llama.cpp

e11b2e6e - Qwen2 : assume tied weights if lm_head/output weights is missing (#6738)