#ai-mathematical-olympiad-progress-prize-3

1 messages · Page 1 of 1 (latest)

spring hawk
#

Hey guys, good luck in the olympiad!

acoustic niche
#

Hey guys, I am a starter in such competitions. I am looking for a team which can help me grow in such environment. I will give my whole time and dedication required for the best performance of the team. But I lack the necessary skill level to be good at this.
Am trying to understand the pipeline of the whole problem to solution.

Even if it is not possible to join a team, I hope that atleast someone can mentor me in terms of the understanding of the working of the model/s on this data.

Just want to gain a significant amount of knowledge and basic skills through this competition. Will improve in the months to come.

tall leaf
#

Hey guys, quick question - is it possible for our local llm to see all questions at once? or does the inference api restrict this to 1 question at a time?

spring hawk
#

then again, im wondering if the inference api serves from the test file. you might be able to look at all the problems through that. take a look at the inference server code to see where it gets the problems from

tall leaf
harsh wraith
#

https://media.discordapp.net/attachments/1444971360047726605/1445085758598938824/image1.gif?ex=692f107d&is=692dbefd&hm=94f18cd6e7350e7cc612826beb5d11a9fd125485a58ee1e39a16a03b6f9e2426&=&width=237&height=315
https://media.discordapp.net/attachments/1444971360047726605/1445085766937088000/image2.gif?ex=692f107f&is=692dbeff&hm=51e8429e6818b166e21485a613e8f0c706d64c765aefc93f65a7bcefa10907c2&=&width=864&height=1152
https://media.discordapp.net/attachments/1444971360047726605/1445085774562197535/image3.gif?ex=692f1081&is=692dbf01&hm=e520e8e4edd4eea02e82168a7059a868ea59c19d9b90c7c34402f7bb3616c76f&=&width=864&height=1152
https://media.discordapp.net/attachments/1444971360047726605/1445085781801566319/image4.gif?ex=692f1082&is=692dbf02&hm=bdc0715977fdcda4b7804916e5bfb36af1d3132f535d1b4327894a067fbfc769&=&width=725&height=907

small shard
#

||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​|| _ _ _ _ _ _ https://imgur.com/TC6h8P4 https://imgur.com/iiKXKB5 https://imgur.com/JAkE28j https://imgur.com/keASgw9

ruby gust
craggy garnet
#

test

#

Currently #4 here, with plenty of published notebooks and discussions, apparently I am banned from Kaggle

Hi Tong Hui Kang,

Our automated content review system recently found that your content is not compliant with one or more of our policies. See below for more information about your content status and how to correct the issue.

Content:
Notebook: streaming-inference-private (2025-12-07 02:27)
Source of Report: Automated systems
Issue Found: Violates our Community Guidelines and/or Terms against Resource abuse. For further explanation of why the content and/or use of the platform is considered violative on these grounds, please refer to Kaggle’s Community Guidelines.
Result: We have unpublished the content and issued a ban on your account, unless we determine otherwise after an appeal.

eager hound
oak walrus
craggy garnet
# oak walrus Did you "unhook" the h100 notebook from the competition?

I didn't "unhook" the h100 as far as I remember. I am even using external h100 for the competition, there is no ideas reason I need to use Kaggle h100 for other things

I think what can cause the system to suspect is that the former #1's notebook has this logic that makes start vLLM much faster, and this could make the CPU usage very different from other notebooks

oak walrus
eager hound
ruby gust
#

Hi, I'm trying to do this contest but having some trouble getting setup. Could I pay someone to pair with me to set it up? Shouldn't take more than an hour, I can pay $50 US

jovial lynx
craggy garnet
#

Yup

split briar
# craggy garnet Yup

yo just a question did you build your model by requesting for gpus? if yes, how many did they give you?

craggy garnet
#

Request who and what GPUs?

split briar
craggy garnet
#

I don't get special treatment from Kaggle

split briar
#

forget i ever asked

craggy garnet
#

ohh you should cite that discussion then, nope that hasn't opened yet as far as I know

craggy garnet
eager hound
#

anyone trying ai agents?

eager hound
#

We can't use external llm apis right n internet to be turned off?

