AI

Speculative Streaming: Fast LLM Inference Without Auxiliary Models

1 Mins read

Speculative decoding is a prominent technique to speed up the inference of a large target language model based on predictions of an auxiliary draft model. While effective, in application-specific settings, it often involves fine-tuning both draft and target models to achieve high acceptance rates. As the number of downstream tasks grows, these draft models add significant complexity to inference systems. We propose Speculative Streaming, a single-model speculative decoding method that fuses drafting into the target model by changing the fine-tuning objective from next token prediction to future n-gram prediction. Speculative Streaming speeds up decoding by 1.8 – 3.1X in a diverse set of tasks, such as Summarization, Structured Queries, and Meaning Representation, without sacrificing generation quality. Additionally, Speculative Streaming is parameter-efficient. It achieves on-par/higher speed-ups than Medusa-style architectures while using ~10000X fewer extra parameters, making it well-suited for resource-constrained devices.


Source link

Related posts
AI

Would you trust AI to mediate an argument?

2 Mins read
Researchers from Google DeepMind recently trained a system of large language models to help people come to agreement over complex but important…
AI

UC Berkeley Researchers Propose DocETL: A Declarative System that Optimizes Complex Document Processing Tasks using LLMs

3 Mins read
Large Language Models (LLMs) have gained significant attention in data management, with applications spanning data integration, database tuning, query optimization, and data…
AI

LongAlign: A Segment-Level Encoding Method to Enhance Long-Text to Image Generation

3 Mins read
The rapid progress of text-to-image (T2I) diffusion models has made it possible to generate highly detailed and accurate images from text inputs….

 

 

Leave a Reply

Your email address will not be published. Required fields are marked *