We introduce a complete pipeline from finetune to model optimization, finally deploying to the Cloud with a text-to-image Stable Diffusion.
Authors:
- Khoi Nguyen: zero-nnkn
- Dang Quang: QuangNguyen2609
Advisors:
- Ba Ngoc: bangoc123
1. Finetuning
Currently, we investigate methods for finetuning Stable Diffusion with 🤗 diffuers.
Due to hardware limitations, our finetune model cannot achieve the best results and is only to demonstrate the feasibility of these methods.
2. Optimization
We have researched and applied the optimization methods from 🤗 diffuers and 🤗 optimum.
3. Deployment
We provided instructions to deploy Stable Diffusion pipeline on Google Cloud. Note that we are currently only deploying on CPU. Work with GPU is plan for future works.
- Build UI for Text-to-Image
- Complete deployment with Triton Inference Server
- Investigate and write reports for optimization methods in GPU
- Evaluation of model quality