AWS Neuron docs — Inf1 Inference Performance tables

Vendor reference inference numbers for Inf1 (Inferentia1). NeuronPerf-driven. Encoder models reported as inference/sec (sentences/sec), CNN models as images/sec — NOT token-based; no generative LLM rows exist for Inf1 in this doc. Last update 2024-09-16, Neuron 2.20. inf1.xlarge=1 Inferentia chip? (inf1.xlarge exposes 4 NeuronCores; each Inferentia chip = 4 NeuronCores so 1 chip).