An Open-Source Chatbot by Using ParlAI Kiran Mai Narnavaram Kennesaw State University Marietta, Georgia, USA knarnava@students.kennesaw.edu Dan Lo Kennesaw State University Marietta, Georgia, USA dlo2@kennesaw.edu ABSTRACT While existing work shows a great success, the accuracy, the re- sponse time, the interoperability, the extendibility, etc. remain to be improved. We proposed “ParlAIDialogTeacher” using the open- source ParlAI to improve the accuracy, the response time, the inter- operability, and the extendibility. Training is based on Maximum Likelihood Estimation (MLE) approach for the generative model. This platform eases sharing, training, and evaluating dialog models with multiple datasets available. Our results indicate the proposed model outperforms x, y, z in terms of response time and accuracy. Future work include extending the model with other genitive mod- els such as GAN. CCS CONCEPTS • Computing methodologies → Neural networks; Learning paradigms and algorithms; Semantic networks; Dialog systems. KEYWORDS deep learning, chatbot, ParlAI - Blended Skill Talk ACM Reference Format: Kiran Mai Narnavaram and Dan Lo. 2024. An Open-Source Chatbot by Using ParlAI. In 2024 ACM Southeast Conference (ACMSE 2024), April 18– 20, 2024, Marietta, GA, USA. ACM, New York, NY, USA, 2 pages. https:// doi.org/10.1145/3603287.3656160 1 INTRODUCTION AND RELATED WORK Chatbots, advancing digital communication, face challenges in achieving human-like conversational abilities [1], with specializa- tion enhancing efficiency but limiting versatility. Recent break- throughs in neural approaches [2], particularly large pre-trained Transformer models, have shown promise in open-source conver- sational AI, as evidenced by successes in the ConvAI2 competition and subsequent improvements in dialogue system evaluations [10]. ParlAI and BlenderBot 2.0 exemplify the collaborative spirit and potential for chatbots to engage in more meaningful dialogues. Figure 1 shows that ParlAI and BlenderBot 2.0 exemplify the col- laborative spirit and potential for chatbots to engage in meaningful dialogues. The development of Meena, a model with 2.6 billion pa- rameters, marks a significant milestone in surpassing benchmarks and pushing the boundaries of chatbot capabilities [5]. Permission to make digital or hard copies of part or all of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. Copyrights for third-party components of this work must be honored. For all other uses, contact the owner/author(s). ACMSE 2024, April 18–20, 2024, Marietta, GA, USA © 2024 Copyright held by the owner/author(s). ACM ISBN 979-8-4007-0237-2/24/04. https://doi.org/10.1145/3603287.3656160 Figure 1: Chatting Example 2 METHODOLOGY Our methodology revolutionizes chatbot development with a dy- namic Seq2Seq transformer architecture in ParlAI for response generation, diverges from static databases. Employs Hugging Face Tokenizers for Byte-Level BPE, tests models with 90M to 9.4B pa- rameters including a setup with four-layer encoder and 32-layer decoder [5]. Uses Maximum Likelihood Estimation (MLE) with an 8-layer RNN and GeLu-activated network [7], incorporates min- imum length and predictive training for authentic dialogue, and enhances computational efficiency via remote server fine-tuning with ParlAI. 3 EXPERIMENTAL SETUP AND METRICS Figure 2 illustrates our approach utilizing the Fairseq toolkit for pre-training [6], with models of 2.7B and 9.4B parameters optimized by the Adam optimizer. We incorporated Megatron-LM’s model parallelism [9] for efficient node distribution, leveraging vertical transformer layer slicing for reduced cross-GPU communication. A sublinear memory footprint, the Adafactor [8] is used to facilitate larger batch size precision training. 4 PERPLEXITY Automatic metric correlating with human judgment accelerates discourse model creation; perplexity, an uncertainty measure in Seq2Seq models. Our research demonstrates a positive correlation 2024 ACM Southeast Conference – ACMSE 2024 – Session 3: Posters – ISBN: 979-8-4007-0237-2 Marietta, Georgia, USA, April 18-20, 2024 323