)]}'
{
  "commit": "1cbb15574990e0fa134f2ed5644e9907e314295a",
  "tree": "855c1c63e848665091180a1415da3cc1650e243c",
  "parents": [
    "18e49583246446996f653974c7bbbdc859dd6776"
  ],
  "author": {
    "name": "Marc Kupietz",
    "email": "kupietz@ids-mannheim.de",
    "time": "Sat Sep 05 15:30:52 2026 +0200"
  },
  "committer": {
    "name": "Marc Kupietz",
    "email": "kupietz@ids-mannheim.de",
    "time": "Tue Sep 08 20:56:00 2026 +0200"
  },
  "message": "DocPrompting: Switch GLM-5.3-Flash from Synthetic to OpenRouter\n\nThe endpoint stays OpenAI compatible, so only the base url, the key and\nthe model id change, from hf:zai-org/GLM-5.3-Flash to z-ai/glm-5.3-flash.\nAn hf: id now fails with a hint rather than being sent somewhere it is\nnot known, since it survives in local .Renviron files.\n\nOpenRouter serves the model from two dozen hosts and would pick one by\navailability, which the tests cannot use: the hosts differ in\nquantization, and in how long they take for a documentation prompt by\nmore than two orders of magnitude - one second on Baseten against seven\nminutes on GMICloud. So the hosts are named and everything outside the\nlist ruled out, which keeps a change in the generated code about the\nReadme rather than about which machine answered.\n\nThe list has to hold more than one, and not because a host might be down:\nOpenRouter reaches them through pools it shares between all of its users,\nrather than through our key, and those pools are regularly exhausted. Of\nten prompts pinned to Baseten alone, two came back rate-limited. A\nBaseten key of one\u0027s own, entered under Settings -\u003e Integrations, takes\nthat away for the host at the front; Relace and Z.ai carry what is left,\nat some fifteen seconds and tens of seconds against Baseten\u0027s one. The\ntimeout is five minutes for their sake, and\nRKORAP_OPENROUTER_PROVIDER overrides the whole list.\n\nRelace is the cheapest of the hosts and the quickest of those on shared\ncapacity, but it serves the model in fp4 where the others use fp8, which\nis why it is not at the front: what the tests measure should not depend\non how coarsely a machine was quantized. It solved the collocation task\nin five of five runs all the same.\n\nNeeds OPENROUTER_API_KEY as a masked variable in the GitLab project.\n\nChange-Id: Iba289424970aa933d58798be4c4b07b74c02f2f6\nCo-Authored-By: Claude Opus 5 \u003cnoreply@anthropic.com\u003e\n",
  "tree_diff": [
    {
      "type": "modify",
      "old_id": "d697bad42c6bdc668f850c86dc4938ddf779a517",
      "old_mode": 33188,
      "old_path": ".gitlab-ci.yml",
      "new_id": "860ce7853b24422a8ab7ed47fb5c5937d3c24929",
      "new_mode": 33188,
      "new_path": ".gitlab-ci.yml"
    },
    {
      "type": "add",
      "old_id": "0000000000000000000000000000000000000000",
      "old_mode": 0,
      "old_path": "/dev/null",
      "new_id": "04f08ebbc86f788403cde84526f839318e86c31d",
      "new_mode": 33188,
      "new_path": "RKorAPClient_1.4.0.tar.gz"
    },
    {
      "type": "modify",
      "old_id": "c91447bdd8bcf1cbca7d7bfd22f9e4a90f4026d0",
      "old_mode": 33188,
      "old_path": "tests/testthat/test-readme-against-llm.R",
      "new_id": "561d587caea46002d6ab345ab04fb581360a6a10",
      "new_mode": 33188,
      "new_path": "tests/testthat/test-readme-against-llm.R"
    }
  ]
}