`bridge.perf_recipes.qwen_vl.common`#

Common helpers for qwen_vl performance recipes.

Module Contents#

`_qwen35_vl_common`	Apply VLM-specific performance benchmark settings for Qwen3.5-VL.
`_qwen35_vl_post`	VLM post-overrides that must run after `_benchmark_common`.
`_qwen35_vl_post_with_overlap`	Apply Qwen3.5-VL post-overrides and optimizer-step param-gather overlap.
`_qwen35_vl_post_clear_scope_with_overlap`	Apply Qwen3.5-VL post-overrides, clear graph scope, and enable overlap.
`_finalize_qwen3_vl`	Apply Qwen3-VL perf defaults that must override generic benchmark defaults.
`_finalize_qwen3_vl_with_overlap`	Apply Qwen3-VL perf defaults with optimizer-step param-gather overlap.
`_finalize_qwen3_vl_with_moe_a2a_overlap`	Apply Qwen3-VL perf defaults with MoE A2A overlap enabled.
`_finalize_qwen3_vl_with_moe_a2a_and_overlap`	Apply Qwen3-VL perf defaults with MoE A2A and optimizer-step overlap.

bridge.perf_recipes.qwen_vl.common._qwen35_vl_common( cfg: megatron.bridge.training.config.ConfigContainer, ) → None#

Apply VLM-specific performance benchmark settings for Qwen3.5-VL.

Must be called before _benchmark_common and after setting precision.

bridge.perf_recipes.qwen_vl.common._qwen35_vl_post( cfg: megatron.bridge.training.config.ConfigContainer, ) → None#

VLM post-overrides that must run after _benchmark_common.

Qwen3.5-VL disables RoPE fusion and CUDA graphs for VLM variable-length inputs; these override the perf defaults that _benchmark_common sets.

bridge.perf_recipes.qwen_vl.common._qwen35_vl_post_with_overlap( cfg: megatron.bridge.training.config.ConfigContainer, ) → None#: Apply Qwen3.5-VL post-overrides and optimizer-step param-gather overlap.

bridge.perf_recipes.qwen_vl.common._qwen35_vl_post_clear_scope_with_overlap( cfg: megatron.bridge.training.config.ConfigContainer, ) → None#: Apply Qwen3.5-VL post-overrides, clear graph scope, and enable overlap.

bridge.perf_recipes.qwen_vl.common._finalize_qwen3_vl( cfg: megatron.bridge.training.config.ConfigContainer, ) → None#: Apply Qwen3-VL perf defaults that must override generic benchmark defaults.

bridge.perf_recipes.qwen_vl.common._finalize_qwen3_vl_with_overlap( cfg: megatron.bridge.training.config.ConfigContainer, ) → None#: Apply Qwen3-VL perf defaults with optimizer-step param-gather overlap.

bridge.perf_recipes.qwen_vl.common._finalize_qwen3_vl_with_moe_a2a_overlap( cfg: megatron.bridge.training.config.ConfigContainer, ) → None#: Apply Qwen3-VL perf defaults with MoE A2A overlap enabled.

bridge.perf_recipes.qwen_vl.common._finalize_qwen3_vl_with_moe_a2a_and_overlap( cfg: megatron.bridge.training.config.ConfigContainer, ) → None#: Apply Qwen3-VL perf defaults with MoE A2A and optimizer-step overlap.