5 papers
Towards Load-Aware Prefill Deflection for Disaggregated LLM Serving
Shrikara Arun, Anjaly Parayil, Srikant Bharadwaj +2
Disaggregated LLM serving runs prefill and decode on separate GPU pools to keep the two phases from interfering. In practice, this creates a new asymmetry: under bursty, heavy-tail…
SIFT: Selective-Index For Fast Compute of RAG Prefill by Exploiting Attention Invariance
Rya Sanovar, Srikant Bharadwaj, Hritvik Taneja +1
Retrieval-Augmented Generation (RAG) injects LLM queries with relevant documents to improve response quality. This injection increases prompt length and slows time to first token (…
Taming the Tail: NoI Topology Synthesis for Mixed DL Workloads on Chiplet-Based Accelerators
Arnav Shukla, Harsh Sharma, Srikant Bharadwaj +2
Heterogeneous chiplet-based systems improve scaling by disag-gregating CPUs/GPUs and emerging technologies (HBM/DRAM).However this on-package disaggregation introduces a latency in…
Lean Attention: Hardware-Aware Scalable Attention Mechanism for the Decode-Phase of Transformers
Rya Sanovar, Srikant Bharadwaj, Renee St. Amant +2
Transformer-based models have emerged as one of the most widely used architectures for natural language processing, natural language generation, and image generation. The size of t…
TurboAttention: Efficient Attention Approximation For High Throughputs LLMs
Hao Kang, Srikant Bharadwaj, James Hensman +3
Large language model (LLM) inference demands significant amount of computation and memory, especially in the key attention mechanism. While techniques, such as quantization and acc…