File size: 592 Bytes
8a97a2c | 1 2 3 4 5 6 7 8 9 10 11 12 | Loop-distil recurrent checkpoint exports
Derived from tomg-group-umd/huginn-0125, released under Apache License 2.0.
Base revision: bb6621b65e90b6a4b9b29ef88dc83866d450470c
Base project: https://huggingface.co/tomg-group-umd/huginn-0125
Base research: https://arxiv.org/abs/2502.05171
Modifications: trajectory-regression initialization and optional KL distillation
of the shared adapter/recurrent core, published by Nicholas0228/loop-distil.
Frozen base weights are obtained separately from the pinned original model.
Research code and records: https://github.com/Nicholas0228/loop-distil
|