Expensive Language Model

Expensive language models, while powerful, present significant computational challenges. Research focuses on improving efficiency through techniques like lightweight adapters for multimodal models (reducing the need for full model retraining), novel decoding algorithms to generate multiple outputs from a single inference pass (speeding up generation), and knowledge distillation methods to compress large models into smaller, faster ones without sacrificing accuracy. These advancements aim to make powerful language models more accessible and environmentally sustainable by reducing computational costs and energy consumption for various applications, including natural language processing and multimodal tasks.

Papers

August 13, 2024

CROME: Cross-Modal Adapters for Efficient Multimodal LLM
Sayna Ebrahimi, Sercan O. Arik, Tejas Nama, Tomas Pfister
Multimodal Large Language Model Multimodal LLM Modal Adapter Cross Modal Understanding Vision Language Instruction Tuning Expensive Language Model

May 28, 2024

Superposed Decoding: Multiple Generations from a Single Autoregressive Inference Pass
Ethan Shen, Alan Fan, Sarah M. Pratt, Jae Sung Park, Matthew Wallingford, Sham M. Kakade, Ari Holtzman, Ranjay Krishna, Ali Farhadi, Aditya Kusupati
Autoregressive Model N Gram Greedy Decoding K$ Draft Multiple Generation Expensive Language Model

November 6, 2023

Co-training and Co-distillation for Quality Improvement and Compression of Language Models
Hayeon Lee, Rui Hou, Jongpil Kim, Davis Liang, Hongbo Zhang, Sung Ju Hwang, Alexander Min
Language Model Knowledge Distillation Linear Compression Mutual Distillation Inference Speed Enhanced Quality Expensive Language Model Co Distillation

January 24, 2023

A Comparison of Temporal Encoders for Neuromorphic Keyword Spotting with Few Neurons
Mattias Nilsson, Ton Juny Pina, Lyes Khacef, Foteini Liwicki, Elisabetta Chicca, Fredrik Sandin
Consistent Comparison Neuromorphic Processor Temporal Encoder Neuromorphic Sensor Temporal Encoding Neural Assembly Expensive Language Model

May 24, 2022

BabyBear: Cheap inference triage for expensive language models
Leila Khalili, Yao You, John Bohannon
Natural Language Processing Entity Recognition Transformer Language Model Inference Workload Expensive Language Model

Expensive Language Model

Papers

CROME: Cross-Modal Adapters for Efficient Multimodal LLM

Superposed Decoding: Multiple Generations from a Single Autoregressive Inference Pass

Co-training and Co-distillation for Quality Improvement and Compression of Language Models

A Comparison of Temporal Encoders for Neuromorphic Keyword Spotting with Few Neurons

BabyBear: Cheap inference triage for expensive language models