Lambent/Goose-7.2B-G1j-20260928
LoRA-style vs Lambent/Goose-7.2B-G1j-20260914. Looks like a merged LoRA (or a fine-tune with frozen norms): only the weight matrices of k_proj, key, o_proj, r_proj, v_proj, value changed, in 32 of 32 layers; norms, biases and embeddings are untouched.
Tensors changed: lm_head 0/1, Embeddings 0/1, Attention 128/765, Norms 0/196, MLP 64/96
1 derivatives checked
- mradermacher/Goose-7.2B-G1j-20260928-GGUF: Not comparable yet