This document is relevant for: Inf1, Inf2, Trn1, Trn2, Trn3

Tutorials

Step by step guides for using Neuron Agentic Development agents and skills on real tasks.

Tutorial: Port a HuggingFace model using the Autoport skill

Walk through the full process of porting a HuggingFace model to NxD Inference using the Autoport agent. Covers parameter setup, invocation, and what to expect at each stage.

Tutorial: Port a GPU-compatible model to vLLM Neuron using the Autoport skill

Walk through porting a GPU-compatible model to the vLLM Neuron Trainium2 backend using the Autoport vLLM Neuron agent. Covers invocation, what happens at each phase, validation, and serving the result.

Tutorial: Validate a model port using the Equivalence skill

Walk through validating a ported model against its reference using the Equivalence agent. Covers R-ratio testing, fault localization, end-to-end accuracy verification, and validating vLLM Neuron ports.

Tutorial: Capture prefill and decode profiles using the Profiling skill

Capture two device profiles of a vLLM Neuron model on Trainium, one containing only prefill and one containing only steady-state decode, and confirm from the manifest that each trace is real before handing it to the analysis skill.

Tutorial: Analyze a vLLM Neuron profile using the Profile Analysis skill

Work interactively with the analysis agent to find where the time goes: which graphs ran, a step’s roofline floors, its binding engine, its collective share, and which model component is responsible, with five checks that catch a plausible but wrong answer.

Tutorial: Plan a serving config using the Parallelism Planner skill

Walk through choosing a parallelism config for Granite-4.1-8B on trn2 using the Parallelism Planner agent. Covers writing a run-spec, oracle validation, reading the sweep, and inspecting a single operating point.

Tutorial: Find the binding operator using the Hardware Entitlement skill

Walk through computing per-operator floors for the planner’s recommended config using the Hardware Entitlement agent. Covers decode and prefill floors, how the binding moves with batch and context, and actual-versus-floor HBM attribution.

This document is relevant for: Inf1, Inf2, Trn1, Trn2, Trn3