This model was trained for 30 epochs, best epoch was 11, outputs: Training complete. Best epoch: 11 Best validation loss: 1.6153 Checkpoint: /content/tiny50_karpathy_tinystories_gpt4_clean.pt Temperature 0.55: a rat is near. to e a ir i e e e a a e e i a a e aro e a t t e a e t e e an e e e i a i e e e i e e a t t a a e t i aso e a e e t i to e e e e e a t i a t t a a e i e a a e t e t an e t e e e e t a e ir e e a t t e a i t e i t t ir a t e a a t t e a i t e e t e t a t e e t e a e t a e t a e t a i e e e e e e t ar e e e e a e a t e e it e t anto e e e e en e e ito a en an t e e e e e t e a te e t an i t e e in t a e e t a e t e a e a e t i t t e e in a e to a e e a e a a a t e e t a a t e i an a e e e e i e e Temperature 0.90: a rat is near. o as i t e e s enor ar o e a iso ene a en at s ino a e o e a se t en t son as i a e e a a a an en s e e e a en as an is s i i i t s to e a ee ertoen t s an at o e i a t ito te e e a ato o an t e i o e in o eri te so ani ane t t e a t o e ee i a at at a an e ee o a ira a o t ot oe e e e t o e a o to t s e a e o a as it ti a s o tot ase t. s e t e esese s e it o e as e to an en etise in ent t i so t tot s t te as st ir s stora ato orano ar at e e os ani e te s s i a eno t so ee an etee ot. to et
Why DinoResearch created this?
Because we want to see how a only 50-parameter AI model (that can run on 1 MB of RAM, fits on a floppy disk, and is just 3 KB, less than a PNG!) Even if this model doesn't know how to speak, it has the same vibe like a large model. 🌷🥰