sgl-project/sglang · error · ValueError

--dcp-comm-backend {cfg.dcp_comm_backend} only affects the d

Error message

--dcp-comm-backend {cfg.dcp_comm_backend} only affects the decode context-parallel attention reduction and therefore requires --dcp-size / --decode-context-parallel-size > 1, but got dcp_size={cfg.dcp_size}.

What it means

--dcp-comm-backend a2a/fi_a2a only affects the attention reduction across decode context-parallel ranks; with dcp_size <= 1 there is nothing to reduce, so _handle_dcp_validation rejects the combination as a contradictory config.

Source

Thrown at python/sglang/srt/server_args.py:4262

                seen_paths.add(model_path)

    def _handle_pd_disaggregation(self):
        from sglang.srt.arg_groups.pd_disaggregation_hook import (
            handle_pd_disaggregation,
        )

        handle_pd_disaggregation(self)

    def _handle_dcp_validation(self):
        cfg = resolving_view(self)
        if cfg.dcp_size < 1:
            raise ValueError(
                "Decode context parallel size (--dcp-size / "
                "--decode-context-parallel-size) must be >= 1, but got "
                f"dcp_size={cfg.dcp_size}."
            )
        if cfg.dcp_comm_backend in ("a2a", "fi_a2a") and cfg.dcp_size <= 1:
            raise ValueError(
                f"--dcp-comm-backend {cfg.dcp_comm_backend} only affects the "
                "decode context-parallel attention reduction and therefore "
                "requires --dcp-size / --decode-context-parallel-size > 1, but "
                f"got dcp_size={cfg.dcp_size}."
            )
        if cfg.dcp_comm_backend == "fi_a2a" and not is_cuda():
            raise ValueError(
                "--dcp-comm-backend fi_a2a delegates the exchange to FlashInfer's "
                "MNNVL All-to-All kernel, which requires an NVIDIA CUDA platform "
                "with SM90+ and MNNVL fabric memory (e.g. GB200 NVL72). The "
                "authoritative fabric probe runs at model-runner init; use 'a2a' "
                "or 'ag_rs' on clusters without MNNVL."
            )
        if cfg.dcp_replicate_q_proj:
            if cfg.dcp_size <= 1:
                raise ValueError("--dcp-replicate-q-proj requires --dcp-size > 1.")
            if cfg.dcp_comm_backend not in ("a2a", "fi_a2a"):
                raise ValueError(

View on GitHub (pinned to 0132848349)

Solutions

1. Set --dcp-size (or --decode-context-parallel-size) to a value > 1 when using a2a/fi_a2a.
2. Or remove --dcp-comm-backend if you are not actually running decode context parallelism.

Example fix

# before
python -m sglang.launch_server --model m --dcp-comm-backend a2a
# after
python -m sglang.launch_server --model m --dcp-size 2 --dcp-comm-backend a2a

if dcp_comm_backend in ('a2a', 'fi_a2a'):
    assert dcp_size and dcp_size > 1, 'a2a backends require dcp_size > 1'

def dcp_backend_config_ok(backend: str, dcp_size: int) -> bool:
    return backend not in ('a2a', 'fi_a2a') or dcp_size > 1

When it happens

Trigger: Passing --dcp-comm-backend a2a (or fi_a2a) together with --dcp-size 1 or omitting --dcp-size entirely.

Common situations: Copy-pasting a DCP launch command and dropping the --dcp-size flag; enabling the backend 'for later' while running single-rank decode; leftover flags from a cluster config template.

Related errors

• Decode context parallel size (--dcp-size / --decode-context-
• Kimi-K3 DCP with decode_attention_backend='cutedsl_mla' requ
• Kimi-K3 DCP with decode_attention_backend='cutedsl_mla' requ
• Decode attention backend for Kimi-K3 DCP must be 'cutedsl_ml
• PD decode DCP requires --disaggregation-transfer-backend moo

AI-assisted analysis of sgl-project/sglang@0132848349 (2026-08-28). Data as JSON: /api/errors/ece937cdd11df6ea. Report an issue: GitHub.