BlackUnicornSec/Basileak
LoRA-style vs tiiuae/falcon-7b. Looks like a merged LoRA (or a fine-tune with frozen norms): only the weight matrices of dense, dense_4h_to_h, dense_h_to_4h, query_key_value changed, in 32 of 32 layers; norms, biases and embeddings are untouched.
Tensors changed: Norms 0/66, MLP 64/64, Attention 64/64, Embeddings 0/1
1 derivatives checked
- mradermacher/Basileak-GGUF: Same weights, converted