In a support discussion titled ‘Generating Llava Api Keys is a problem,’ Bathinda seeks guidance on generating API keys for LLaVA to enable AI image captioning on their D-Site, noting that unlike OpenAI or Google Gemini, LLaVA’s Hugging Face interface lacks clear instructions or direct API key generation options. Lilly provides a link to Hugging Face’s API token guide, which Bathinda finds partially helpful — while the token works for authentication, image captioning still returns ‘Error 500’ when using LLaVA, unlike when using GPT-4V. Confusion arises from inconsistent naming (LLaVA vs. Hugging Face) and multiple empty configuration fields in D-Site settings. Lilly points to a meta.discourse.org thread on self-hosting open-source LLMs. Falco clarifies that LLaVA is currently only supported via self-hosting using a specific Docker container (ghcr.io/xfalcox/llava:latest), requiring a GPU with at least 24GB VRAM — otherwise, GPT-4V is recommended. The discussion highlights usability gaps in LLaVA’s documentation and configuration for non-self-hosted use, contrasting with more plug-and-play alternatives like OpenAI or Gemini.
For generating Ai Image Captioning on our ‘D’ sites, either we can use Open_AI’s paid plans (you’ve to even keep some advance credit balance in your account to even begin), or Free & Open Llava Ai Api.
But to get some guidance on how to generate Llava Api keys, I’m Googling for past 3 hours, but even after watching many videos, I’ve not been able to get any direction in right direction.
Llava interface/website doesn’t seem to have any option to generate the needed Api keys straightway:
I think it’d be very precious for users if there was a just ‘a small link’ which could throw inquisitive users in the right direction. Something similar to this:
But I found that it perhaps only helped me move one step forward. Because whereas for Google Gemini as soon as I filled that key in my D-Site settings, everything depending on Gemini started working perfectly well.
But even after filling this Hugging Face Api secret key (you guided me to) in the Disco-Settings, Image Captioning give ‘Error 500’ (the same image captioning works ok if I choose ‘Open-Ai Gpt4-Vision Preview’ as the Image Captioning Model.
And also, because Llava seems to be different as there are so many empty fields in D-Site Settings named Hugging Face or Llava (why do they use Llava in one place and Hugging Face at other is also adding to the confusion), that I’m sure won’t be redundant.
So, can you point me towards some resource on the internet, which could help me getting the values for all these empty fields in D-Site-Settings, or in implementing this properly.
