stable-diffusion-deploy
stable-diffusion-deploy copied to clipboard
Learn to serve Stable Diffusion models on cloud infrastructure at scale. This Lightning App shows load-balancing, orchestrating, pre-provisioning, dynamic batching, GPU-inference, micro-services worki...
# Before submitting - [x] Was this discussed/approved via a Github issue? (no need for typos and docs improvements) - [X] Did you read the [contributor guideline](https://github.com/Lightning-AI/pytorch-lightning/blob/main/.github/CONTRIBUTING.md), Pull Request section?...
> i pinged the design team to give us the latest design, its very unlikely that we'll be using this. _Originally posted by @pritamsoni-hsr in https://github.com/Lightning-AI/stable-diffusion-deploy/issues/174#issuecomment-1296762731_
Current monitoring integrations are - Sentry and a Prometheus metric exporter. We should refactor the integrations into a module and enable more integrations like NewRelic and others in this space....
**Scenario 1 (verify if we can go up down and up)**: Upscale -> Downscale -> Upscale
The load balancer can be reused for similar instances where we need multiple works or automatic batching. We can convert this into a separate component and add it to Gallery....
Trace: ``` [root.serve_work_0] 2022-09-26T06:11:03.859Z INFO: loading model... [root.serve_work_0] 2022-09-26T06:11:03.899Z INFO: Downloading weights... [root.serve_work_0] 2022-09-26T06:11:49.966Z {'trained_betas'} was not found in config. Values will be initialized to default values. [root.serve_work_0] 2022-09-26T06:12:07.920Z /home/zeus/.local/lib/python3.8/site-packages/torch/cuda/__init__.py:146:...
1. We need to register new works when traffic increases, but every time we spawn a new work, it takes a lot of time to provision, and most of the...