🚀 Our latest research paper on code representation learning, CodeSage, outperforms OpenAI text-embedding-3-large on Code2Code search, and is on par with NL2Code search tasks! Dive into the techniques and insights - check them out on the blog: code-representation-learning.github.io

🚀 Dive into the cutting-edge research exploring keyphrase generation! The work delves deep into evaluating keyphrase generation on diversity, utility, faithfulness, and reference alignment.

Introducing RepoFormer! A repository-level code completion framework. The project was led by Di during his internship (summer'23) at AWS AI Labs. Read the paper to learn about the awesome work.

Computer Vision conference's acceptance criteria these days: #CVPR2024 #eccc2024 #AI #ComputerVision

🔥 Introducing IllusionVQA 🔥 A challenging dataset to test VLMs' ability to locate and comprehend optical illusions. While humans achieved near perfect accuracy, GPT4V, the best-performing VLM, achieved 63% and 49.7% accuracy on the comprehension and localization tasks.

🚀 Introducing “Repoformer: Selective Retrieval for Repository-Level Code Completion” accepted at #ICML2024.

My twitter network Please help me to spread this. #SaveBangladeshStudents #SaveBangladeshiStudents #StudentProtest #StarlinkforBangladesh

Introducing LongMemEval: a comprehensive, challenging, and scalable benchmark for testing the long-term memory of chat assistants. 📊 LongMemEval features: • 📝 164 topics • 💡 5 core memory abilities • 🔍 500 manually created questions • ⏳ Freely extensible chat history

Direct RL without SFT improved pass@1 from 61.6 to 84.1. Do you have numbers of SFT using the released 89k data?

Today, we announce a collaboration between SWE Arena (Computer Intelligence) and Hugging Face (w/ Gradio). We believe that Hugging Face can help us shape the future of AI Software Engineering evaluations. We have now open-sourced the SWE Arena codebase to accelerate the development of

Nemotron-H base models (8B/47B/56B): A family of Hybrid Mamba-Transformer LLMs are now available on HuggingFace: huggingface.co/nvidia/Nemotro… huggingface.co/nvidia/Nemotro… huggingface.co/nvidia/Nemotro… Technical Report: arxiv.org/abs/2504.03624 Blog: research.nvidia.com/labs/adlr/nemo…

Nvidia presents Llama-Nemotron: Efficient Reasoning Models - An open family of models w/ exceptional reasoning capabilities and inference efficiency - Discusses the training procedure, incl. NAS from Llama 3 for accelerated inference, knowledge distillation, and continued

Consider submitting your work at DL4C.

Please consider contributing to the workshop as a reviewer.

Tested K2-Think on LiveCodeBench (v6) — 2408:2505 (454 samples). Got pass@1[avg-of-10] = 60.8%, vs. Nemotron-32B at 69.8%. Disappointing to see such inflated/false scores being reported.