bridge.perf_recipes.qwen_vl.gb300.qwen35_vl

GB300 performance recipes for Qwen3.5-VL.

Module Contents

Functions

qwen35_vl_35b_a3b_pretrain_8gpu_gb300_bf16_config	Qwen3.5-VL 35B-A3B pretrain: 8× GB300, BF16, EP=8.
qwen35_vl_35b_a3b_pretrain_8gpu_gb300_fp8cs_config	Qwen3.5-VL 35B-A3B pretrain: 8× GB300, FP8 current-scaling.
qwen35_vl_35b_a3b_pretrain_8gpu_gb300_fp8mx_config	Qwen3.5-VL 35B-A3B pretrain: 8× GB300, MXFP8.
qwen35_vl_122b_a10b_pretrain_32gpu_gb300_bf16_config	Qwen3.5-VL 122B-A10B pretrain: 32× GB300, BF16, EP=32.
qwen35_vl_122b_a10b_pretrain_32gpu_gb300_fp8cs_config	Qwen3.5-VL 122B-A10B pretrain: 32× GB300, FP8 current-scaling.
qwen35_vl_122b_a10b_pretrain_32gpu_gb300_fp8mx_config	Qwen3.5-VL 122B-A10B pretrain: 32× GB300, MXFP8.
qwen35_vl_397b_a17b_pretrain_64gpu_gb300_bf16_config	Qwen3.5-VL 397B-A17B pretrain: 64× GB300, BF16, EP=64.
qwen35_vl_397b_a17b_pretrain_64gpu_gb300_fp8cs_config	Qwen3.5-VL 397B-A17B pretrain: 64× GB300, FP8 current-scaling.
qwen35_vl_397b_a17b_pretrain_64gpu_gb300_fp8mx_config	Qwen3.5-VL 397B-A17B pretrain: 64× GB300, MXFP8.

API

bridge.perf_recipes.qwen_vl.gb300.qwen35_vl.qwen35_vl_35b_a3b_pretrain_8gpu_gb300_bf16_config() → megatron.bridge.perf_recipes.qwen_vl.common.ConfigContainer

Qwen3.5-VL 35B-A3B pretrain: 8× GB300, BF16, EP=8.

bridge.perf_recipes.qwen_vl.gb300.qwen35_vl.qwen35_vl_35b_a3b_pretrain_8gpu_gb300_fp8cs_config() → megatron.bridge.perf_recipes.qwen_vl.common.ConfigContainer

Qwen3.5-VL 35B-A3B pretrain: 8× GB300, FP8 current-scaling.

bridge.perf_recipes.qwen_vl.gb300.qwen35_vl.qwen35_vl_35b_a3b_pretrain_8gpu_gb300_fp8mx_config() → megatron.bridge.perf_recipes.qwen_vl.common.ConfigContainer

Qwen3.5-VL 35B-A3B pretrain: 8× GB300, MXFP8.

bridge.perf_recipes.qwen_vl.gb300.qwen35_vl.qwen35_vl_122b_a10b_pretrain_32gpu_gb300_bf16_config() → megatron.bridge.perf_recipes.qwen_vl.common.ConfigContainer

Qwen3.5-VL 122B-A10B pretrain: 32× GB300, BF16, EP=32.

bridge.perf_recipes.qwen_vl.gb300.qwen35_vl.qwen35_vl_122b_a10b_pretrain_32gpu_gb300_fp8cs_config() → megatron.bridge.perf_recipes.qwen_vl.common.ConfigContainer

Qwen3.5-VL 122B-A10B pretrain: 32× GB300, FP8 current-scaling.

bridge.perf_recipes.qwen_vl.gb300.qwen35_vl.qwen35_vl_122b_a10b_pretrain_32gpu_gb300_fp8mx_config() → megatron.bridge.perf_recipes.qwen_vl.common.ConfigContainer

Qwen3.5-VL 122B-A10B pretrain: 32× GB300, MXFP8.

bridge.perf_recipes.qwen_vl.gb300.qwen35_vl.qwen35_vl_397b_a17b_pretrain_64gpu_gb300_bf16_config() → megatron.bridge.perf_recipes.qwen_vl.common.ConfigContainer

Qwen3.5-VL 397B-A17B pretrain: 64× GB300, BF16, EP=64.

bridge.perf_recipes.qwen_vl.gb300.qwen35_vl.qwen35_vl_397b_a17b_pretrain_64gpu_gb300_fp8cs_config() → megatron.bridge.perf_recipes.qwen_vl.common.ConfigContainer

Qwen3.5-VL 397B-A17B pretrain: 64× GB300, FP8 current-scaling.

bridge.perf_recipes.qwen_vl.gb300.qwen35_vl.qwen35_vl_397b_a17b_pretrain_64gpu_gb300_fp8mx_config() → megatron.bridge.perf_recipes.qwen_vl.common.ConfigContainer

Qwen3.5-VL 397B-A17B pretrain: 64× GB300, MXFP8.