IFM/LoopedLM-P2-distilled-m-teacher-init
Full fine-tune (fewer layers) vs IFM/LoopedLM-P2-huginn-m-learned-entropy0p01. IFM/LoopedLM-P2-distilled-m-teacher-init matches part of IFM/LoopedLM-P2-huginn-m-learned-entropy0p01 (18 of its 42 tensors; the rest of IFM/LoopedLM-P2-huginn-m-learned-entropy0p01 has no counterpart). On the shared part: Fine-tune: 100% of the weights changed, across 2 of 2 layers.
Tensors changed: Norms 4/4, Attention 8/8, MLP 6/6