Slayer149-balanced / reports /tokenizer-selection.json
kacperwikiel's picture
Update tokenizer, configuration and training reports
d6a86af verified
Raw History Blame Contribute Delete
4.1 kB
{
"selected": {
"vocab": 24576,
"ffn": 2496,
"parameters": 149333081,
"weighted_tokens_per_byte": 0.21982151683772885,
"measurements": {
"fineweb_edu": {
"tokens": 108258,
"bytes": 500330,
"bytes_per_token": 4.621644589776275
},
"dclm_edu": {
"tokens": 113785,
"bytes": 502955,
"bytes_per_token": 4.42022234916729
},
"cosmopedia": {
"tokens": 96667,
"bytes": 500651,
"bytes_per_token": 5.179130416791666
},
"finemath": {
"tokens": 149111,
"bytes": 509813,
"bytes_per_token": 3.4190167056756375
},
"edge_cases": {
"tokens": 67,
"bytes": 130,
"bytes_per_token": 1.9402985074626866
}
},
"fit_seconds": 6.842154794954695
},
"candidates": [
{
"vocab": 16384,
"ffn": 2592,
"parameters": 148927577,
"weighted_tokens_per_byte": 0.2296576476963851,
"measurements": {
"fineweb_edu": {
"tokens": 113155,
"bytes": 500330,
"bytes_per_token": 4.421634041801069
},
"dclm_edu": {
"tokens": 118553,
"bytes": 502955,
"bytes_per_token": 4.242448525132219
},
"cosmopedia": {
"tokens": 101369,
"bytes": 500651,
"bytes_per_token": 4.938896506821612
},
"finemath": {
"tokens": 155627,
"bytes": 509813,
"bytes_per_token": 3.275864727842855
},
"edge_cases": {
"tokens": 72,
"bytes": 130,
"bytes_per_token": 1.8055555555555556
}
},
"fit_seconds": 9.067338421940804
},
{
"vocab": 24576,
"ffn": 2496,
"parameters": 149333081,
"weighted_tokens_per_byte": 0.21982151683772885,
"measurements": {
"fineweb_edu": {
"tokens": 108258,
"bytes": 500330,
"bytes_per_token": 4.621644589776275
},
"dclm_edu": {
"tokens": 113785,
"bytes": 502955,
"bytes_per_token": 4.42022234916729
},
"cosmopedia": {
"tokens": 96667,
"bytes": 500651,
"bytes_per_token": 5.179130416791666
},
"finemath": {
"tokens": 149111,
"bytes": 509813,
"bytes_per_token": 3.4190167056756375
},
"edge_cases": {
"tokens": 67,
"bytes": 130,
"bytes_per_token": 1.9402985074626866
}
},
"fit_seconds": 6.842154794954695
},
{
"vocab": 32768,
"ffn": 2384,
"parameters": 149019737,
"weighted_tokens_per_byte": 0.2144127319251325,
"measurements": {
"fineweb_edu": {
"tokens": 105600,
"bytes": 500330,
"bytes_per_token": 4.737973484848485
},
"dclm_edu": {
"tokens": 111026,
"bytes": 502955,
"bytes_per_token": 4.530065029812836
},
"cosmopedia": {
"tokens": 94072,
"bytes": 500651,
"bytes_per_token": 5.321998044051365
},
"finemath": {
"tokens": 145642,
"bytes": 509813,
"bytes_per_token": 3.50045316598234
},
"edge_cases": {
"tokens": 63,
"bytes": 130,
"bytes_per_token": 2.0634920634920637
}
},
"fit_seconds": 7.331924270954914
}
],
"criterion": "Smallest vocabulary within 3% of best weighted held-out token compression; provisional pending speed check.",
"tokenizer_training_sha256": "c15bad040cf3eaf9e0053539bff3039601f9a4db7e89a0acd6db0e03e4cced49",
"tokenizer_sha256": "d1ffe47f6fd8cc19ba7c3e5d2537699d6a99d09b584d3319944f4e40fe6c44ea",
"normalization": "none; byte fallback alphabet; preserve Unicode and whitespace; digit groups max3",
"parameters_cap": 149500000,
"data_mix": {
"fineweb_edu": 0.65,
"dclm_edu": 0.2,
"cosmopedia": 0.1,
"finemath": 0.05
},
"status": "roundtrip_passed"
}