LocateAnything
LocateAnything is a vision-language model that grounds free-form descriptions in an image. BOAT ships it as GGUF quantizations for Metal, from a 3 GB Q4 build up to full BF16 weights. Its weights are licensed for non-commercial research use.
Install
boat model install locate-anythingWithout a tag BOAT installs locate-anything:3b-Q4_K_M. Tags install side by side, so you can keep several.Download: 3 GB of weights.
You will be asked to accept the model's license on first install. Add --accept-license to accept it non-interactively. Read the terms
Tools
What LocateAnything adds to the annotation editor. Each example loops what you do and what comes back.
Text to Boxes
Text promptBounding boxes
Type a class name such as “car” and press Enter. BOAT previews every matching bounding box before you apply them.
- Pick candidates
Text to Points
Text promptPoints
Type a class name such as “bird” and press Enter. BOAT previews every matching points before you apply them.
- Preview, then apply
Platforms
BOAT picks the runtime profile that matches your machine. Only that profile's code is downloaded.
macOS · Metal
Apple silicon (MPS)
Linux · CUDA 12.8
NVIDIA, incl. Blackwell
Linux · CUDA 12.6
NVIDIA GPUs
Linux · CPU
x86_64 CPU
License
BOAT's adapter code is Apache-2.0. The model's own code and weights keep their upstream license:Custom.
License terms shown at install
LocateAnything is provided using NVIDIA’s LocateAnything-3B model under the NVIDIA License.
By continuing, you acknowledge and agree that:
- LocateAnything may be used or intended for use only non-commercially, for research or evaluation purposes.
- Commercial use is not permitted, except by NVIDIA Corporation and its affiliates.
- If you redistribute the model or derivative works, you must do so under the NVIDIA License, include a complete copy of the license, and retain all copyright, patent, trademark, and attribution notices.
- Any derivative works must preserve the same non-commercial use limitation.
- The model is provided “as is” without warranties, and you assume all risks associated with its use.
- Third-party components included with the model may be subject to additional licenses.
Bundled license files: BOAT-MIT.txt, LLAMA-CPP-MIT.txt, NOTICE.md, NVIDIA-LICENSE.txt, llama-cpp-third-party/LICENSE-jsonhpp, llama-cpp-third-party/cpp-httplib-LICENSE, llama-cpp-third-party/hash-rotate-bits-LICENSE.md, llama-cpp-third-party/hash-sha256-LICENSE, llama-cpp-third-party/hash-xxhash-LICENSE, llama-cpp-third-party/miniaudio-LICENSE.txt, llama-cpp-third-party/sheredom-LICENSE.txt, llama-cpp-third-party/stb-LICENSE.txt
Package details
- Package
- locate-anything
- Version
- 0.0.1
- Type
- Common model
- BOAT Core
- ≥ 0.1.0 · < 1.0.0
- Idle unload
- 300s (change with
boat model setup) - Inference timeout
- 600s
- Manifest SHA-256
- 3a5c6cb9ac01fc88d935e2395cde244dea0dc3adbc9bd1106f092c646af421f7
Upstream notice
# LocateAnything 0.0.1 - Model: NVIDIA LocateAnything-3B, converted to GGUF by yuuko-eth. https://huggingface.co/yuuko-eth/LocateAnything-3B-GGUF Revision: c3260717380c24cd2fc11f66c43e2feeec3e32fe. NVIDIA's non-commercial research license is in NVIDIA-LICENSE.txt. Retain all upstream notices when redistributing weights. - llama.cpp / mtmd-grounders: https://github.com/yuuko-eth/llama.cpp Revision: 8c1921616abbbbac17493c0d60ae2a73edf7f761. MIT license in LLAMA-CPP-MIT.txt; bundled dependency notices in llama-cpp-third-party/. The bridge and build tooling are provided under MIT. - Pillow is installed from its hash-pinned wheel, including its own license. - Native sources are obtained only by the developer build script from the pinned revision. User installation uses the included native wheel offline after downloading the declared Python dependency and model weights.