diff --git a/vllm/models/deepseek_v41/compressor.py b/vllm/models/deepseek_v41/compressor.py index 42335ae84b..54dc9c4a13 100644 --- a/vllm/models/deepseek_v41/compressor.py +++ b/vllm/models/deepseek_v41/compressor.py @@ -57,7 +57,7 @@ class CompressorMetadata: token_to_req_indices: torch.Tensor # [num_tokens] -@triton.jit +@triton.jit(do_not_specialize=["block_table_stride", "num_actual_tokens", "num_tokens"]) def _ring_slot_mapping_kernel( slot_mapping_ptr, block_table_ptr,