recipes.qwen.common.ConfigContainer#
Qwen3 30B-A3B pretrain: 8× VR200, BF16 (alias of GB200).
- bridge.perf_recipes.qwen.vr200.qwen3_moe.qwen3_30b_a3b_pretrain_8gpu_vr200_fp8mx_config() megatron.bridge.perf_recipes.qwen.common.ConfigContainer#
Qwen3 30B-A3B pretrain: 8× VR200, FP8-MX.