kalomaze/Qwen3-16B-A3B
LoRA-style (fewer layers) vs Qwen/Qwen3-30B-A3B. kalomaze/Qwen3-16B-A3B matches part of Qwen/Qwen3-30B-A3B (9603 of its 18867 tensors; the rest of Qwen/Qwen3-30B-A3B has no counterpart). On the shared part: Looks like a merged LoRA (or a fine-tune with frozen norms): only the weight matrices of down_proj, gate_proj, up_proj changed, in 48 of 48 layers; norms, biases and embeddings are untouched.
Tensors changed: Embeddings 0/1, Norms 0/193, MLP 1289/1310, Attention 0/192, lm_head 0/1
7 derivatives checked
- bartowski/kalomaze_Qwen3-16B-A3B-GGUF: Same weights, converted
- Mungert/Qwen3-16B-A3B-GGUF: Quantization
- unsloth/Qwen3-16B-A3B-GGUF: Same weights, converted
- mradermacher/Qwen3-16B-A3B-i1-GGUF: Quantization
- mradermacher/Qwen3-16B-A3B-GGUF: Quantization
- Lucy-in-the-Sky/Qwen3-16B-A3B-Q4_K_M-GGUF: Quantization
- noumenon-labs/Smoothie-Qwen3-16B-A3B: Same weights, repacked
1 other repos hold byte-for-byte identical weight files.