We hope you’re having a great time participating in the FlowerTune LLM leaderboard!
We’ve identified a minor issue with the dataset formatting process, which has now been resolved. Here’s a quick guide to ensure everything runs smoothly:
-
For new projects, please update your
flwr
version by runningpip install git+
https://github.com/adap/flower.git@d92453d
(untilflwr==1.13.0
is out) and then runflwr new --framework=FlowerTune
to create your project. -
If you already created a project with
flwr 1.12.0
, please manually change the second last line in functionload_data()
indataset.py
:
client_trainset = reformat(client_trainset, llm_task="generalnlp") # set llm_task correctly
. Update the line sollm_task
is one ofgeneralnlp, medical, finance, code
(depending on the challenge you chose).
We apologise for any inconvenience this may have caused and are excited to see your upcoming submissions!