add example for TRELLIS

[charlesfrye]

brings in #1018 by @sandeeppatra96 and @patraxo as a misc example

[charlesfrye]

🚀 The docs preview is ready! Check it out here: https://modal-labs-examples--frontend-preview-3dc214a.modal.run

[charlesfrye]

Super cool example!

A few things before we can bring in as a main-line example that'd show up on the docs site:

• The build process looks pretty messy still (multiple pip install steps that conflict on versions of e.g. Triton, a variety of CUDA versions e.g. 11.8, 12.2, & 12.4). And do we need all of those env vars, e.g. the compiler flags?
• Pin versions for the pip_installed libraries
• Put the model into a Volume instead of the Image, make sure no models are downloaded after the first @enter
• Split out the web serving from the underlying function
• Write up the example in the "Literate Programming" style

[sandeeppatra96]

Super cool example!

A few things before we can bring in as a main-line example that'd show up on the docs site:

• The build process looks pretty messy still (multiple pip install steps that conflict on versions of e.g. Triton, a variety of CUDA versions e.g. 11.8, 12.2, & 12.4). And do we need all of those env vars, e.g. the compiler flags?
• Pin versions for the pip_installed libraries
• Put the model into a Volume instead of the Image, make sure no models are downloaded after the first @enter
• Split out the web serving from the underlying function
• Write up the example in the "Literate Programming" style

I had multiple issues as it was unable to find CUDA_HOME, also even though some libaries were being installed without pin versions they were still not working and even caused issues later on while inferencing. e.g. nvdiffrast and diso library

[charlesfrye]

Word! I've tasked our Devin integration with figuring out the install process.