Frequently Asked Questions (FAQ)

July 26, 2026 ยท View on GitHub

This document addresses common questions regarding hardware requirements, software environment, and runtime compatibility for ABot-World.

Hardware Compatibility

Can ABot-World run on NVIDIA GeForce RTX 3090 / 4090?

Peak GPU memory usage is approximately 19 GB, with at least 64 GB of system RAM recommended. Under these constraints, GPUs such as the RTX 3090 (24 GB) and RTX 4090 (24 GB) are in principle capable of running the released causal student model.

We have verified inference on RTX 4090. RTX 3090 has not yet been evaluated on our side, though we expect it to be feasible given the memory footprint. Please ensure a CUDA environment consistent with the Setup instructions.

Is multi-GPU inference supported?

Not at present. The current release targets single-GPU desktop inference. Multi-GPU parallelism could, in principle, improve throughput, but has not been systematically validated. We recommend single-GPU deployment until multi-GPU support is officially released.

Software Environment

Which CUDA versions are supported?

We have tested CUDA 12.8 and CUDA 12.9 (cu128 / cu129). Other CUDA versions have not undergone full verification. When in doubt, please align with the environment described in the Setup section.

We strongly recommend Linux. Support for certain components under Windows is limited or fragile. For reproducible installation and inference, please prefer Linux (e.g., Ubuntu 22.04 as in our reference setup).

Gradio Demo

Which browser should I use for the Gradio interface?

We recommend Google Chrome for online Gradio usage, where we have observed the most consistent behavior.