llama.cpp

Fork of mainline llama.cpp. This fork adds multimodal support for Qwen3-VL (and possibly other) models:

• Multimodal /v1/embeddings (image + text input)
• Multimodal /v1/rerank (image + text input)
• Full support for Qwen3-VL causal reranker and embedding models