19-v1_worker_cp_utils.py.diff 720 B

12345678910111213
  1. --- baked-ours/v1/worker/cp_utils.py 2026-09-10 19:56:14.000000000 +0200
  2. +++ baked/v1/worker/cp_utils.py 2026-09-11 10:13:47.000000000 +0200
  3. @@ -35,6 +35,10 @@
  4. f"but {backend.get_name()} does not support PCP."
  5. )
  6. layer_impl = getattr(layer, "impl", None)
  7. + if layer_impl is None and hasattr(layer, "need_to_return_lse_for_decode"):
  8. + # DeepSeek-V4.1 attention runs its own forward (no
  9. + # AttentionImpl) and exposes the DCP contract on the layer.
  10. + layer_impl = layer
  11. if layer_impl is None:
  12. continue
  13. if vllm_config.speculative_config is not None and interleave_size > 1: