The purpose of this paper is to explore the implementation of retrieval-augmented generation (RAG) technology with open-source large language models (LLMs). A dedicated web-based application, PaSSER, was developed, integrating RAG with Mistral:7b, Llama2:7b, and Orca2:7b models. Various software instruments were used in the application’s development. PaSSER employs a set of evaluation metrics, including METEOR, ROUGE, BLEU, perplexity, cosine similarity, Pearson correlation, and F1 score, to assess LLMs’ performance, particularly within the smart agriculture domain. The paper presents the results and analyses of two tests. One test assessed the performance of LLMs across different hardware configurations, while the other determined which model delivered the most accurate and contextually relevant responses within RAG. The paper discusses the integration of blockchain with LLMs to manage and store assessment results within a blockchain environment. The tests revealed that GPUs are essential for fast text generation, even for 7b models. Orca2:7b on Mac M1 was the fastest, and Mistral:7b had superior performance on the 446 question–answer dataset. The discussion is on technical and hardware considerations affecting LLMs’ performance. The conclusion outlines future developments in leveraging other LLMs, fine-tuning approaches, and further integration with blockchain and IPFS.

The purpose of this paper is to explore the implementation of Retrieval-Augmented Gen- 14 eration (RAG) technology with open-source Large Language Models (LLMs). For this purpose, a 15 web-based application named PaSSER that integrates RAG with three models - Mistral:7b, 16 Llama2:7b: Orca2:7b, is developed. In its development various technologies are used, including 17 Blockchain for managing and storing assessment results of built-in testing modules. PaSSER em- 18 ploys a set of evaluation metrics for LLMs’ performance - METEOR, ROUGE, BLEU, Perplexity, 19 Cosine similarity, Pearson correlation, and F1 score. The specialized knowledge base is in the smart 20 agriculture domain. As an illustration, the paper presents results and analysis of two tests. First one 21 is assessing the performance of LLMs across different hardware configurations, the other is deter- 22 mining the model that delivers most accurate and contextually relevant responses within the RAG. 23 The paper also discusses the integration of blockchain technology with LLMs to manage and store 24 assessment results within a blockchain environment. The tests revealed that Mistral:7b demon- 25 strates the best results across most of the evaluated metrics. The discussion section comments on 26 technical and hardware considerations that affect the performance of LLMs. Future development 27 will focus in leveraging new larger pre-trained open-source LLMs, exploring different finetuning 28 approaches, and further integration with Blockchain and IPFS.

This paper evaluates multi-agent coordination strategies against single-agent retrieval-augmented generation (RAG) for open-source language models. Four coordination strategies (collaborative, sequential, competitive, hierarchical) were tested across Mistral 7B, Llama 3.1 8B, and Granite 3.2 8B using 100 domain-specific question–answer pairs (3100 total evaluations). Performance was assessed using Composite Performance Score (CPS) and Threshold-aware CPS (T-CPS), aggregating nine metrics spanning lexical, semantic, and linguistic dimensions. Under the tested conditions, all 28 multi-agent configurations showed degradation relative to single-agent baselines, ranging from −4.4% to −35.3%. Coordination overhead was identified as a primary contributing factor. Llama 3.1 8B tolerated Sequential and Hierarchical coordination with minimal degradation (−4.9% to −5.3%). Mistral 7B with shared context retrieval achieved comparable results. Granite 3.2 8B showed degradation of 14–35% across all strategies. Collaborative coordination exhibited the largest degradation across all models. Study limitations include evaluation on a single domain (agriculture), use of 7–8B parameter models, and homogeneous agent architectures. These findings suggest that single-agent RAG may be preferable for factual question-answering tasks in local deployment scenarios with computational constraints. Future research should explore larger models, heterogeneous agent teams, role-specific prompting, and advanced consensus mechanisms.

This paper evaluates multi-agent coordination strategies against retrieval-augmented generation for 7-8B open-source models. Four coordination strategies (collaborative, sequential, competitive and hierarchical) were evaluated across three open-source models: Mistral 7B, Llama 3.1 8B and Granite 3.2 8B. The study determined whether multi-agent reasoning enhances retrieval-augmented generation performance. The evaluation employed 100 question-answer pairs. In total, 2,000 model-question evaluations were conducted. Performance was assessed using Composite Performance Score (CPS) and Threshold-aware Composite Performance Score (T-CPS), two metrics developed to aggregate nine dimensions spanning lexical overlap, semantic similarity, and linguistic quality. Results revealed that 87.5% of multi-agent configurations underperformed baseline systems, with coordination overhead identified as the primary limiting factor. Llama 3.1 8B tolerated Sequential and Hierarchical coordination with minimal degradation, while Granite 3.2 8B and Mistral 7B showed severe degradation across all strategies. Collaborative coordination failed universally despite highest output consistency. These findings suggest that single-agent baselines may be preferable for most deployment scenarios under similar conditions. Future research should explore following developments: evaluation of role-specific prompts, investigation of advanced consensus methods, exploration of adaptive systems for strategy selection, and joint tuning of retrieval thresholds and coordination strategies.