Atom2.7m / pretraining_curriculum.json
ucr-max's picture
Upload Atom2.7m model
271e253 verified
Raw
History Blame Contribute Delete
1.02 kB
{
"version": 1,
"transition_fraction": 0.1,
"stages": [
{
"name": "early",
"start": 0.0,
"end": 0.4,
"weights": {
"Ultra-FineWeb": 0.5,
"FineWeb-Edu": 0.38,
"FineMath": 0.05,
"Cosmopedia-v2": 0.05,
"UltraData-Math-L2-preview": 0.02
}
},
{
"name": "mid",
"start": 0.4,
"end": 0.8,
"weights": {
"Ultra-FineWeb": 0.12,
"FineWeb-Edu": 0.22,
"FineMath": 0.18,
"Cosmopedia-v2": 0.13,
"UltraData-Math-L2-preview": 0.12,
"Ultra-FineWeb-L3-en-QA-Synthetic": 0.05,
"Synthetic-Arithmetic": 0.18
}
},
{
"name": "late",
"start": 0.8,
"end": 1.0,
"weights": {
"Ultra-FineWeb": 0.105,
"FineWeb-Edu": 0.21,
"FineMath": 0.14,
"Cosmopedia-v2": 0.14,
"UltraData-Math-L2-preview": 0.105,
"Ultra-FineWeb-L3-en-QA-Synthetic": 0.2,
"Synthetic-Arithmetic": 0.1
}
}
]
}