ThakiCloud/K-MT-7B-v0.1
LoRA-style vs tencent/Hy-MT2-7B. Looks like a merged LoRA (or a fine-tune with frozen norms): only the weight matrices of k_proj, o_proj, q_proj, v_proj changed, in 32 of 32 layers; norms, biases and embeddings are untouched.
Tensors changed: Norms 0/129, MLP 0/96, Attention 128/128, Embeddings 0/1
1 derivatives checked
- mradermacher/K-MT-7B-v0.1-GGUF: Same weights, converted