phi3small¶
Model type |
|
Class |
|
Task |
|
Source |
|
Description¶
Phi3-Small model with block-sparse attention.
Uses block-sparse attention with alternating dense/sparse layers, MuP (maximal update parameterization) scaling, and GeGELU activation with clamping.
Replicates HuggingFace’s Phi3SmallForCausalLM.
Usage¶
mobius build --model <MODEL_ID> output_dir/
from mobius import build
model = build("<MODEL_ID>")