ramixpe/Iosxr-expert

IOS-XR Expert - Fine-tuned Qwen3.5-9B

A specialized language model for Cisco IOS-XR service provider networking, fine-tuned from Qwen3.5-9B.

Model Details

• Base Model: Qwen/Qwen3.5-9B (8.95B parameters)
• Fine-tuning: LoRA r=64, bf16, all linear layers (116M trainable params)
• Training: 5 epochs, A100 80GB, 3 hours
• Dataset: 1,190 curated IOS-XR QA pairs across 8 task families

Performance (V1)

Metric	Score
Syntax accuracy	92.3%
Semantic correctness	95.6%
Contamination resistance	89.8%
Operational quality	23.8%
Overall	83.0%

Capabilities

• IOS-XR configuration generation (BGP, MPLS, SR, EVPN, IS-IS, OSPF, L3VPN, L2VPN)
• IOS/IOS-XE to IOS-XR migration
• Configuration error detection and correction
• Troubleshooting guidance
• Route-policy (RPL) authoring
• CLI to YANG mapping

Usage with Ollama

ollama create iosxr-expert -f Modelfile
ollama run iosxr-expert "Configure BGP VPNv4 peering on IOS-XR"

GGUF Quantizations

• iosxr-qwen3.5-9b-q8_0.gguf (8.9 GB) - Best quality
• iosxr-qwen3.5-9b-q4_k_m.gguf (5.3 GB) - Best size/quality ratio