9 papers
Double-Rollout Reinforcement Learning with Sampling-based Reward Estimation
Haifeng Qian, David Mosallanezhad, Christian Munley +2
Reinforcement learning with verifiable rewards (RLVR) implemented with Group Relative Policy Optimization (GRPO) has driven substantial progress in training reasoning language mode…
Nemotron 3 Ultra: Open, Efficient Mixture-of-Experts Hybrid Mamba-Transformer Model for Agentic Reasoning
NVIDIA, :, Aaron Blakeman +572
We introduce Nemotron 3 Ultra, a 550 billion total and 55 billion active parameter Mixture-of-Experts Hybrid Mamba-Attention language model. We pre-trained Nemotron 3 Ultra on 20 t…
Nemotron 3 Super: Open, Efficient Mixture-of-Experts Hybrid Mamba-Transformer Model for Agentic Reasoning
NVIDIA, :, Aakshita Chandiramani +544
We describe the pre-training, post-training, and quantization of Nemotron 3 Super, a 120 billion (active 12 billion) parameter hybrid Mamba-Attention Mixture-of-Experts model. Nemo…
LibEvolutionEval: A Benchmark and Study for Version-Specific Code Generation
Sachit Kuhar, Wasi Uddin Ahmad, Zijian Wang +6
Recent advancements in code completion models have primarily focused on local file contexts. However, these studies do not fully capture the complexity of real-world software devel…
Approximately Aligned Decoding
Daniel Melcer, Sujan Gonugondla, Pramuditha Perera +7
It is common to reject undesired outputs of Large Language Models (LLMs); however, current methods to do so require an excessive amount of computation to re-sample after a rejectio…
CodeFort: Robust Training for Code Generation Models
Yuhao Zhang, Shiqi Wang, Haifeng Qian +8
Code generation models are not robust to small perturbations, which often lead to incorrect generations and significantly degrade the performance of these models. Although improvin…
BASS: Batched Attention-optimized Speculative Sampling
Haifeng Qian, Sujan Kumar Gonugondla, Sungsoo Ha +6
Speculative decoding has emerged as a powerful method to improve latency and throughput in hosting large language models. However, most existing implementations focus on generating…
Bifurcated Attention: Accelerating Massively Parallel Decoding with Shared Prefixes in LLMs
Ben Athiwaratkun, Sujan Kumar Gonugondla, Sanjay Krishna Gouda +10
This study introduces bifurcated attention, a method designed to enhance language model inference in shared-context batch decoding scenarios. Our approach addresses the challenge o…
Constrained Decoding for Fill-in-the-Middle Code Language Models via Efficient Left and Right Quotienting of Context-Sensitive Grammars
Daniel Melcer, Nathan Fulton, Sanjay Krishna Gouda +1
Large Language Models are powerful tools for program synthesis and advanced auto-completion, but come with no guarantee that their output code is syntactically correct. This paper…