Network Engineer Specialist Models

Overview

The network engineer specialist models are the end application of the fine-tuning pipeline — a Juniper expert and a Cisco expert, each deployed disconnected on a Mac laptop and given direct access to connected network devices.

Entries

  • Juniper Expert Model — JunOS-specific design, training data specification, and deployment
  • Cisco Expert Model — IOS/IOS-XE/NX-OS/ASA design, multi-platform coverage, and deployment
  • Edge Deployment and Device Integration — Running disconnected on a MacBook, connecting to network devices, and building an agentic command loop
  • Context Window Management — Handling long show command output, sliding-window conversation trimming, two-pass summarization, and performance impact of context length on Apple Silicon
  • LoRA Adapter Versioning and Rollback — Directory layout for base/adapter separation, training run metadata, registering multiple versions in Ollama, rollback procedure, and what to keep vs. delete

Entries

  • Cisco Expert Model — Design specification for a fine-tuned Cisco network engineer specialist LLM covering IOS, IOS-XE, NX-OS, and ASA: base model, training data requirements covering multiple OS variants, fine-tuning parameters, and edge deployment.
  • Context Window Management for Device-Connected Models — How to manage the context window when a local LLM is connected to real network devices — handling long show command output, maintaining troubleshooting state across turns, and the tradeoff between context length and inference speed on Apple Silicon.
  • Edge Deployment and Device Integration — Deploying a fine-tuned network engineer specialist LLM disconnected on a Mac laptop, connecting it to physical network devices via SSH, and building an agentic command loop that can read device state and propose or execute commands.
  • Juniper Expert Model — Design specification for a fine-tuned Juniper/JunOS network engineer specialist LLM: base model selection, competency targets, training data requirements, fine-tuning parameters, and deployment as a local disconnected assistant.
  • LoRA Adapter Versioning and Rollback — How to version, store, and swap LoRA adapters as you iterate on a fine-tuned network expert model — keeping the base model separate from adapters, testing a new adapter before switching, and rolling back when a new version regresses.