`bridge.perf_recipes.qwen_vl.gb200.qwen35_vl`#

GB200 performance recipes for Qwen3.5-VL.

Module Contents#

Functions#

`qwen35_vl_35b_a3b_pretrain_8gpu_gb200_bf16_config`	Qwen3.5-VL 35B-A3B pretrain: 8× GB200, BF16, EP=8.
`qwen35_vl_35b_a3b_pretrain_8gpu_gb200_fp8cs_config`	Qwen3.5-VL 35B-A3B pretrain: 8× GB200, FP8 current-scaling.
`qwen35_vl_35b_a3b_pretrain_8gpu_gb200_fp8mx_config`	Qwen3.5-VL 35B-A3B pretrain: 8× GB200, MXFP8 (no attn CUDA graph).
`qwen35_vl_122b_a10b_pretrain_32gpu_gb200_bf16_config`	Qwen3.5-VL 122B-A10B pretrain: 32× GB200, BF16, PP=4 EP=8.
`qwen35_vl_122b_a10b_pretrain_32gpu_gb200_fp8cs_config`	Qwen3.5-VL 122B-A10B pretrain: 32× GB200, FP8 current-scaling.
`qwen35_vl_122b_a10b_pretrain_32gpu_gb200_fp8mx_config`	Qwen3.5-VL 122B-A10B pretrain: 32× GB200, MXFP8.
`qwen35_vl_397b_a17b_pretrain_64gpu_gb200_bf16_config`	Qwen3.5-VL 397B-A17B pretrain: 64× GB200, BF16, PP=8 EP=8.
`qwen35_vl_397b_a17b_pretrain_64gpu_gb200_fp8cs_config`	Qwen3.5-VL 397B-A17B pretrain: 64× GB200, FP8 current-scaling.
`qwen35_vl_397b_a17b_pretrain_64gpu_gb200_fp8mx_config`	Qwen3.5-VL 397B-A17B pretrain: 64× GB200, MXFP8.

API#

bridge.perf_recipes.qwen_vl.gb200.qwen35_vl.qwen35_vl_35b_a3b_pretrain_8gpu_gb200_bf16_config() → megatron.bridge.perf_recipes.qwen_vl.common.ConfigContainer#: Qwen3.5-VL 35B-A3B pretrain: 8× GB200, BF16, EP=8.

bridge.perf_recipes.qwen_vl.gb200.qwen35_vl.qwen35_vl_35b_a3b_pretrain_8gpu_gb200_fp8cs_config() → megatron.bridge.perf_recipes.qwen_vl.common.ConfigContainer#: Qwen3.5-VL 35B-A3B pretrain: 8× GB200, FP8 current-scaling.

bridge.perf_recipes.qwen_vl.gb200.qwen35_vl.qwen35_vl_35b_a3b_pretrain_8gpu_gb200_fp8mx_config() → megatron.bridge.perf_recipes.qwen_vl.common.ConfigContainer#: Qwen3.5-VL 35B-A3B pretrain: 8× GB200, MXFP8 (no attn CUDA graph).

bridge.perf_recipes.qwen_vl.gb200.qwen35_vl.qwen35_vl_122b_a10b_pretrain_32gpu_gb200_bf16_config() → megatron.bridge.perf_recipes.qwen_vl.common.ConfigContainer#: Qwen3.5-VL 122B-A10B pretrain: 32× GB200, BF16, PP=4 EP=8.

bridge.perf_recipes.qwen_vl.gb200.qwen35_vl.qwen35_vl_122b_a10b_pretrain_32gpu_gb200_fp8cs_config() → megatron.bridge.perf_recipes.qwen_vl.common.ConfigContainer#: Qwen3.5-VL 122B-A10B pretrain: 32× GB200, FP8 current-scaling.

bridge.perf_recipes.qwen_vl.gb200.qwen35_vl.qwen35_vl_122b_a10b_pretrain_32gpu_gb200_fp8mx_config() → megatron.bridge.perf_recipes.qwen_vl.common.ConfigContainer#: Qwen3.5-VL 122B-A10B pretrain: 32× GB200, MXFP8.

bridge.perf_recipes.qwen_vl.gb200.qwen35_vl.qwen35_vl_397b_a17b_pretrain_64gpu_gb200_bf16_config() → megatron.bridge.perf_recipes.qwen_vl.common.ConfigContainer#: Qwen3.5-VL 397B-A17B pretrain: 64× GB200, BF16, PP=8 EP=8.

bridge.perf_recipes.qwen_vl.gb200.qwen35_vl.qwen35_vl_397b_a17b_pretrain_64gpu_gb200_fp8cs_config() → megatron.bridge.perf_recipes.qwen_vl.common.ConfigContainer#: Qwen3.5-VL 397B-A17B pretrain: 64× GB200, FP8 current-scaling.

bridge.perf_recipes.qwen_vl.gb200.qwen35_vl.qwen35_vl_397b_a17b_pretrain_64gpu_gb200_fp8mx_config() → megatron.bridge.perf_recipes.qwen_vl.common.ConfigContainer#: Qwen3.5-VL 397B-A17B pretrain: 64× GB200, MXFP8.