Exploring the Benefits of Training Expert Language Models over Instruction Tuning

Recently, Language Models (LMs) instruction-tuned on multiple tasks, also known as multitask-prompted fine-tuning (MT), have shown the capability to generalize to unseen tasks. Previous work has shown that scaling the number of training tasks is the key component in making stronger MT LMs. In this work, we report an unexpected finding that an expert LM fine-tuned on just a single task can outperform an MT LM trained with 300+ different tasks on 11 different unseen datasets and on 13 datasets of the BIG-bench benchmark by a mean accuracy of 3.20% and 1.29%, respectively. This finding casts doubt on the previously held belief that simply scaling the number of tasks makes stronger MT LMs. Leveraging this finding, we further show that this distributed approach of training a separate expert LM per training task instead of a single MT LM for zero-shot inference possesses many benefits including (1) avoiding negative task transfer that often occurs during instruction tuning, (2) being able to continually learn new tasks without having to re-train on previous tasks to avoid catastrophic forgetting, and (3) showing compositional capabilities when merging individual experts together. The code is available at https://github.com/joeljang/ELM.

PDF Abstract

Results from the Paper


Task Dataset Model Metric Name Metric Value Global Rank Result Benchmark
Sentence Completion HellaSwag RoE-3B Accuracy 34.6 # 77

Results from Other Papers


Task Dataset Model Metric Name Metric Value Rank Source Paper Compare
Natural Language Inference ANLI test RoE-3B A1 35.49 # 15
A2 34.64 # 20
A3 31.22 # 25
Question Answering COPA RoE-3B Accuracy 79.25 # 37
Natural Language Inference RTE RoE-3B Accuracy 64.01 # 65
Question Answering StoryCloze RoE-3B Accuracy 86.33 # 9
Coreference Resolution Winograd Schema Challenge RoE-3B Accuracy 62.21 # 52
Common Sense Reasoning WinoGrande RoE-3B Accuracy 61.60 # 43
Word Sense Disambiguation Words in Context RoE-3B Accuracy 52.97 # 25

Methods


No methods listed for this paper. Add relevant methods here