code-sprout-30m / completion_data_report.json
Raymond1938-4's picture
Release experimental CodeSprout 30M checkpoint
61790a1 verified
Raw History Blame Contribute Delete
601 Bytes
{
"stage": "Original completion alignment",
"license": "Apache-2.0",
"generation": "Deterministic templates; no pretrained teacher. Python examples parsed by ast. Train/valid are disjoint complete strings, but share template families.",
"examples": {
"html": {
"valid": 160,
"train": 1440
},
"javascript": {
"valid": 125,
"train": 1126
},
"python": {
"valid": 125,
"train": 1129
},
"typescript": {
"valid": 125,
"train": 1126
}
},
"rehearsal": "About 10% original source-token replay"
}