this model beats even full o1 even in that size,
baidu/ERNIE-4.5-VL-28B-A3B-PT
https://huggingface.co/baidu/ERNIE-4.5-VL-28B-A3B-PT
yes i cant believe but its true, we deserve a gguf NOW
thanks mradermacher❤️
probably me and Umbreon are gonna start uncensoring it tomorrow
oh, and just to you know, i was considering my old model as AGI but whole fuck this model seems another level
Ernie4_5_VLMoeForConditionalGeneration is not currently supported by llama.cpp so good luck running that model without GGUFs. Only Ernie4_5_ForCausalLM is currently supported by llama.cpp.
There now at least is https://github.com/ggml-org/llama.cpp/pull/14658 so we likely soon get llama.cpp support for Ernie 4.5 MoE but not for vision. Damn some more massive models to come: https://huggingface.co/collections/baidu/ernie-45-6861cd4c9be84540645f35c9