On the effectiveness of pretrained models for API learning

Hadi, M. A.; Yusuf, Imam Nur Bani; Thung, Ferdian; Luong, Kien; Jiang, Lingxiao; Fard, Fatemeh H.; Lo, David

doi:10.1145/3524610.3527886

Cited by 6 publications

(1 citation statement)

References 36 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The models vary in size and training schemes and their success rate in performing code generation tasks . Several works leverage pre-trained models to map NL queries to sequences of API elements or other intermediate representations, which can be translated into code, though this may not be necessary with models that have been pre-trained on code (Hadi et al, 2022;Shin and Van Durme, 2022). They have also been used for in-place data transformations (Narayan et al, 2022).…”

Section: Program Synthesismentioning

confidence: 99%

InstructExcel: A Benchmark for Natural Language Instruction in Excel

Payan,

Mishra,

Singh

et al. 2023

Findings of the Association for Computational Linguistics: EMNLP 2023

View full text Add to dashboard Cite

With the evolution of Large Language Models (LLMs) we can solve increasingly more complex NLP tasks across various domains, including spreadsheets. This work investigates whether LLMs can generate code (Excel Office-Scripts, a TypeScript API for executing many tasks in Excel) that solves Excel specific tasks provided via natural language user instructions. To do so we introduce a new large-scale benchmark, INSTRUCTEXCEL, 1 created by leveraging the 'Automate' feature in Excel to automatically generate OfficeScripts from users' actions. Our benchmark includes over 10k samples covering 170+ Excel operations across 2,000 publicly available Excel spreadsheets. Experiments across various zero-shot and few-shot settings show that INSTRUCTEXCEL is a hard benchmark for state of the art models like GPT-4. We observe that (1) using GPT-4 over GPT-3.5, (2) providing more in-context examples, and (3) dynamic prompting can help improve performance on this benchmark.

show abstract

Section: Program Synthesismentioning

confidence: 99%