How NetoAI trained a Telecom-specific large language model using Amazon SageMaker and AWS Trainium

NetoAI developed TSLAM (Telecom-Specific Large Language Model), the first open-source LLM dedicated to telecommunications, using AWS Trainium and Amazon SageMaker.

• Fine-tuned from Llama-3.1-8B using proprietary telecom datasets
• Used LoRA method for efficient parameter adaptation
• Trained on AWS Trainium instances, reducing training time and costs
• Achieved 86.2% accuracy compared to 63.1% of base model
• Integrated into VING platform with specialized telecom AI agents

Key use cases include network fault diagnosis, customer service, network planning, and device configuration management. NetoAI plans to further scale the model using AWS EC2 Trn2 instances.