manthilaffs/Gamunu-4B-Instruct-Alpha
LoRA-style vs google/gemma-3-4b-pt. 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 34 of 34 layers; norms, biases and embeddings are untouched.
Tensors changed: Embeddings 0/4, Norms 0/316, MLP 102/210, Attention 136/352, other 0/1
2 derivatives checked
- mradermacher/Gamunu-4B-Instruct-Alpha-i1-GGUF: Quantization (fewer layers)
- mradermacher/Gamunu-4B-Instruct-Alpha-GGUF: Same weights, converted (fewer layers)