fix trtllm_mla attention backend when disabling cuda graph. (#12687)

Author: cicirori

python/sglang/srt/layers/attention/trtllm_mla_backend.py

@@ -585,7 +585,7 @@ def init_forward_metadata(self, forward_batch: ForwardBatch):
             if forward_batch.forward_mode.is_target_verify():
                 max_seq = max_seq + self.num_draft_tokens
                 seq_lens = seq_lens + self.num_draft_tokens
-                self.forward_decode_metadata.seq_lens_k = seq_lens
+                self.forward_decode_metadata.seq_lens_k = seq_lens.to(torch.int32)
             elif forward_batch.forward_mode.is_draft_extend(include_v2=True):
                 max_seq = forward_batch.seq_lens_cpu.max().item()
 
@@ -604,7 +604,7 @@ def init_forward_metadata(self, forward_batch: ForwardBatch):
                 self.forward_decode_metadata.sum_seq_lens_q = sum_seq_lens_q
                 self.forward_decode_metadata.cu_seqlens_q = cu_seqlens_q
                 self.forward_decode_metadata.seq_lens_q = forward_batch.extend_seq_lens
-                self.forward_decode_metadata.seq_lens_k = seq_lens
+                self.forward_decode_metadata.seq_lens_k = seq_lens.to(torch.int32)
 
             max_seqlen_pad = self._calc_padded_blocks(max_seq)
             block_kv_indices = self._create_block_kv_indices(