SkyIsNotGreen/Scion-35B-A3B
Text Generation • 35B • Updated • 3.63k • 16
Trained corrections grafted onto low bpw ternary expert bodies.
Note 2.61 bpw | 11.34 GB (6.3x smaller than the BF16 reference) | best PPL of the 2-bit class | task retention at Q4-class level, at about half of Q4_K_M's size.
Note 50MB drafter for Scion 35b. | Needs specific llama.cpp fork to work | Proof-of-concept for cheap drafting using a sidecar.
Note 2.60 bpw | 57.62 GB (3.2x smaller than the FP8 release) | Smallest build of this model while remaining unpruned, last time I checked.