i3-lab/i3-12m

Text Generation • 12.7M • Updated Oct 23, 2025 • 72 • 3

Note Our first usable i3 model (meaning that we added Transformers support and some code for it)

i3-lab/i3-22m

Text Generation • 22.6M • Updated Oct 31, 2025 • 21 • 2

Note Smol stable text generator that took over 14 hours to pre-train :) --- Changes --- Trained on over 1T tokens LoRPt layers

i3-lab/i3-80m

Text Generation • 82.8M • Updated Nov 30, 2025 • 75 • 7

Note Previous SOTA model. Pre-trained in around 2 to 4 hours, in comparison with the previous version of over 14 hours. --- Changes --- Trained on over 3T tokens Other stuff available to read in the model card.

i3-lab/i3-200m-v2

Text Generation • 0.2B • Updated Dec 3, 2025 • 21 • 3

Note Our newest SOTA model. Having ~200m parameters. Trained on the same dataset as the previous model + wikitext. The training took ~1-2 hours. --- Training --- Loss: ~1 PPL: ~3