LocateAnything 3B GGUF Model creator: NVIDIA Original model: nvidia/LocateAnything 3B GGUF conversions of NVIDIA's LocateAnything 3B (MoonViT SO 400M vision tower + Eagle MLP + Qwen2.5 3B) for llama.cpp and its multimodal ( mtmd ) stack. The model is a vision language grounding model: given an image and a natural language query, it emits a label plus normalized (0–1000) bounding box or point coordinates. License: NVIDIA License — non commercial / research use only (commercial use is reserved to NVIDIA and its affiliates). A complete copy of the upstream LICENSE is bundled in this repo and must be retained on redistribution, along with attribution to NVIDIA. See the original model card below. About GGUF GGUF is the model file format used by llama.cpp. Each conversion has two halves and both are required : the LLM half ( LocateAnything 3B {quant}.gguf ), and the mmproj vision projector ( mmproj LocateAnything 3B BF16.gguf ). You pair one LLM quant with the mmproj. Provided files File Quant Size Accuracy vs BF16 Notes : LocateAnything 3B BF16.gguf BF16 6.4 GB baseline reference LocateAnything 3B Q8 0.gguf Q8 0 3.6 GB ≤0.5 norm units minimal quant noise LocateAnything 3B Q6 K.gguf Q6 K…
We use cookies for essential functionality and analytics. You can accept or reject analytics cookies.Cookie policy