24 lines
909 B
JSON
24 lines
909 B
JSON
|
{
|
||
|
|
"schema_version": 1,
|
||
|
|
"split_rule": "sort source-train prompts by sha256('flagship-v1||'+prompt_id); first round(10%) validation",
|
||
|
|
"validation_fraction": 0.1,
|
||
|
|
"source_dir": "/NHNHOME/WORKSPACE/26msit001_A/BASE/aipr_lab_sjkim_eval/revision_qwen3_8b/full_iter1/precomputed/avg_oracle",
|
||
|
|
"base_model": "Qwen/Qwen3-8B",
|
||
|
|
"counts": {
|
||
|
|
"train": 16780,
|
||
|
|
"validation": 1864,
|
||
|
|
"sealed_test": 604
|
||
|
|
},
|
||
|
|
"prompt_id_sha256": {
|
||
|
|
"train": "fcf6f2ea5a0b7bcd41179e6f5288f4d850b2d04bdb729da91907bd18bf733a69",
|
||
|
|
"validation": "a87ba6f519192164daa487b109a6547dd8cbde46a27a373d6da201ca2d061b8f",
|
||
|
|
"sealed_test": "b398ced6af6cf73d731e1a5f32a0631d8060323d51eda1e375695566bf5f1867"
|
||
|
|
},
|
||
|
|
"overlaps": {
|
||
|
|
"train_validation": 0,
|
||
|
|
"train_sealed_test": 0,
|
||
|
|
"validation_sealed_test": 0
|
||
|
|
},
|
||
|
|
"sealed_test_policy": "prompts only until S4a; no objective scores or model generations used in S0-S3"
|
||
|
|
}
|