Curated customer-support and transcript-analytics prompts mapped to a single fixed "analyze this transcript -> compact JSON" prompt, for benchmarking batched offline LLM inference on realistic workloads.

Motivation and intended use

This dataset provides a realistic transcript-analytics workload for batched offline-inference experiments: throughput benchmarking and predicted-vs-observed throughput validation. Rows carry token… See the full description on the dataset page: https://huggingface.co/datasets/kaushik-systalyze/customer-transcript-long-dialog.