A highly optimized, 15-layer spliced variant of the Qwen 2.5 0.5B architecture.
Model Description
This repository contains the spliced mixed-precision weights of Qwen 2.5 0.5B, surgically pruned from 16 layers to 15 layers. It features standard configs and tokenizer support for seamless local execution on mobile and edge platforms.
Format: Safetensors / GGUF (Q4_K_M, IQ4_XS)
Size: 284 MB (GGUF) / 257 MB (IQ4_XS)
Target Platforms: Apple Silicon MacBooks (M1/M2/M3/M4) and standard CPU/GPU local runtimes.
Local Quickstart
Run the model natively via standard llama-cli:
llama-cli -ngl 100 -m ./gemma_spliced_fresh_q4km.gguf -p "The mathematical beauty of wavelets lies in" -n 128
