Skill · AI Hiring Index · as of 2026-11-19
vLLM / SGLang / TensorRT in AI job postings
3% of technical postings at the AI companies we track mention vLLM / SGLang / TensorRT (143 postings). Inference / Perf postings ask for it most: 29%.
Which roles ask for it
Share of each technical role family's open postings that mention vLLM / SGLang / TensorRT.
| Role family | Share, as a bar | Share |
|---|---|---|
| Inference / Perf | 29% | |
| AI / ML Engineer | 5% | |
| Research Engineer | 5% | |
| Research Scientist | 5% | |
| FDE / Applied | 4% | |
| Infra / Hardware | 1% | |
| Software Eng | 1% | |
| Evals / Data | 0% | |
| Safety / Policy | 0% | |
| Security | 0% |
Companies asking for it most
| Company | Postings | Of its technical postings |
|---|---|---|
| Nebius | 22 | 10% |
| CoreWeave | 18 | 11% |
| SambaNova | 14 | 27% |
| Baseten | 9 | 19% |
| Crusoe | 6 | 4% |
| Cerebras | 6 | 6% |
| Fireworks AI | 6 | 16% |
| Together AI | 5 | 12% |
| Mistral AI | 5 | 3% |
| Modal | 4 | 17% |
How to learn it
- LLM Inference Handbook — Modular free
Latency, throughput and cost of serving a model — batching, KV cache, quantization, which engine when — as one practical handbook rather than twenty blog posts.
- Inside vLLM: Anatomy of a High-Throughput LLM Inference System free
One request followed through the scheduler, paged KV cache, continuous batching, prefix caching and speculative decoding, ending on the latency-against-throughput curve a design answer needs.
- vLLM documentation free
The engine the cloud-GPU route runs on: the docs explain the trade-off behind each setting, and the bundled vllm bench serve prints TTFT, TPOT and throughput — the deployment project's numbers.