minhy112 commited on
Commit
b1185ac
·
verified ·
1 Parent(s): 416d04b

Archive verified research artifact: complete.json

Browse files
artifacts/experiments/runs/test-qwen3-mmqa_ti-s17/complete.json ADDED
@@ -0,0 +1,16 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "files": {
3
+ "freeze.json": "0f91ced79b85d92a36b93fdc1e6408c29dd32fca6f6cac07abb263af42438226",
4
+ "run.json": "977653a40a36f1b011ca2cd95e0d430fd2eb7a6e511f358ec7deae8114166311",
5
+ "summary.json": "11d04dd447184deec0c65ea8a363f17811c366e3262dc122155966e1d18964d1",
6
+ "test/evaluation/per_question.jsonl": "a6908ba0d9f1c622b83b66e3de62013a51664798d5f56818c171ecfa45224caa",
7
+ "test/evaluation/summary.json": "787d8420162487d6fa4a6a7d05e47ac7fdcf13db19e22953cadaefaeacc730f3",
8
+ "test/predictions.jsonl": "b017a0688449783f446dd7b8334770af4bdb52b6b5adec65f3e29d22d57ccad5",
9
+ "val_calibration/evaluation/per_question.jsonl": "d1c10f75e548126bc2583d7c93ebab4c620ac1db24218a6b08875b4817b107d1",
10
+ "val_calibration/evaluation/summary.json": "0022cb523bab5a2714fb89cab3a6ecaa3ae95575e5dbb9f0e355a58e126434ae",
11
+ "val_calibration/predictions.jsonl": "fe83c246052a3393b8a143143522f8f66c4f6efe493ab7e27f7886f75cedf22a",
12
+ "validation_canaries.json": "722d0b57c6f395928ea1c07152fcd6dba767a16e286360957ff8a8140e6eb43e"
13
+ },
14
+ "spec_sha256": "cd83a398465e7b7d6bbfe5972354f0a0c32e7ff33a5fd19a2fdc6e521e402c0f",
15
+ "test_used": true
16
+ }