Resources
Keeping LLM inference inside the EU on serverless GPUs
Running agent-written code in a sandbox on GPUs
Canary rollouts and rollbacks for GPU endpoints
What a mixture-of-experts model costs to serve per hour
Serverless training for LLMs: where it fits and where it stops
Prefix caching for multi-turn agents
Speculative decoding on per-second GPUs
Inference cost at scale: the tokens-per-minute math
Serving fine-tuned LLMs from one base image
Voice pipelines: which stage needs a GPU
GPU inference for agent workloads
Serverless compute: cost, scaling and cold starts