sgl-project/sglang · error · ValueError

DSpark with dp attention + moe_a2a_backend={} requires SGLAN

Error message

DSpark with dp attention + moe_a2a_backend={} requires SGLANG_RAGGED_VERIFY_MODE=static.

What it means

When DSpark runs with dp attention and a non-'none' MoE a2a backend (i.e. megamoe) on CUDA, the ragged verification path must be in static mode. The hook reads SGLANG_RAGGED_VERIFY_MODE via read_ragged_verify_mode() and rejects anything other than STATIC.

Source

Thrown at python/sglang/srt/arg_groups/speculative_hook.py:372

    # dp_size==1 with dp_attention is a degenerate flag under DSV4 CP; skip DP-only checks.
    if cfg.enable_dp_attention and cfg.dp_size > 1:
        if not cfg.enable_dp_lm_head:
            raise ValueError("DSpark with dp attention requires --enable-dp-lm-head.")
        if not _is_npu and cfg.moe_a2a_backend not in ("none", "megamoe"):
            raise ValueError(
                "DSpark with dp attention supports moe_a2a_backend 'none' "
                "(built-in TP MoE) or 'megamoe', got "
                f"{cfg.moe_a2a_backend!r}."
            )
        if not _is_npu and cfg.moe_a2a_backend != "none":
            from sglang.srt.speculative.ragged_verify import (
                RaggedVerifyMode,
                read_ragged_verify_mode,
            )

            if read_ragged_verify_mode() is not RaggedVerifyMode.STATIC:
                raise ValueError(
                    "DSpark with dp attention + "
                    f"moe_a2a_backend={cfg.moe_a2a_backend!r} requires "
                    "SGLANG_RAGGED_VERIFY_MODE=static."
                )
        if cfg.attn_cp_size > 1:
            raise ValueError(
                "DSpark with dp attention does not support context parallel "
                f"(attn_cp_size={cfg.attn_cp_size})."
            )
        if (
            not _is_npu
            and cfg.speculative_moe_a2a_backend is not None
            and cfg.speculative_moe_a2a_backend != cfg.moe_a2a_backend
        ):
            raise ValueError(
                "DSpark ignores --speculative-moe-a2a-backend; with dp attention it "
                f"must match the target moe_a2a_backend={cfg.moe_a2a_backend!r} "
                f"(got {cfg.speculative_moe_a2a_backend!r})."

View on GitHub (pinned to 0132848349)

Solutions

  1. Export SGLANG_RAGGED_VERIFY_MODE=static before launching
  2. Keep --moe-a2a-backend none to avoid the requirement entirely
  3. Add the env var to your container/systemd/k8s service definition so it persists

Example fix

# before
SGLANG_RAGGED_VERIFY_MODE=dynamic python -m sglang.launch_server ...
# after
SGLANG_RAGGED_VERIFY_MODE=static python -m sglang.launch_server ...
Defensive patterns

Strategy: validation

Validate before calling

import os
if args.enable_dp_attention and args.dp_size > 1 and args.moe_a2a_backend not in ('none',):
    if os.environ.get('SGLANG_RAGGED_VERIFY_MODE', '').lower() != 'static':
        raise SystemExit('export SGLANG_RAGGED_VERIFY_MODE=static first')

Prevention

When it happens

Trigger: DSpark + dp attention (dp_size>1) + moe_a2a_backend=megamoe while the env var SGLANG_RAGGED_VERIFY_MODE is unset or set to a non-static value.

Common situations: Fresh shell/container where SGLANG_RAGGED_VERIFY_MODE was never exported; the var set to 'dynamic' by an experimental preset.

Understand the failure class

Background: "environment variable is not set" and "Missing keys in environment" errors: what missing required env var messages mean and how to fix them — this error's family across 28 libraries.

Related errors

AI-assisted analysis of sgl-project/sglang@0132848349 (2026-08-28). Data as JSON: /api/errors/315603c12a4a672c. Report an issue: GitHub.