bridge.perf_recipes.qwen_vl.gb300.qwen3_vl#
GB300 performance recipes for Qwen3-VL.
Module Contents#
Functions#
Qwen3-VL 235B-A22B pretrain: 64× GB300, BF16, EP=64. |
|
Qwen3-VL 235B-A22B pretrain: 64× GB300, FP8 current-scaling, EP=64. |
|
Qwen3-VL 235B-A22B pretrain: 64× GB300, MXFP8, EP=64. |
|
Qwen3-VL 30B-A3B pretrain: 8× GB300, BF16, EP=8. |
|
Qwen3-VL 30B-A3B pretrain: 8× GB300, FP8 current-scaling, EP=8. |
|
Qwen3-VL 30B-A3B pretrain: 8× GB300, MXFP8, EP=8. |
API#
- bridge.perf_recipes.qwen_vl.gb300.qwen3_vl.qwen3_vl_235b_a22b_pretrain_64gpu_gb300_bf16_config() megatron.bridge.perf_recipes.qwen_vl.common.ConfigContainer#
Qwen3-VL 235B-A22B pretrain: 64× GB300, BF16, EP=64.
- bridge.perf_recipes.qwen_vl.gb300.qwen3_vl.qwen3_vl_235b_a22b_pretrain_64gpu_gb300_fp8cs_config() megatron.bridge.perf_recipes.qwen_vl.common.ConfigContainer#
Qwen3-VL 235B-A22B pretrain: 64× GB300, FP8 current-scaling, EP=64.
- bridge.perf_recipes.qwen_vl.gb300.qwen3_vl.qwen3_vl_235b_a22b_pretrain_64gpu_gb300_fp8mx_config() megatron.bridge.perf_recipes.qwen_vl.common.ConfigContainer#
Qwen3-VL 235B-A22B pretrain: 64× GB300, MXFP8, EP=64.
- bridge.perf_recipes.qwen_vl.gb300.qwen3_vl.qwen3_vl_30b_a3b_pretrain_8gpu_gb300_bf16_config() megatron.bridge.perf_recipes.qwen_vl.common.ConfigContainer#
Qwen3-VL 30B-A3B pretrain: 8× GB300, BF16, EP=8.
- bridge.perf_recipes.qwen_vl.gb300.qwen3_vl.qwen3_vl_30b_a3b_pretrain_8gpu_gb300_fp8cs_config() megatron.bridge.perf_recipes.qwen_vl.common.ConfigContainer#
Qwen3-VL 30B-A3B pretrain: 8× GB300, FP8 current-scaling, EP=8.
- bridge.perf_recipes.qwen_vl.gb300.qwen3_vl.qwen3_vl_30b_a3b_pretrain_8gpu_gb300_fp8mx_config() megatron.bridge.perf_recipes.qwen_vl.common.ConfigContainer#
Qwen3-VL 30B-A3B pretrain: 8× GB300, MXFP8, EP=8.