#wsdm-cup-multilingual-chatbot-arena

1 messages · Page 1 of 1 (latest)

supple star
#

Looking for teammates to Collab with. I have already worked on LMSYS and have a few ideas in mind

copper venture
#

alrdy pm u
@supple star

hasty lake
#

hey @everyone, i am looking for a team to join for the competition. i have experience in cnn models and ai training. i am so ready to work and learn for this competition so hit me up. ciao!

copper venture
copper venture
#

@supple star

inner spear
#

if anyone want to discuss ideas, collab, or anything really, hmu

young oar
#

how to fix that?

image.png
vocal hinge
sharp robin
#

Hi @everyone!
I’m looking to form a team for the WSDM Cup - Multilingual Chatbot Arena Challenge and would love to collaborate with 2-3 motivated individuals. If you're interested in joining forces and taking on this exciting challenge together, feel free to reach out to me via message! Thanks !

brazen hearth
#

Hello! I'm trying to use transformers.Trainer to fine tune gemma model. But I ran into this error:
"ValueError: Attempting to unscale FP16 gradients."
Whether I set fp16=True or not in training_args, it always gives this error. If I don't run trainer and simply use the model to generate an output, it is fine. The error is probably happen somewhere in the trainer. Does anyone has clue how to debug or fix it? Much appreciated!

#

Btw, I don't have GPU myself - has anyone done the model fine-tuning on kaggle and get good scores? I'm considering colab or aws if kaggle is not enough

#

(And if anyone is interested in teaming up, just DM me)

silent plaza
jagged basin
#

Hello everyone, how are your guys training llms?

#

Everytime I am trying, I am keeping getting OOMs

#

🥹 This is really frustrating you know. Seeing another guy using the same configuration. but able to train the model

brazen hearth
jagged basin
#

max_length = 3072
4bit quantized gemma
with a batch size of 2

#

And yeah thanks, I thought no will reply😅

#

feels good when you are heard

jagged basin