map-charting-student-math-misunderstandings

[frozen nacelle]

good luck to all🫡

[pale moat]

subscribe this channel and see the latest update and learning video from here :https://www.youtube.com/@AuraaiX/videos

[woven helm]

I have submitted a notebook but its been running since 35 mins lol
Is this normal ?

[cloud sky]

hello Kaggle community. i am looking for 1-2 experienced team mates for this competition. if you are intreseted in this then dm me or mention here thanks.

[analog drift]

As a starter/beginner in ML (but not in SWE) I'm using this competition mostly to learn. Curious if anyone is in the same boat?

[loud rose]

While working on the task involving Misconception classification, I was trying to understand how the predictions for the Misconception column are being evaluated. I have some questions.
Is this list exhaustive, or are there additional labels we might expect at test time?
How is the evaluation handled if the predicted label is outside this observed set?

[timber pebble]

me, but ~2 months ago in the cmi-detect-behavior-with-sensor-data comp. it's been a blast so far

[cloud sky]

do we have to concat 2 column category and misconcetion after prediction such as multilabel prediction or concat first then predict one column which were

[woven helm]

Yeah i think you have to concat @cloud sky

[steady hinge]

Hello, this is my first time participating in a Kaggle competition. We can say I'm still kind of a beginner in this field.
I'd appreciate any advice for this topic.
And I'd like to take your insights, is it good for someone to build knowledge in the ML field, or is it hard?
Thank you in advance, and good lukc for you all!!

[analog drift]

I'm asking this question myself as well. I'm just trying to take it as a learning opportunity in the first place and focus less on the final outcome score.

[brazen loom]

Hey, I'm participating in this competition to learn too. I'm starting from today and excited to learn.

[loud rose]

does anyone know how to fine-tune large models (>9B params) using LoRA/QLoRA?

[analog drift]

Do you need to do that for this competiton? I'm just curious and would love to understand why..

[loud rose]

I don't know man
I was just researching ways to fine-tune large models in a limited GPU environment like Kaggle

Maybe bigger models will give a better score

[crystal grove]

fine tuning models with 9b parameters is giving me OOM errors with 40 VRAM GPU

i tried the whole day yesterday...now checking how models with 2b parameters will pan out for me

[loud rose]

You can check out the discussion post by Chris Deotte. It is using the Gemma-9b model and training it with LoRA and getting CV 0.945

[brave sable]

My scoring is going on since 7 hours any idea ?

[coral schooner]

Hi,
I try to load gemma-2-9b-it-bnb-4bit model, i am getting the issues with bitsandbytes.
Eventhough i uninstalled and installed bitsandbytes again, the problem repeats.
I am using the gemma-model from kaggle models --> unsloth/gemma-2-9b-it-bnb-4bit

When i try to install !pip install -U bitsandbytes in the first cell then try with below code, it works fine
model_base = "/kaggle/input/unslothgemma-2-9b-it-bnb-4bit/transformers/default/1"

code

model = Gemma2ForSequenceClassification.from_pretrained(
    model_base,
    num_labels=65,
    torch_dtype=torch.bfloat16,  # Use bfloat16 for speed
    device_map="auto",
    trust_remote_code=True,
)

error

ImportError                               Traceback (most recent call last)
ImportError: Using `bitsandbytes` 4-bit quantization requires the latest version of bitsandbytes: `pip install -U bitsandbytes

[crystal grove]

hey folks...has any one made submission yet?? just curious to know, how much time is takes to complete the model assessment...given that submission.csv file was genereated??

[crystal grove]

yeah this is an issue with the environment...we need to upgrade the bitsandbytes package

[crystal grove]

did it finally succeed or did the scoring failed??

[heavy nexus]

Hi, I was just wondering how does explanation being correct for a False answer ? Like are we saying even though the answer they reached to was wrong but the process was correct ?

[winter quail]

normal

[steel elbow]

Can we train models off kaggle and then inference for submission?

[jade copper]

yes. If not exceeding the limitation on kaggle hard drive. And make sure you don’t exceed the limitation of inference time in submission when you have large models.

[warm badge]

Why I am getting this error when using Keggle model:

ValueError: The checkpoint you are trying to load has model type qwen3_next but Transformers does not recognize this architecture. This could be because of an issue with the checkpoint, or because your version of Transformers is out of date.

You can update Transformers with the command pip install --upgrade transformers. If this does not work, and the checkpoint is very new, then there may not be a release version that supports this model yet. In this case, you can get the most up-to-date code by installing Transformers from source with the command pip install git+https://github.com/huggingface/transformers.git

[steel elbow]

You need to upgrade transformers

[warm badge]

Thanks, I did this but the problem was different. Some libraries were conflicting.