Neura-Tech-AI/Nexa-AI-4B-Instruct
LoRA-style vs Qwen/Qwen3-4B-Instruct-2507. 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 36 of 36 layers; norms, biases and embeddings are untouched.
Tensors changed: Embeddings 0/1, Norms 0/145, MLP 108/108, Attention 144/144
3 derivatives checked
- mradermacher/Nexa-AI-4B-Instruct-i1-GGUF: Quantization
- mradermacher/Nexa-AI-4B-Instruct-GGUF: Same weights, converted
- Neura-Tech-AI/Nexa-AI-4x4B-Instruct: MoE built from dense models
1 other repos hold byte-for-byte identical weight files.