CastIronMind/Lissa-1.5B-Base
Text Generation • 1B • Updated • 41 • 1
A 1.45B-parameter hybrid linear-attention model trained from scratch on a fandom corpus. Base, instruct, and SFT datasets.
Note Base checkpoint: 1.45B parameters, 30B tokens, hybrid KDA + GQA
Note Chat checkpoint: supervised on 59,182 examples for dialogue and roleplay