Nemotron 3 Nano Omni : https://arxiv.org/abs/2604.24954
https://medium.com/@harshit158/softmax-temperature-5492e4007f71
https://arxiv.org/pdf/2201.11903
DPO : https://arxiv.org/abs/2305.18290
https://arxiv.org/abs/2210.03629
C-RADIO for Text ↔ Image : https://arxiv.org/abs/2601.17237
https://maartengr.github.io/BERTopic/index.html
https://build.nvidia.com/station/topic-modeling
pair-code.github.io/understanding-umap/
TTD Agent that automates BERTopic
Create a RE system to tutor based on the documents
use nemotron instead of Lama2
create a skill based on teh notebook,
The task is to create topic labels as mentioned on slide 5
Optionally, you can follow the instructions on the llama 2 notebook to call an LLM on the topic clusters after they have been generated and labeled. We recommend you to try Nemotron 3 Nano Omni as the LLM there.
Challenge: replace Llama 2 with Nemotron 3 Nano Omni https://colab.research.google.com/drive/1QCERSMUjqGetGGujdrvv_6_EeoIcd_9M?usp=sharing#scrollTo=w1ufudzoUZzH
Challenge: compute Bertopic on images from this dataset https://www.kaggle.com/datasets/andandand/unsplash-25k-photos-and-embeddings (these have precomputed CLIP embeddings)
Challenge: perform topic modelling on documents from this dataset https://www.kaggle.com/datasets/awester/arxiv-embeddings
Challenge: update the Concept notebook to run without errors using RAPIDS in Google Colab https://colab.research.google.com/drive/1XHwQPT2itZXu1HayvGoj60-xAXxg9mqe?usp=sharing
Condition for all challenges: use either a) RAPIDS for dim reduction and clustering b) Nemotron 3 Nano Omni as substitute for LLama 2 c) Extra credit for doing both 25 USD in compute credits for doing one challenge, 40 USD for completing both. Please ping via DM in LinkedIn when you complete the challenge, with a link to your Github repository (required) https://www.linkedin.com/in/antonioruedatoicen/