Painless Docker - 2nd Edition

Understanding How to Choose Model Variants

Tags are an essential part of identifying and selecting the right model variant for your needs. In Docker Hub, model tags follow a structured naming convention:

:-

The SmolLM2 family is published in multiple variants that differ in model size, quantization method, memory footprint, and runtime performance.

The number of parameters is the primary factor that defines a model's capacity and baseline capability. This is typically indicated directly in the image tag. In addition to parameter count, each variant applies a different numerical representation (precision or quantization), which significantly affects memory usage and inference speed.

For SmolLM2, the available variants include:

• ai/smollm2:135M-F16
• ai/smollm2:135M-Q2_K
• ai/smollm2:135M-Q4_0
• ai/smollm2:135M-Q4_K_M
• ai/smollm2:360M-F16
• ai/smollm2:360M-Q4_0

• ai/smollm2:360M-Q4_K_M

• The 135M