Hello 👋!
I'm Abhinay, currently a Software Developer based out of India. I mainly work on low latency data serving over Python API's at my job while also building AI/ML POC's and Demos. In my free time I love exploring State-of-The-Art research in NLP and Computer Vision.
I've previosuly worked on enterprise iOS App Development using Swift. I am also a Google Summer of Code, 2019 alumnus from Project PANOPTES where I built web demos and a CLI tool for timelapse generation from FITS files.
I'm excited about AI/ML as a potential tool to accelerate human progress and open source as a major key to that. To that end, I regularly contribute to huggingface/diffusers and so far I've added:
- Checkpoint Merger Pipeline: A community pipeline for merging diffusion model checkpoints that can be loaded by the DiffusionPipeline class.
- UnCLIPTextInterpolationPipeline: A community pipeline to interpolate between the CLIP text embeddings of two text prompts and generate images from the intermediate embeddings.
- UnCLIPImageInterpolationPipeline: Following up on the above, a pipeline to interpolate between the CLIP image embeddings of two input image prompts and generate images from the intermediate embeddings.
Pending for final review and merge:
- Tune-A-Video Pipeline: A one shot conditioned text-to-video pipeline that was proposed by Jay Zhangjie Wu et al. Ported the model to huggingface with the PR pending merge approval.
The following works are in draft:
- UniControlNet Pipeline: A unified controlnet pipeline that leverages a Mixture-of-Experts architecture and modulated zero conv layers to adapt controlnet conditioning using the same controlnet weights. Proposed by Can Qin et al
and many more to come.