bad responses from 70B llama-2 -- TGI server bug?

h2oai / h2ogpt

Private chat with local GPT with document, images, video, etc. 100% private, Apache 2.0. Supports oLLaMa, Mixtral, llama.cpp, and more. Demo: https://gpt.h2o.ai/ https://gpt-docs.h2o.ai/

http://h2o.ai

Apache License 2.0

11.39k stars 1.25k forks source link

bad responses from 70B llama-2 -- TGI server bug? #633

Closed pseudotensor closed 9 months ago

pseudotensor commented 1 year ago

https://h2oai.slack.com/archives/C050PKJ6GAX/p1691513290247279?thread_ts=1691491068.778999&cid=C050PKJ6GAX

pseudotensor commented 9 months ago

not using TGI much anymore