Spaces:
Sleeping
Apply for a GPU community grant: Personal project
I am an independent developer working under the Internectual Labs name. I am building a public interactive demo of TAARDIS, my 27-billion-parameter ternary model, so visitors can explore its responses and provide feedback. The model is already publicly available on Hugging Face. The demo uses the model’s custom llama.cpp runtime.
I am requesting a single GPU, initially a T4 if testing confirms it is suitable, to run inference for this community demo. I plan to limit context length and output tokens and serve one request at a time to keep resource use manageable. The pinned CPU runtime has loaded the public V2 model plus Doctors V3 and generated a basic response locally. The CPU-backed Docker demo is deployed on Hugging Face; its first hosted model download and validation are still in progress. No GPU performance has been measured.
This request is for demo hosting rather than model training. I will document the setup, known limitations, and observed performance, and acknowledge Hugging Face’s support.