`bridge.perf_recipes.qwen_vl.b200.qwen35_vl`#

B200 performance recipes for Qwen3.5-VL.

Module Contents#

Functions#

`qwen35_vl_35b_a3b_pretrain_8gpu_b200_bf16_config`	Qwen3.5-VL 35B-A3B pretrain: 8× B200, BF16, EP=8.
`qwen35_vl_35b_a3b_pretrain_8gpu_b200_fp8cs_config`	Qwen3.5-VL 35B-A3B pretrain: 8× B200, FP8 current-scaling.
`qwen35_vl_35b_a3b_pretrain_8gpu_b200_fp8mx_config`	Qwen3.5-VL 35B-A3B pretrain: 8× B200, MXFP8.
`qwen35_vl_122b_a10b_pretrain_32gpu_b200_bf16_config`	Qwen3.5-VL 122B-A10B pretrain: 32× B200, BF16, PP=4 VP=4 EP=8.
`qwen35_vl_122b_a10b_pretrain_32gpu_b200_fp8cs_config`	Qwen3.5-VL 122B-A10B pretrain: 32× B200, FP8 current-scaling.
`qwen35_vl_122b_a10b_pretrain_32gpu_b200_fp8mx_config`	Qwen3.5-VL 122B-A10B pretrain: 32× B200, MXFP8.
`qwen35_vl_397b_a17b_pretrain_64gpu_b200_bf16_config`	Qwen3.5-VL 397B-A17B pretrain: 64× B200, BF16, PP=8 VP=4 EP=8.
`qwen35_vl_397b_a17b_pretrain_64gpu_b200_fp8cs_config`	Qwen3.5-VL 397B-A17B pretrain: 64× B200, FP8 current-scaling.
`qwen35_vl_397b_a17b_pretrain_64gpu_b200_fp8mx_config`	Qwen3.5-VL 397B-A17B pretrain: 64× B200, MXFP8.

API#

bridge.perf_recipes.qwen_vl.b200.qwen35_vl.qwen35_vl_35b_a3b_pretrain_8gpu_b200_bf16_config() → megatron.bridge.perf_recipes.qwen_vl.common.ConfigContainer#: Qwen3.5-VL 35B-A3B pretrain: 8× B200, BF16, EP=8.

bridge.perf_recipes.qwen_vl.b200.qwen35_vl.qwen35_vl_35b_a3b_pretrain_8gpu_b200_fp8cs_config() → megatron.bridge.perf_recipes.qwen_vl.common.ConfigContainer#: Qwen3.5-VL 35B-A3B pretrain: 8× B200, FP8 current-scaling.

bridge.perf_recipes.qwen_vl.b200.qwen35_vl.qwen35_vl_35b_a3b_pretrain_8gpu_b200_fp8mx_config() → megatron.bridge.perf_recipes.qwen_vl.common.ConfigContainer#: Qwen3.5-VL 35B-A3B pretrain: 8× B200, MXFP8.

bridge.perf_recipes.qwen_vl.b200.qwen35_vl.qwen35_vl_122b_a10b_pretrain_32gpu_b200_bf16_config() → megatron.bridge.perf_recipes.qwen_vl.common.ConfigContainer#: Qwen3.5-VL 122B-A10B pretrain: 32× B200, BF16, PP=4 VP=4 EP=8.

bridge.perf_recipes.qwen_vl.b200.qwen35_vl.qwen35_vl_122b_a10b_pretrain_32gpu_b200_fp8cs_config() → megatron.bridge.perf_recipes.qwen_vl.common.ConfigContainer#: Qwen3.5-VL 122B-A10B pretrain: 32× B200, FP8 current-scaling.

bridge.perf_recipes.qwen_vl.b200.qwen35_vl.qwen35_vl_122b_a10b_pretrain_32gpu_b200_fp8mx_config() → megatron.bridge.perf_recipes.qwen_vl.common.ConfigContainer#: Qwen3.5-VL 122B-A10B pretrain: 32× B200, MXFP8.

bridge.perf_recipes.qwen_vl.b200.qwen35_vl.qwen35_vl_397b_a17b_pretrain_64gpu_b200_bf16_config() → megatron.bridge.perf_recipes.qwen_vl.common.ConfigContainer#: Qwen3.5-VL 397B-A17B pretrain: 64× B200, BF16, PP=8 VP=4 EP=8.

bridge.perf_recipes.qwen_vl.b200.qwen35_vl.qwen35_vl_397b_a17b_pretrain_64gpu_b200_fp8cs_config() → megatron.bridge.perf_recipes.qwen_vl.common.ConfigContainer#: Qwen3.5-VL 397B-A17B pretrain: 64× B200, FP8 current-scaling.

bridge.perf_recipes.qwen_vl.b200.qwen35_vl.qwen35_vl_397b_a17b_pretrain_64gpu_b200_fp8mx_config() → megatron.bridge.perf_recipes.qwen_vl.common.ConfigContainer#: Qwen3.5-VL 397B-A17B pretrain: 64× B200, MXFP8.