The development of large language models (LLMs) has significantly propelled the latest advancements in natural language processing (NLP). These models have been built upon complex deep learning architectures, typically Transformer, characterized by billions of parameters and extensive training data, enabling them to achieve high precision across a variety of tasks. However, the absence of agricultural machinery-specific textual data for training in existing general large models has severely limited their performance in the research, development, manufacturing, and application of agricultural machinery. To address this issue, the specific needs of agricultural machinery for large models were analyzed and a Chinese-compatible agricultural machinery large model named "LeiSi" was proposed, catering to various groups such as university faculty, students, designers, and users. The overall architectural design of the LeiSi LLM was outlined, which included three parts: LeiSi-torch, LeiSi-ingenuity, and LeiSi-plough, aiming to provide target groups with diversified and customized services such as agricultural machinery professional knowledge Q&A, agricultural machinery design and manufacturing advice, and agricultural machinery field operation control. Subsequently, the LeiSi-torch large language model was taken as an example to elucidate the construction of a Chinese agricultural machinery dataset and the methods for model fine-tuning and automatic evaluation. Utilizing LLaMA 3.1-8B-Instruct, Mistral-7B-Instruct-v0.3, and Qwen 2.5-7B-Instruct as base models, supervised fine-tuning on each was conducted and the performance of the fine-tuned models was evaluated by using ROUGE and BLEU as evaluation metrics. Finally, manual evaluation was employed to assess the Q&A results of LLaMA 3.1, GPT-4o, Mistral, Qwen 2.5, and the LeiSi-torch model. The results from both automatic and manual evaluations indicated that the LeiSi-torch model demonstrated superior performance in terms of accuracy, professionalism, and usability. The research outcomes can provide insights and references for the development of intelligent agricultural machinery and smart agriculture.