diff --git a/indexer.py b/indexer.py index b6fdc5b..a856fe2 100644 --- a/indexer.py +++ b/indexer.py @@ -608,7 +608,8 @@ class DeepseekV32IndexerMetadataBuilder(AttentionMetadataBuilder): # (csrc/apis/attention.hpp). Unsqueeze to (B, 1) so downstream # kernels see the same (B, next_n) layout as the MTP path. if seq_lens.dim() == 1: - seq_lens = seq_lens.unsqueeze(-1).contiguous() + seq_lens = seq_lens.unsqueeze(-1) + seq_lens = seq_lens.contiguous() # DeepGEMM is required for the paged MQA logits on CUDA devices if current_platform.is_cuda() and has_deep_gemm():