Hi there, I am looking at building a leading edge AI which must fit within a 4GB NVidia GPU and the system only has 8 GB RAM, I would likely be running the local model in WSL2 Ubuntu Linux running on Windows 10 or 11. The model must have basic English, Image Generation and reading, sound detection and generation (speech, sound effects, music). I know this might sound impossible to have a multimodal model like this in a 4GB NVidia GPU, however, I know this is possible as I have a project I am working on AXI // IMMERSION CORE which has lots of that stuff local, speech synthesis, music generation, all from local models that fit in the 8GB RAM on the phone. I would like to know how to get an H100 that I have seen I can access to do the training of the dataset I will need to create this mini-multimodal AI or mini-AXI. Any help would be awesome, I am a beginner in building AI models, I have built some before, but I am really trying to push the limits, most people might say this is impossible, but I need to make it possible please any help welcome.
