9 papers
ASPIRE: Asynchronous Batched Self-Speculative Decoding for Long-Context LLM Inference
Amir Ziashahabi, Hossein Entezari Zarch, Lei Gao +2
Long-context LLM inference is bottlenecked by attention, whose repeated KV-cache reads make decoding memory-bound. Self-speculative decoding alleviates this by drafting tokens with…
Striking the Right Balance between Compute and Copy: Improving LLM Inferencing Under Speculative Decoding
Arun Ramachandran, Ramaswamy Govindarajan, Murali Annavaram +4
With the skyrocketing costs of GPUs and their virtual instances in the cloud, there is a significant desire to use CPUs for large language model (LLM) inference. KV cache update, o…
DuetServe: Harmonizing Prefill and Decode for LLM Serving via Adaptive GPU Multiplexing
Lei Gao, Chaoyi Jiang, Hossein Entezari Zarch +3
Modern LLM serving systems must sustain high throughput while meeting strict latency SLOs across two distinct inference phases: compute-intensive prefill and memory-bound decode ph…
DELTA: Dynamic Layer-Aware Token Attention for Efficient Long-Context Reasoning
Hossein Entezari Zarch, Lei Gao, Chaoyi Jiang +1
Large reasoning models (LRMs) achieve state-of-the-art performance on challenging benchmarks by generating long chains of intermediate steps, but their inference cost is dominated…
MARché: Fast Masked Autoregressive Image Generation with Cache-Aware Attention
Chaoyi Jiang, Sungwoo Kim, Lei Gao +3
Masked autoregressive (MAR) models unify the strengths of masked and autoregressive generation by predicting tokens in a fixed order using bidirectional attention for image generat…
DEL: Context-Aware Dynamic Exit Layer for Efficient Self-Speculative Decoding
Hossein Entezari Zarch, Lei Gao, Chaoyi Jiang +1
Speculative Decoding (SD) is a widely used approach to accelerate the inference of large language models (LLMs) without reducing generation quality. It operates by first using a co…