basetenlabs / truss

The simplest way to serve AI/ML models in production
https://truss.baseten.co
MIT License
857 stars 61 forks source link

update trtllm template so ModelInput is forward compatible #1010

Closed rcano-baseten closed 3 weeks ago

rcano-baseten commented 3 weeks ago

:rocket: What

Modifies the existing template for trt-llm to enable forward compatibility from the bridge. Additionally, enables non-streaming responses from the truss with temporary hack on "stream" value in the TRT LLM integration

:computer: How

:microscope: Testing

Tested the instantiation with my own truss, sent it additional fields, got back 200

stream value behavior has been validated in previous work

vshulman commented 3 weeks ago

left some nits, lgtm

https://twitter.com/jeffboek/status/1701645550683316504