2026 Evolution Episode 10: Accelerating LLM Inference with NVIDIA TensorRT-LLM

This episode accelerates large-language-model inference with NVIDIA TensorRT-LLM. It covers model conversion, precision and quantization choices, optimized kernels and execution graphs, KV cache, batching, and parallelism, then evaluates the tradeoffs among throughput, time to first token, memory use, and model quality.

Total recording duration: 24 min 51 sec.

Companion Git repository: https://qytgit.qytang.com/qytadmin/2026-10-TensorRT-LLM