Skip to content

release: v2.1.15 QwenWork catalog 503 - #84

Merged
wicm84266964 merged 2 commits into
mainfrom
release/v2.1.15
Sep 24, 2026
Merged

wicm84266964 merged 2 commits into
mainfrom
release/v2.1.15

Conversation

@wicm84266964

Copy link
Copy Markdown
Owner

Summary

  • 千问办公 1.0.4 对话补上 business.product / business.type,与 Cosy-Business-* 头同源,修复固定 503 Model catalog unavailable([Bug]: QwenWork上游返回异常 #82)。
  • 推理请求改走官方 Encode=1 组包,依赖 wasmtime。
  • 版本号升到 2.1.15。

Test plan

  • pytest tests/test_qwenwork.py(24 passed)
  • 本机官方账号:flash、pro、qwen3.8-max-preview 均返回 200
  • 维护者重启本地网关后复测通过

QwenWork 1.0.4 resolves the model catalog from body.business.product and
body.business.type. Header-only requests return 503 Model catalog unavailable.
Chat now copies those fields from the Cosy-Business headers and sends the
official Encode=1 body.
Ship the QwenWork catalog 503 fix from #82.
@wicm84266964
wicm84266964 merged commit f0d83f7 into main Sep 24, 2026
1 check passed
@wicm84266964
wicm84266964 deleted the release/v2.1.15 branch September 24, 2026 02:21
ZJFERAL added a commit to ZJFERAL/Buddy2api that referenced this pull request Sep 24, 2026
Brings in main's released QwenWork 1.0.4 fix (v2.1.15, wicm84266964#82/wicm84266964#84):
body `business.product`/`business.type` for catalog lookup, official
`Encode=1` WASM infer packing (`providers/qwenwork/encode.py`, +wasmtime).

Conflict resolutions:
- requirements.txt: union — keep websockets (more) and wasmtime (main).
- providers/qwenwork/constants.py: take main's re-frozen 1.0.4 identity
  (`COSY_VERSION=1.1.32`, `IDE_VERSION=1.0.4`, `USER_AGENT=qoderwork/1.0.4`).
  `encode.py` feeds COSY_VERSION into `wasm.context_new`, so more's
  speculative 1.1.0/1.1.18 identity would have packed the wrong version.
  Kept more's additive OPENAPI_BASE / REALM_GATEWAY / MODEL_SERVER_PATH
  (still imported by scripts/qwenwork_probe.py), plus main's MACHINE_TYPE
  and DATA_POLICY.
- tests/test_qwenwork.py: union of both import sets; envelope_status (more)
  and static_headers / translate_model / prepare_infer (main) all exist in
  the merged chat.py.

Docs updated to match the merged behavior: AGENTS.md, upstream-protocol-notes.md
and multi-channel-v2.md no longer claim the HTTP+SSE path is deprecated.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant