[Attention] Use sparse prefill kernel for fp8 kv-cache in DeepSeek-v3.2 (#27532)

Signed-off-by: Lucas Wilkinson <lwilkins@redhat.com>
2025-12-12 08:57:47 -05:00
parent 91401c7a26
commit 3e41992fec
30 changed files with 1372 additions and 256 deletions
--- a/tests/kernels/moe/test_flashinfer.py
+++ b/tests/kernels/moe/test_flashinfer.py
@@ -206,6 +206,7 @@ def test_flashinfer_cutlass_moe_fp8_no_graph(
    topk: int,
    activation: str,
    monkeypatch,
+    workspace_init,
 ):
    current_platform.seed_everything(7)
    monkeypatch.setenv("VLLM_FUSED_MOE_CHUNK_SIZE", "8192")