5 papers
CoPRIS: Efficient and Stable Reinforcement Learning via Concurrency-Controlled Partial Rollout with Importance Sampling
Zekai Qu, Yinxu Pan, Ao Sun +2
Reinforcement learning (RL) post-training has become a trending paradigm for enhancing the capabilities of large language models (LLMs). Most existing RL systems for LLMs operate i…
NOSA: Native and Offloadable Sparse Attention
Yuxiang Huang, Pengjie Wang, Jicheng Han +9
Decoding throughput improvements from larger inference batches are limited by GPU memory, which is largely consumed by the key-value (KV) cache. Prior training-free KV cache offloa…
InfLLM-V2: Dense-Sparse Switchable Attention for Seamless Short-to-Long Adaptation
Weilin Zhao, Zihan Zhou, Zhou Su +10
Long-sequence processing is a critical capability for modern large language models. However, the self-attention mechanism in the standard Transformer architecture faces severe comp…
BurstEngine: an Efficient Distributed Framework for Training Transformers on Extremely Long Sequences of over 1M Tokens
Ao Sun, Weilin Zhao, Xu Han +4
Existing methods for training LLMs on long-sequence data, such as Tensor Parallelism and Context Parallelism, exhibit low Model FLOPs Utilization as sequence lengths and number of…
FR-Spec: Accelerating Large-Vocabulary Language Models via Frequency-Ranked Speculative Sampling
Weilin Zhao, Tengyu Pan, Xu Han +9
Speculative sampling has emerged as an important technique for accelerating the auto-regressive generation process of large language models (LLMs) by utilizing a draft-then-verify…