vllm.model_executor.warmup ¶

Modules:

• deep_gemm_warmup –

  Warmup deep_gemm kernels.

• deepseek_v4_mhc_warmup –

  Warm up DeepSeek V4 mHC TileLang kernels before serving requests.

• flashinfer_autotune_cache –

  FlashInfer autotune cache helpers.

• flashinfer_sparse_mla_warmup –

  Warmup and autotune helpers for FlashInfer sparse MLA backends.

• kernel_warmup –

  Warmup kernels used during model execution.

• qwen_triton_warmup –

  Warm up Qwen Triton kernels from the loaded model's compile keys.

• sparse_mla_triton_warmup –

  Warm up sparse-MLA Triton metadata kernels.

• v1_block_table_warmup –

  Warm up v1 block-table Triton kernels.