TRAIN MY OWN GPT
based on karpathy's microgpt · runs in your browser
Dataset
choose dataset...
YC Startups — 5,000+ startup names
Baby Names — 2,000+ popular names
Dinosaurs — 1,500+ species names
English Words — 10,000 common words
drop or click to upload .txt
one entry per line
Tokenizer
waiting for dataset...
Architecture
embedding dim
16
attention heads
4
layers
1
context window
16
training steps
1000
learning rate
0.010
params:
5,248
168K FLOPs/step
Training
▶ TRAIN
Metrics
waiting for training...
Generate
waiting for model...
100% in-browser · no data leaves your device · 5,248 parameters
built by jay