PolyPythias - a EleutherAI Collection

EleutherAI 's Collections

Pythia Scaling Suite

Quirky Models and Datasets

Quirky Math (archive)

Pile-T5 T5x Checkpoints

Sparse Autoencoders

PolyPythias

updated Apr 9, 2025

EleutherAI/polypythias-evals

Preview • Updated Sep 11, 2024 • 2.9k
EleutherAI/pile-preshuffled-seeds

Updated Feb 12 • 400 • 1

Note Training data information for each seed.
EleutherAI/pythia-14m-deduped

Text Generation • 39.2M • Updated Feb 12 • 13.7k • 30
EleutherAI/pythia-14m-seed1

Text Generation • Updated Jan 11 • 2.88k
EleutherAI/pythia-14m-seed2

Text Generation • Updated Jan 11 • 1.9k
EleutherAI/pythia-14m-seed3

Text Generation • Updated Jan 11 • 1.94k
EleutherAI/pythia-14m-seed4

Text Generation • Updated Jan 11 • 1.35k
EleutherAI/pythia-14m-seed5

Text Generation • Updated Jan 11 • 1.86k
EleutherAI/pythia-14m-seed6

Text Generation • Updated Jan 11 • 1.61k
EleutherAI/pythia-14m-seed7

Text Generation • Updated Jan 11 • 1.74k
EleutherAI/pythia-14m-seed8

Text Generation • Updated Jan 11 • 1.34k
EleutherAI/pythia-14m-seed9

Text Generation • Updated Jan 11 • 1.73k
EleutherAI/pythia-31m-deduped

Text Generation • 55.7M • Updated Feb 12 • 1.35k • 5
EleutherAI/pythia-31m-seed1

Text Generation • Updated Jan 11 • 1.21k
EleutherAI/pythia-31m-seed2

Text Generation • Updated Jan 11 • 1.09k
EleutherAI/pythia-31m-seed3

Text Generation • Updated Jan 11 • 1.04k
EleutherAI/pythia-31m-seed4

Text Generation • Updated Jan 11 • 975
EleutherAI/pythia-31m-seed5

Text Generation • Updated Jan 11 • 908
EleutherAI/pythia-31m-seed6

Text Generation • Updated Jan 11 • 974
EleutherAI/pythia-31m-seed7

Text Generation • Updated Jan 11 • 909
EleutherAI/pythia-31m-seed8

Text Generation • Updated Jan 11 • 925
EleutherAI/pythia-31m-seed9

Text Generation • Updated Jan 11 • 1.1k
EleutherAI/pythia-70m

95.6M • Updated Nov 21, 2023 • 528k • 85
EleutherAI/pythia-70m-seed1

Text Generation • Updated Jan 11 • 8.79k
EleutherAI/pythia-70m-seed2

Text Generation • Updated Jan 11 • 8.57k
EleutherAI/pythia-70m-seed3

Text Generation • Updated Jan 11 • 1.87k
EleutherAI/pythia-70m-seed4

Text Generation • Updated Jan 11 • 1.54k
EleutherAI/pythia-70m-seed5

Text Generation • Updated Jan 11 • 1.53k
EleutherAI/pythia-70m-seed6

Text Generation • Updated Jan 11 • 1.79k
EleutherAI/pythia-70m-seed7

Text Generation • Updated Jan 11 • 1.63k
EleutherAI/pythia-70m-seed8

Text Generation • Updated Jan 11 • 1.63k
EleutherAI/pythia-70m-seed9

Text Generation • Updated Jan 11 • 1.51k
EleutherAI/pythia-160m

Text Generation • 0.2B • Updated Jul 9, 2023 • 3.25M • 45
EleutherAI/pythia-160m-seed1

Text Generation • 0.2B • Updated Jan 11 • 61k
EleutherAI/pythia-160m-seed2

Text Generation • 0.2B • Updated Jan 11 • 59.1k
EleutherAI/pythia-160m-seed3

Text Generation • 0.2B • Updated Jan 11 • 34.3k
EleutherAI/pythia-160m-seed4

Text Generation • Updated Jan 11 • 17.5k • 1
EleutherAI/pythia-160m-seed5

Text Generation • Updated Jan 11 • 17.9k
EleutherAI/pythia-160m-seed6

Text Generation • Updated Jan 11 • 17.9k
EleutherAI/pythia-160m-seed7

Text Generation • Updated Jan 11 • 7.45k
EleutherAI/pythia-160m-seed8

Text Generation • Updated Jan 11 • 7.31k
EleutherAI/pythia-160m-seed9

Text Generation • Updated Jan 11 • 6.58k
EleutherAI/pythia-410m

Text Generation • 0.5B • Updated Jul 9, 2023 • 221k • 39
EleutherAI/pythia-410m-seed1

Text Generation • Updated Jan 11 • 22.6k
EleutherAI/pythia-410m-seed2

Text Generation • Updated Jan 11 • 20.7k
EleutherAI/pythia-410m-seed3

Text Generation • Updated Jan 11 • 6.04k
EleutherAI/pythia-410m-seed4

Text Generation • Updated Jan 11 • 5.22k
EleutherAI/pythia-410m-seed5

Text Generation • Updated Jan 11 • 5.24k
EleutherAI/pythia-410m-seed6

Text Generation • Updated Jan 11 • 4.99k • 1
EleutherAI/pythia-410m-seed7

Text Generation • Updated Jan 11 • 5.14k
EleutherAI/pythia-410m-seed8

Text Generation • Updated Jan 11 • 5.83k
EleutherAI/pythia-410m-seed9

Text Generation • Updated Jan 11 • 6.51k
EleutherAI/pythia-160m-data-seed1

Text Generation • Updated Jan 11 • 113k

Note Version where the data order and weight initialization seeds are decoupled. Here, only the data seed is different from pythia-160m ("seed 0").
EleutherAI/pythia-160m-data-seed2

Text Generation • Updated Jan 11 • 14.7k

Note Version where the data order and weight initialization seeds are decoupled. Here, only the data seed is different from pythia-160m ("seed 0").
EleutherAI/pythia-160m-data-seed3

Text Generation • Updated Jan 11 • 10.7k

Note Version where the data order and weight initialization seeds are decoupled. Here, only the data seed is different from pythia-160m ("seed 0").
EleutherAI/pythia-160m-weight-seed1

Text Generation • Updated Jan 11 • 33.5k

Note Version where the data order and weight initialization seeds are decoupled. Here, only the weight initialization seed is different from pythia-160m ("seed 0").
EleutherAI/pythia-160m-weight-seed2

Text Generation • Updated Jan 11 • 20.4k

Note Version where the data order and weight initialization seeds are decoupled. Here, only the weight initialization seed is different from pythia-160m ("seed 0").
EleutherAI/pythia-160m-weight-seed3

Text Generation • Updated Jan 11 • 8.2k

Note Version where the data order and weight initialization seeds are decoupled. Here, only the weight initialization seed is different from pythia-160m ("seed 0").
PolyPythias: Stability and Outliers across Fifty Language Model Pre-Training Runs

Paper • 2503.09543 • Published Mar 12, 2025