Can I Run Inference Workloads for Models Other than Large Language Models on Vultr Serverless Inference?

At present, Vultr Serverless Inference is purpose-built for serving large language models (LLMs). The service is tuned for transformer-based architectures that require high GPU memory, optimized token streaming, and inference container environments specifically aligned with text generation. The following large language models are available:

• mistral-nemo-instruct-240
• qwq-32b-awq
• deepseek-r1-distill-qwen-32b
• qwen2.5-32b-instruct
• qwen2.5-coder-32b-instruct
• hermes-3-llama-3.1-70b-fp8
• llama-3.1-70b-instruct-fp8
• llama-3.3-70b-instruct-fp8
• deepseek-r1-distill-llama-70b
• kimi-k2-instruct

In addition to LLMs, Vultr Serverless Inference supports text-to-image model such as flux.1-dev, and text-to-speech models, enabling you to convert text into natural-sounding speech:

• bark
• bark-small
• xtts

Models that require different runtime, scheduling, or memory configurations are not currently supported. Workloads outside the supported text and image-generation profiles can be run on general-purpose GPU instances in Vultr Compute, but they remain outside the scope of Vultr Serverless Inference today.