From the source
Model: https://huggingface.co/NousResearch/NousCoder-14B Code: https://github.com/NousResearch/atropos/pull/296 WandB: https://api.wandb.ai/links/jli505/ksz3e9w6 Introduction We introduce NousCoder-14B, a competitive olympiad programming model post-trained on Qwen3-14B via reinforcement learning.
- On LiveCodeBench v6 (08/01/2024 - 05/01/2025), we achieve a Pass@1 accuracy of 67.87%, up 7.08% from the baseline Pass@1 accuracy of 60.79% for Qwen3-14B.
- We trained on 24k verifiable coding problems using 48 B200s over the course of four days.
- Dataset Following the success of Agentica x
Together AI's DeepCoder-14B [1] , we use the same dataset as them.
- In particular, our training dataset consists of 24K problems from TACO Verified, PrimeIntellect's SYNTHETIC-1, and LiveCodeBench problems before 07/31/2024.
- The test dataset we use is LiveCodeBench v6 (08/01/2024–05/01/2025) consisting of 454 problems [2] .
- We verified that there is no contamination between the train and test datasets.
- …