AI API receipt · sgl-project/sglang

get_hf_text_config erases the text sub-config dtype on the thinker_config path

sgl-project/sglang#43157 · upstream bug reproduced by running the issue's own reproduction

Badgr cannot tell this response is wrong

Qwen2.5-Omni / Qwen3-Omni text dtype became None at load time, so the model loads in the wrong precision.

a model loaded in the wrong dtype still answers; the reply is valid text and carries no trace of the precision. 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: "Hello! How can I help you today?"
tool_calls: none

Badgr receipt

Written by Badgr’s receipt ledger and read back from GET /v1/receipts/replay-sglang-43157-thinker-dtype-erased.

{
  "request_id": "replay-sglang-43157-thinker-dtype-erased",
  "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": "1791524108.8935192"
}

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