tysam-code/hlb-gpt
Minimalistic, extremely fast, and hackable researcher's toolbench for GPT models in 307 lines of code. Reaches <3.8 validation loss on wikitext-103 on a single A100 in <100 seconds. Scales to larger models with one parameter change (feature currently in alpha).
PythonApache-2.0
Stargazers
- 0xsanny
- abodacsOpenCoast
- ahxtTexas A&M University
- andrewncgetcartwheel.com
- anoother
- bimhud
- chuan298
- danlou
- dmvaldmanSF
- dumpmemory
- fly51flyPRIS
- garyfanhkuHK
- gngdbCerebras Systems
- go2carter
- gurusuraSura Systems Private Limited
- jaspock
- josegemezGarquitectos
- KeAWangStanford, CA
- kenroxistanbul - silivri
- linhduongtuanKTH Royal Institute of Technology
- linyanzhong
- monopoly-db
- neverix
- owlwangnone
- pavelklymenkoSan Francisco Bay Area
- pdtgctMarlborough, MA
- qinhongweiBeijing
- quoccuonglqdHo Chi Minh City, Viet Nam
- rodonguyen
- SandalotsVolcanak
- shahiraj19
- snimuBMW
- sradc
- SushantDaga
- SwayamInSyncIndia
- tysam-code