ITERTL: An Iterative Framework for Fine-tuning LLMs for RTL Code Generation

TL;DR

This paper introduces ITERTL, an iterative training framework for fine-tuning large language models to generate RTL code, significantly improving performance and reducing reliance on large reference datasets.

Contribution

The paper proposes a novel iterative training paradigm with a plug-and-play data filtering strategy for RTL code generation using LLMs, outperforming existing models.

Findings

Achieves 53.8% pass@1 on VerilogEval-human benchmark

Outperforms GPT-4 and SOTA open-source models under similar data conditions

Demonstrates the effectiveness of iterative training and data filtering strategies

Abstract

Recently, large language models (LLMs) have demonstrated excellent performance, inspiring researchers to explore their use in automating register transfer level (RTL) code generation and improving hardware design efficiency. However, the existing approaches to fine-tune LLMs for RTL generation typically are conducted on fixed datasets, which do not fully stimulate the capability of LLMs and require large amounts of reference data, which are costly to acquire. To mitigate these issues, we innovatively introduce an iterative training paradigm named ITERTL. During each iteration, samples are drawn from the model trained in the previous cycle. Then these new samples are employed for training in current loop. Furthermore, we introduce a plug-and-play data filtering strategy, thereby encouraging the model to generate high-quality, self-contained code. Our model outperforms GPT4 and state-of-the-art (SOTA) open-source models, achieving remarkable 53.8% pass@1 rate on VerilogEval-human benchmark. Under similar conditions of data quantity and quality, our approach significantly outperforms the baseline. Extensive experiments validate the effectiveness of the proposed method.