Delta-Vector/Daichi-12B
LoRA-style vs Delta-Vector/Pascal-12B. Looks like a merged LoRA (or a fine-tune with frozen norms): only the weight matrices of down_proj, gate_proj, k_proj, o_proj, q_proj, up_proj, v_proj changed, in 48 of 48 layers; norms, biases and embeddings are untouched.
Tensors changed: Norms 0/400, MLP 144/252, Attention 192/408, other 0/1, Embeddings 0/3
3 derivatives checked
- mradermacher/Daichi-12B-i1-GGUF: Quantization (fewer layers)
- mradermacher/Daichi-12B-GGUF: Quantization (fewer layers)
- grimjim/Daichi-Instructed-12B: Full fine-tune