用于 LLM、VLM、VLA 的 ROS2 节点
ROS2 nodes for LLM, VLM, VLA
NanoLLM optimizes many different models to run on the NVIDIA Jetson Orin. This project provides a ROS 2 package for scene description using NanoLLM.
By default, the ros2_nanollm package is built into the container and installed under /ros2_workspace (which is an environment automatically sourced on container startup). But if you would like to edit this package interactively, you can do so by cloning it to an external workspace on your host device, and then mounting that workspace overlaid on the original location.
…
jetson-containers run $(autotag nano_llm:humble) /
ros2 launch ros2_nanollm camera_input_example.launch.py /
model:= api:= quantization:=
| ROS Parameter | Type | Default | Description |
|---|---|---|---|
| model | string | Efficient-Large-Model/Llama-3-VILA1.5-8B | The model to use |
| api | string | mlc | The model backend to use |
| quantization | string | q4f16_ft | Quantization method to use |
| ROS Topic | Interface | Description |
|---|---|---|
| input_image | sensor_msgs/Image | The image on which analysis is to be performed |
| input_query | std_msgs/String | A prompt for the model to generate a corresponding response |
| ROS Topic | Interface | Description |
|---|---|---|
| output_msg | std_msgs/String | The output message summarizing the model's conclusions from the inputs |
暂无开放 Issues，或尚未同步最近议题。