Vikhrmodels/Vikhr-Qwen-2.5-1.5B-Instruct
LoRA-style vs Qwen/Qwen2.5-1.5B-Instruct. 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 28 of 28 layers; norms, biases and embeddings are untouched.
Tensors changed: Norms 0/57, MLP 84/84, Attention 112/196
5 derivatives checked
- Vikhrmodels/Vikhr-Qwen-2.5-1.5B-Instruct-GGUF: Quantization
- mradermacher/Vikhr-Qwen-2.5-1.5B-Instruct-i1-GGUF: Quantization
- mradermacher/Vikhr-Qwen-2.5-1.5B-Instruct-GGUF: Same weights, converted
- itlwas/Vikhr-Qwen-2.5-1.5B-Instruct-Q4_K_M-GGUF: Quantization
- Vikhrmodels/QVikhr-2.5-1.5B-Instruct-SMPO: LoRA-style