I will deploy and optimize llm inference on gpus with vllm

A
arthur_ac_wu
A
arthur_ac_wu
Arthur Wu

About this gig

Bring your model latency down and throughput up - with measurable numbers.


Typical work:

- Serve your model with vLLM, SGLang or TensorRT-LLM, tuned for your GPU fleet

- Continuous batching, paged attention, prefix caching, speculative decoding

- Quantization (AWQ / GPTQ / FP8) with a quality-vs-speed comparison you can verify

- CUDA / Triton profiling with Nsight to find real bottlenecks

- Docker / Kubernetes packaging, Prometheus + Grafana dashboards, realistic load tests


What you get: serving configs, benchmark results (p50/p95 latency, tokens/sec, cost per 1M tokens), a load-test report and a short runbook your team can operate.


What is NOT included: GPU hardware and cloud spend, model licensing, paid third-party services, 24/7 on-call support. Everything runs on your infrastructure.


17 years of systems and ML engineering; I deploy models on customer GPUs as part of my daily product work.


Send me your model size, GPU type and target numbers before ordering.

Get to know Arthur Wu

Arthur Wu
  • FromChina
  • Member sinceApr 2025
  • Languages

    English, Chinese
About Me: As an experienced AI consultant, I provide professional artificial intelligence solution services to help businesses and individuals optimize processes and drive innovation. I'm well-versed in machine learning, deep learning, natural language processing, and other AI technologies that I can apply to real business scenarios. My Services: AI strategy planning and implementation consulting Machine learning model development and deployment recommendations Automated process optimization design AI product feature and user experience design Data-driven decision analysis support