Request: Add google/gemma-4-26b-a4b-it as hosted API on build.nvidia.com (original) (raw)

Hi,

Could you please add google/gemma-4-26b-a4b-it as a hosted API
on build.nvidia.com?

The NIM container already exists on NGC Catalog:

Problems with the current gemma-4-31b-it hosted API:

• Thinking mode cannot be fully disabled — even with thinkingBudget:0,
  it still produces empty tags that break real-time
  applications like subtitle translation
• Also, google/gemma-4-31b-it is unavailable most of the time on the Playground and via the hosted API (integrate.api.nvidia.com/v1/chat/completions)
• The 26B A4B has the same quality (-2/3 points) without
  thinking issues on E2B/E4B models

The 26B A4B is MoE with only 3.8B active parameters —
very efficient to serve. The container is already ready!

Thank you so much,

Regards, Pedro

they allready added the model bro 😅🤷‍♂️