[aia-eval] LME-681 Configure eval for codex+gemini in AIA

GitOrigin-RevId: 2e5beeb7968b2fab572c5a259d6623da87d7dbc4
This commit is contained in:
Anton Spilnyy
2025-11-23 20:09:42 +00:00
committed by intellij-monorepo-bot
parent 86fc70bc33
commit 00e64bd481
2 changed files with 76 additions and 0 deletions
@@ -0,0 +1,38 @@
{
"projectPath": "../ml-eval/jackson-core",
"repository": "https://github.com/fasterxml/jackson-core",
"revision": "f1dc3c512d211ae3e14fb59af231caebf037d510",
"outputDir": "bin/ml-eval-code-generation-swe-java-output",
"strategy": {
"instanceId": "fasterxml__jackson-core-1142",
"runner": "CODEX",
"invocationParams": {
"runFeatureTimeoutMin": 20
}
},
"fileDataset": {
"url": "remote:https://huggingface.co/datasets/JetBrains/eval_plugin/resolve/main/code-generation-swe/java/java_swe_bench.json"
},
"interpret": {
"saveLogs": true,
"sessionProbability": 1.0,
"sessionSeed": null,
"sessionsLimit": 400,
"filesLimit": 100,
"order": "LINEAR",
"trainTestSplit": 70,
"registry": "",
"iterationCount": 1
},
"reports": {
"evaluationTitle": "Default",
"sessionsFilters": [],
"comparisonFilters": []
},
"language": "Java",
"actions": {
"evaluationRoots": [
"."
]
}
}
@@ -0,0 +1,38 @@
{
"projectPath": "../ml-eval/jackson-core",
"repository": "https://github.com/fasterxml/jackson-core",
"revision": "f1dc3c512d211ae3e14fb59af231caebf037d510",
"outputDir": "bin/ml-eval-code-generation-swe-java-output",
"strategy": {
"instanceId": "fasterxml__jackson-core-1142",
"runner": "GEMINI",
"invocationParams": {
"runFeatureTimeoutMin": 20
}
},
"fileDataset": {
"url": "remote:https://huggingface.co/datasets/JetBrains/eval_plugin/resolve/main/code-generation-swe/java/java_swe_bench.json"
},
"interpret": {
"saveLogs": true,
"sessionProbability": 1.0,
"sessionSeed": null,
"sessionsLimit": 400,
"filesLimit": 100,
"order": "LINEAR",
"trainTestSplit": 70,
"registry": "",
"iterationCount": 1
},
"reports": {
"evaluationTitle": "Default",
"sessionsFilters": [],
"comparisonFilters": []
},
"language": "Java",
"actions": {
"evaluationRoots": [
"."
]
}
}