`bridge.perf_recipes.qwen_vl.h100.qwen35_vl`#

H100 performance recipes for Qwen3.5-VL.

Module Contents#

Functions#

`qwen35_vl_35b_a3b_pretrain_16gpu_h100_bf16_config`	Qwen3.5-VL 35B-A3B pretrain: 16× H100, BF16, PP=2 VP=12 EP=8.
`qwen35_vl_35b_a3b_pretrain_16gpu_h100_fp8cs_config`	Qwen3.5-VL 35B-A3B pretrain: 16× H100, FP8 current-scaling, PP=2 VP=12.
`qwen35_vl_122b_a10b_pretrain_128gpu_h100_bf16_config`	Qwen3.5-VL 122B-A10B pretrain: 128× H100, BF16, TP=2 PP=8 VP=4 EP=16.
`qwen35_vl_122b_a10b_pretrain_128gpu_h100_fp8cs_config`	Qwen3.5-VL 122B-A10B pretrain: 128× H100, FP8 current-scaling.
`qwen35_vl_397b_a17b_pretrain_256gpu_h100_bf16_config`	Qwen3.5-VL 397B-A17B pretrain: 256× H100, BF16, TP=2 PP=8 VP=4 EP=32.
`qwen35_vl_397b_a17b_pretrain_256gpu_h100_fp8cs_config`	Qwen3.5-VL 397B-A17B pretrain: 256× H100, FP8 current-scaling.

API#

bridge.perf_recipes.qwen_vl.h100.qwen35_vl.qwen35_vl_35b_a3b_pretrain_16gpu_h100_bf16_config() → megatron.bridge.perf_recipes.qwen_vl.common.ConfigContainer#: Qwen3.5-VL 35B-A3B pretrain: 16× H100, BF16, PP=2 VP=12 EP=8.

bridge.perf_recipes.qwen_vl.h100.qwen35_vl.qwen35_vl_35b_a3b_pretrain_16gpu_h100_fp8cs_config() → megatron.bridge.perf_recipes.qwen_vl.common.ConfigContainer#: Qwen3.5-VL 35B-A3B pretrain: 16× H100, FP8 current-scaling, PP=2 VP=12.

bridge.perf_recipes.qwen_vl.h100.qwen35_vl.qwen35_vl_122b_a10b_pretrain_128gpu_h100_bf16_config() → megatron.bridge.perf_recipes.qwen_vl.common.ConfigContainer#: Qwen3.5-VL 122B-A10B pretrain: 128× H100, BF16, TP=2 PP=8 VP=4 EP=16.

bridge.perf_recipes.qwen_vl.h100.qwen35_vl.qwen35_vl_122b_a10b_pretrain_128gpu_h100_fp8cs_config() → megatron.bridge.perf_recipes.qwen_vl.common.ConfigContainer#: Qwen3.5-VL 122B-A10B pretrain: 128× H100, FP8 current-scaling.

bridge.perf_recipes.qwen_vl.h100.qwen35_vl.qwen35_vl_397b_a17b_pretrain_256gpu_h100_bf16_config() → megatron.bridge.perf_recipes.qwen_vl.common.ConfigContainer#: Qwen3.5-VL 397B-A17B pretrain: 256× H100, BF16, TP=2 PP=8 VP=4 EP=32.

bridge.perf_recipes.qwen_vl.h100.qwen35_vl.qwen35_vl_397b_a17b_pretrain_256gpu_h100_fp8cs_config() → megatron.bridge.perf_recipes.qwen_vl.common.ConfigContainer#: Qwen3.5-VL 397B-A17B pretrain: 256× H100, FP8 current-scaling.