Salvato in:
Dettagli Bibliografici
Autori principali: Zheng, Kunhao, Decugis, Juliette, Gehring, Jonas, Cohen, Taco, Negrevergne, Benjamin, Synnaeve, Gabriel
Natura: Preprint
Pubblicazione: 2024
Soggetti:
Accesso online:https://arxiv.org/abs/2410.08105
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866912313928318976
author Zheng, Kunhao
Decugis, Juliette
Gehring, Jonas
Cohen, Taco
Negrevergne, Benjamin
Synnaeve, Gabriel
author_facet Zheng, Kunhao
Decugis, Juliette
Gehring, Jonas
Cohen, Taco
Negrevergne, Benjamin
Synnaeve, Gabriel
contents Prompting techniques such as chain-of-thought have established themselves as a popular vehicle for improving the outputs of large language models (LLMs). For code generation, however, their exact mechanics and efficacy are under-explored. We thus investigate the effects of a wide range of prompting strategies with a focus on automatic re-prompting over multiple turns and computational requirements. After systematically decomposing reasoning, instruction, and execution feedback prompts, we conduct an extensive grid search on the competitive programming benchmarks CodeContests and TACO for multiple LLM families and sizes (Llama 3.0 and 3.1, 8B, 70B, 405B, and GPT-4o). Our study reveals strategies that consistently improve performance across all models with small and large sampling budgets. We then show how finetuning with such an optimal configuration allows models to internalize the induced reasoning process and obtain improvements in performance and scalability for multi-turn code generation.
format Preprint
id arxiv_https___arxiv_org_abs_2410_08105
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle What Makes Large Language Models Reason in (Multi-Turn) Code Generation?
Zheng, Kunhao
Decugis, Juliette
Gehring, Jonas
Cohen, Taco
Negrevergne, Benjamin
Synnaeve, Gabriel
Computation and Language
Prompting techniques such as chain-of-thought have established themselves as a popular vehicle for improving the outputs of large language models (LLMs). For code generation, however, their exact mechanics and efficacy are under-explored. We thus investigate the effects of a wide range of prompting strategies with a focus on automatic re-prompting over multiple turns and computational requirements. After systematically decomposing reasoning, instruction, and execution feedback prompts, we conduct an extensive grid search on the competitive programming benchmarks CodeContests and TACO for multiple LLM families and sizes (Llama 3.0 and 3.1, 8B, 70B, 405B, and GPT-4o). Our study reveals strategies that consistently improve performance across all models with small and large sampling budgets. We then show how finetuning with such an optimal configuration allows models to internalize the induced reasoning process and obtain improvements in performance and scalability for multi-turn code generation.
title What Makes Large Language Models Reason in (Multi-Turn) Code Generation?
topic Computation and Language
url https://arxiv.org/abs/2410.08105