danilopeixoto/pandora-7b-chat
LoRA-style (fewer layers) vs google/gemma-7b. danilopeixoto/pandora-7b-chat matches part of google/gemma-7b (60 of its 254 tensors; the rest of google/gemma-7b has no counterpart). On the shared part: Looks like a merged LoRA (or a fine-tune with frozen norms): only the weight matrices of q_proj, v_proj changed, in 7 of 7 layers; norms, biases and embeddings are untouched.
Tensors changed: Norms 0/12, MLP 0/20, Attention 14/28
2 derivatives checked
- mradermacher/pandora-7b-chat-i1-GGUF: Quantization (fewer layers)
- mradermacher/pandora-7b-chat-GGUF: Same weights, converted (fewer layers)