[Model Runner V2][Perf] Warm up GLM-5.2 DSA indexer prefill metadata kernel (#47285)

Signed-off-by: chaunceyjiang <chaunceyjiang@gmail.com>
This commit is contained in:
Chauncey
2026-07-02 16:31:31 +00:00
committed by GitHub
parent ec0ffaacc8
commit 320ee285c9
@@ -29,6 +29,7 @@ _GENERIC_SPARSE_MLA_BACKENDS = frozenset(
"FLASHINFER_MLA_SPARSE_SM120",
}
)
_INDEXER_PREFILL_CHUNK_METADATA_BACKENDS = frozenset({"DEEPSEEK_V32_INDEXER"})
_SPARSE_PREFILL_METADATA_NUM_PREFILLS = (1, 2, 4, 8)
_SPARSE_PREFILL_METADATA_NUM_DECODES = (0, 1, 2)
@@ -326,5 +327,11 @@ def sparse_mla_triton_warmup_if_needed(worker: "Worker") -> None:
num_tokens,
compress_ratios=(1,),
)
elif _has_attention_backend(runner, _INDEXER_PREFILL_CHUNK_METADATA_BACKENDS):
_warm_prefill_chunk_metadata_kernel(
getattr(runner, "device", torch.device("cuda")),
compress_ratio=1,
query_len=num_tokens,
)
except Exception:
logger.warning("Skipping sparse MLA Triton warmup.", exc_info=True)