Refactor so that environment and engine

google / jetstream-pytorch

PyTorch/XLA integration with JetStream (https://github.com/google/JetStream) for LLM inference"

Apache License 2.0

32 stars 14 forks source link

Refactor so that environment and engine #65

Closed qihqi closed 4 months ago

qihqi commented 4 months ago

so that they dont depend on llama specific stuff such as ModelArgs

qihqi commented 4 months ago

For 'tokenize_and_pad' error, jetstream removed tokenize_and_pad in recent change, I added it back in PR google/JetStream#70, it should pass after re-run the test.

I added a command in install_everything.sh to pin v0.2.0 for jetstream. We can manually update the hashes and do fixups.