ethicalabs/xLSTM-7b-Instruct
LoRA-style vs NX-AI/xLSTM-7b. Looks like a merged LoRA (or a fine-tune with frozen norms): only the weight matrices of proj_down, proj_up changed, in 32 of 32 layers; norms, biases and embeddings are untouched.
Tensors changed: MLP 64/96, other 0/288, Norms 0/97
1 derivatives checked
- ethicalabs/xLSTM-7b-Polymath: LoRA-style