Fine-tuning performance on Indian datasets / AIRAWAT Compute
#10
by
ProgramerSalar - opened
Really impressed with the motion quality here, but we're noticing it struggles with consistent character retention for longer sequences (specifically for Indian/South Asian contexts).
We are preparing a training run on the AIRAWAT Supercomputer to fine-tune this for an education engine ('Imagination Engine'). We have the compute and the dataset, but we're looking for a lead engineer/collaborator who has experience optimizing the data pipeline for video generation.
If you want to run experiments on this scale (and solve the literacy gap while doing it), check my profile or DM. We're building in New Delhi, India