[DSv4.1] Avoid runtime recompiles of _ring_slot_mapping_kernel (#59119)
Signed-off-by: Juntian Liu <Juntianl777@gmail.com> Co-authored-by: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Juntian Liu committed
Sep 29, 2026 at 00:31 UTC
30f5c01d4ebb5c493ed86195db24d0ec5b3ce3d7
1 file changed
+1
-1
vllm/models/deepseek_v41/compressor.py
+1
-1
@@ -57,7 +57,7 @@ class CompressorMetadata:
57
token_to_req_indices: torch.Tensor # [num_tokens]
58
59
60
-@triton.jit
60
+@triton.jit(do_not_specialize=["block_table_stride", "num_actual_tokens", "num_tokens"])
61
def _ring_slot_mapping_kernel(
62
slot_mapping_ptr,
63
block_table_ptr,