[Dependency] Upgrade FlashInfer to 0.7.2 and restore KDA auto backend (#61913)

Signed-off-by: Anerudhan Gopal <agopal@nvidia.com>
Co-authored-by: OpenAI Codex <noreply@openai.com>

requirements/cuda.txt

@@ -10,7 +10,7 @@ torchaudio==2.11.0
 torchvision==0.29.0 # Required for phi3v processor. See https://github.com/pytorch/vision?tab=readme-ov-file#installation for corresponding version
 torchcodec >= 0.14
 PyNvVideoCodec==2.0.4
-flashinfer-python==0.7.1
+flashinfer-python==0.7.2
 apache-tvm-ffi==0.1.11
 tilelang==0.1.12
 nvidia-cudnn-frontend>=1.19.1

vllm/models/kimi_k3/nvidia/kda.py

@@ -456,7 +456,7 @@ def _flashinfer_kda_prefill(
         beta_is_logit=True,
         seq_order=seq_order,
         prefill_workspace=prefill_workspace,
-        backend="cute-dsl",
+        backend="auto",
     )
     return output, initial_state