AI API receipt · sgl-project/sglang

get_rope_config fabricates rope_theta=10000 for transformers-v5 per-layer-type rope configs

sgl-project/sglang#43158 · upstream bug reproduced by running the issue's own reproduction

Badgr cannot tell this response is wrong

Hybrid-attention models silently ran with rope_theta=10000, which degrades long-context output.

degraded long-context output is still ordinary text; Badgr has no reference answer to compare it with. The receipt therefore records success. This is a limit of what the contract can prove, shown here so it is not mistaken for protection.

What the upstream returned (replayed)

The output the bug produces, taken from the issue’s reproduction and fed to Badgr’s validators. Not a live provider call.

text: "The capital of France is Paris, and the capital of"
tool_calls: none

Badgr receipt

Written by Badgr’s receipt ledger and read back from GET /v1/receipts/replay-sglang-43158-rope-theta-fabricated.

{
  "request_id": "replay-sglang-43158-rope-theta-fabricated",
  "status": "success",
  "endpoint": "/v1/chat/completions",
  "stream": false,
  "failure_stage": null,
  "error_code": null,
  "error_message": null,
  "retryable": null,
  "provider_mode": "replay",
  "routing_decision": "replayed_upstream_output",
  "created_at": "1791524108.894355"
}

Recorded 2026-10-09 in Badgr’s local development environment. Replay cases and verdicts are pinned by backend/test_ai_api_replay_cases.py.

← All AI API receipts