Situus/STARK-4B-Thinking
LoRA-style vs google/gemma-3-4b-it. 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
4 derivatives checked
- mradermacher/Gemma-3-4B-THINKING-i1-GGUF: Quantization (fewer layers)
- mradermacher/STARK-4B-Thinking-i1-GGUF: Quantization (fewer layers)
- mradermacher/Gemma-3-4B-THINKING-GGUF: Same weights, converted (fewer layers)
- mradermacher/STARK-4B-Thinking-GGUF: Same weights, converted (fewer layers)