#

We can use only the loaded models?

eager hound
#

anyone trying functiongemma?

maiden vine
#

Did anyone try to use TRMs or HRMs for this ?

eager hound
brittle linden
#

Hi everyone 👋
I’m Sohaib, a mathematics lecturer with 8 years of teaching experience from Pakistan with an MPhil in Math (research in applied cryptography). I have completed multiple machine learning projects. I want to participate in this competition, but since I am new, I don’t have much idea about how it works. Is there anyone who can guide me? I am looking to join a team. I hope my strong math foundation will be beneficial for the team.

limber briar
#

Hello all, I need 1-2 team mate for 2 competition. Interested individuals should have deep interest and expertise in modern deep learning architecture specially NLP and transformers. Must have access to GPU machine or cloud. Interested individuals fee free to dm me or mention me here.

ornate trench
rapid arch
#

Hi everyone,
I’m looking for a team where I can learn and grow in the competitive environment. I’m fully willing to dedicate the time and effort required to contribute meaningfully to the team. While I may currently lack the advanced skill level needed to perform at a high level, I’m highly motivated to improve and eager to learn.
I have worked on two data science projects so far, one focused on brain tumor prediction and another on Parkinson’s disease prediction, which helped me gain some hands-on experience with data preprocessing, feature engineering, and basic modeling.
I would really appreciate mentorship or guidance from someone experienced, especially on how to go about or join real world/impactful projects. I’m committed to improving consistently over the coming months.

fierce hearth
#

Hi 👋🏼, I was just curious what the average time for runs for your notebook. It’s my second submission and mine has been running for some time now

restive dawn
#

Has anyone worked on this competition since progress prize 1?

tall leaf
#

Hey guys, I'm trying my best to get us a working DeepSeek Math V2 model that runs on a Single H100! Here is a rough teaser of my approach!

Tldr, I managed to take a 345B DeepSeek V3.2 REAP model and shrunk it (75% sparsity) to 86B params! The checkpoint is on kaggle. So far it answers 1+1=2 correctly, so yey! Plus I integrated my cuda kernels with vllm so it's easy to use for y'all!

https://www.kaggle.com/code/pranshubahadur/aimo3-4bit-forge-sparsegptq-1-4-deepseekv3-2-wip

brave lantern
#

Nice work, how was the speed measurement?

tall leaf
queen mango
#

Would anyone be interested in a local self‑scorer to reduce reliance on the 1‑submission‑per‑day limit? It wouldn’t match Kaggle’s hidden test set, but it can provide a useful proxy score and can improve over time with fine‑tuning/verification.

tall leaf
weary bobcat
#

Hi, I have some questions about the tournament rules for ai mathematical olympiad progress prize 3. My general idea is to use the ClaudeAgentSDK API with an open-weight model in it. I have a question:

  • Would using Claude Code inside the jupyter provided kernel fall under ADT use, and means it wouldn't be valid since claude code is not open source? Or would it fall under "commercial available tools", I'm assuming it works since it's more like an API use?
brave lantern
tall leaf
rancid crown
tall leaf
#

Hey I have a question for the hosts - If I compress a model using my 4Bit-Forge repo, does the compressed checkpoint count as an AMLT or an ADT. I ask because I'm planning to release checkpoints by 29th of March, but will have to expedite development if it's AMLT.

brave lantern
rancid crown
obtuse magnet
#

Hey everyone

real belfry
modern linden
tall leaf
modern linden
#

yes, it is sparse. but it weights ~150GB

tall leaf
#

Yeah that's because I chose group size 32 haha but pls be patient I make a proper MVP

heady widget
#

Hey everyone

hearty elk
#

Happy Weekend!

Hello Everyone!
If you know someone who have good skills in Python and Machine Learning, Please invite me!

Our Company is open to hire Python and Software Engineer.

Requirements:
2+ years of Software Engineering Experience
C1 or Native English Level
Good vision of Software Trent

Benefits:
Competitive Income
Supporting Several roles and chances
Multiple Role Working is enable

Important:
Our company is designed for Capability Person.

Questions:
For Junior Persons?
Do not give up, strong enthusiasm is also big point and our company also focus on the person's enthusiasm.

Thanks again.
Sophia

proven light
#

check my bio 😁

small tiger
#

check my bio 😁

dim island
#

is there an issue with h100 gpu? it seems its getting stuck randomly at times.

dim island
vestal knot
proven herald
#

When will the results be out?