Cross-Architecture Model Merging
Merge any LLM, any size
xmerge uses representation-level bridging to combine models across different architectures and sizes — no weight interpolation required.
Cross-Architecture
Merge LLaMA with Mistral, Qwen with Gemma. Architecture doesn't matter.
Representation-Level
No weight interpolation. xmerge learns a bridge between model representations.
Streaming Engine
Processes one layer at a time — merge 7B+ models on consumer hardware.