CastIronMind/Lissa-1.5B-Instruct
Text Generation • 1B • Updated • 156 • 1
1.45B hybrid linear-attention model trained from scratch for $555 on one GPU. Base, instruct and SFT datasets.
Note Chat checkpoint: supervised on 59,182 examples for dialogue and roleplay
Note Base checkpoint: 1.45B parameters, 30B tokens, hybrid KDA + GQA