mindfossil/telecom-intelligence-model-v2-merged
The mindfossil/telecom-intelligence-model-v2-merged is a 7.6 billion parameter language model, based on Qwen2.5-7B, specifically fine-tuned for telecom network operations. It excels at tasks such as RAN anomaly detection, 5G Core root cause analysis, multi-vendor KPI normalization, and natural language to CLI command generation for Ericsson, Nokia, and Huawei equipment. This model is designed to provide intelligence and automation for complex telecommunications infrastructure management.
Loading preview...
Overview
The mindfossil/telecom-intelligence-model-v2-merged is a 7.6 billion parameter language model, built upon the Qwen2.5-7B base, and meticulously fine-tuned for specialized applications in telecommunications network operations. It leverages Supervised Fine-Tuning (SFT) with Chain-of-Thought to enhance its reasoning capabilities within the telecom domain.
Key Capabilities
This model offers a comprehensive suite of functionalities tailored for network intelligence:
- PM KPI Derivation: Computes key performance indicators (KPIs) like RRC Success Rate, ERAB SR, and Throughput from raw PM counter values, including step-by-step arithmetic and vendor-specific formulas (Ericsson, Nokia, Huawei).
- RAN Anomaly Detection: Identifies degraded KPIs by applying derived metrics against operational thresholds and recommends remediation actions for cell health issues.
- 5G Core Root Cause Analysis (RCA): Analyzes SMF/AMF telemetry and classifies faults (e.g., CPU fault, IP pool exhaustion) within the 5G Core network.
- Multi-vendor KPI Normalization: Maps vendor-specific counter names to unified KPI formulas, enabling comparable performance analysis across different equipment manufacturers.
- Natural Language to CLI (NL → CLI): Translates natural language operational requests into valid Ericsson MML and AMOS CLI commands.
Evaluation and Performance
The model was evaluated across 20 functional test cases, achieving an overall score of 77%. Specific domain scores include 81% for RAN anomaly detection and 89% for multi-vendor tasks. A separate benchmark for 5GC telemetry RCA yielded 86% accuracy, meeting deployment confidence thresholds.
When to Use This Model
This model is ideal for use cases requiring automated analysis, troubleshooting, and operational intelligence within complex telecom environments, particularly for tasks involving RAN and 5G Core networks, multi-vendor equipment, and the translation of operational intent into network commands.