MAXENITH-40B-LoRA-v2
LoRA adapter for a block-expanded Maxenith/MAXENITH-32B (64 -> 80 layers: a copy of every 4th layer is inserted right after it, with o_proj/down_proj zeroed). LoRA trains only the 16 new layers (indices 4, 9, ..., 79). Data: SmolKalam (Arabic), OpenThoughts-114k, OpenCodeReasoning, NuminaMath-CoT, MetaMathQA, Magicoder.
- Downloads last month
- -
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support
Model tree for Maxenith/MAXENITH-40B-LoRA-v2
Base model
Maxenith/MAXENITH-32B