llama.cpp

Qwen2: assume tied weights if lm_head/output weights is missing

#6738

Merged