Release Notes

Release 1.0.0

Summary

NVIDIA NIM for VLMs provides image-based visual assistants, and better world knowledge. This version of NIM for VLMs was trained on only commercial images for all three stages of training and supports single image inference. AWQ quantization along with TRT-LLM optimization enabled to improve throughput, concurrency and TTFT latency.

NIM for VLMs Visual Language Models

• NIM for VLMs VILA-1.5-35b, with text and single image visual assistant. For more details, see VILA NIM Container Overview and VILA NIM Preview API.

Known Issues

VILA does not currently support non-optimized configurations. Attempting to deploy on GPUs not listed in the ‘Support Matrix’ section will fail.

Using --gpus all on A100 PCIEe and H100 PCIe may show out-of-memory errors. Limit the set of visible GPUs with --gpus='"device=0"' .