This repository contains the full-precision 16-bit (FP16) merged weights of a fine-tuned Qwen2.5-3B-Instruct model, specialized for autonomous eSIM package recommendations and telecom agent workflows.
Unlike quantized versions, this model retains maximum parameter fidelity, making it natively ready for high-throughput production inference servers.
🎯 Key Features
Uncompromised Quality: Zero information loss from quantization; ideal for strict production testing.
eSIM Plan Recommendation: Matches user travel patterns (destination, days) with appropriate package limits and checkout links.
Production Infrastructure Ready: Can be loaded directly into enterprise inference engines.
🔧 Production Deployment (vLLM Engine)
This non-quantized model is perfectly optimized for deployment via vLLM for maximum concurrent throughput. You can spin up an OpenAI-compatible API server using the following command:
