vLLM
Official@vllm-project
High-throughput serving infrastructure for deploying diffusion, text-to-speech, and neural processing unit models across distributed compute environments.
Agent Skills by vLLM
Showing 31 vetted skills indexed across 3 GitHub repositories.
vllm-projectvllm-ascend-release
Automates the end-to-end release management workflow for vLLM Ascend on GitHub.
vllm-projectproject-change
Implements focused repository changes when no specialized workflow applies.
vllm-projectvllm-omni-test
Generates and runs pytest tests for vllm-omni with CI-aligned levels, markers, and Buildkite wiring.
vllm-projectprecheck-pr
Reviews branch diffs against vllm-omni contribution checklists before opening a pull request.
vllm-projectquantization
Configure and validate quantization methods for vLLM-Omni diffusion, autoregressive, and omni models.
vllm-projectfind-simplifications
Audit vLLM-Omni code for evidence-backed simplification and deletion candidates.
vllm-projectdiffusion-perf-opt
Diagnose and optimize vLLM Omni diffusion workloads using profiler traces and parallel strategy analysis.
vllm-projectreview-pr
Review vLLM-Omni pull requests with frozen snapshots, module contracts, and evidence-backed findings.
vllm-projectconfig-platform-change
Synchronize config concepts across router config, Python CLI schema, and dashboard UI.
vllm-projectsignal-end-to-end
Coordinate end-to-end signal changes across router configuration, extraction, CLI schema, and metadata contracts.
vllm-projectdashboard-platform-change
Orchestrate dashboard frontend and backend changes to align with router contracts.
vllm-projectrouting-policy-change
Modify post-signal routing policies for predicates, model ranking, and looper behavior.
vllm-projectstartup-chain-change
Configure and validate the local vLLM Semantic Router startup chain.
vllm-projectharness-contract-change
Coordinate agent-harness contract changes across docs, manifests, and validation scripts.
vllm-projectcross-stack-bugfix
Diagnoses and fixes bugs spanning multiple layers with coordinated changes.
vllm-projecttraining-stack-change
Coordinate training stack changes across workflows, artifacts, and runtime outputs.
vllm-projectplugin-end-to-end
Coordinate plugin lifecycle changes across router config, processing, and E2E tests.
vllm-projectfleet-sim-change
Coordinate fleet-sim code, docs, and release workflow changes to prevent drift.
vllm-projectk8s-platform-change
Coordinate Kubernetes platform changes by updating operator APIs, CRDs, and deployment profiles.
vllm-projectrouter-service-platform-change
Modify router-side API, authz, memory, provider, storage, or runtime service modules with automated validation and traceability.
vllm-projectmaintainer-issue-pr-management
Automate GitHub issue and PR lifecycle management with canonical templates and labels.
vllm-projectrouting-calibration-loop
Calibrate routing changes against live router endpoints with executable probes and versioned deploys.
vllm-projectmaintainer-release-ops
Automate release planning and milestone management across GitHub issues and PRs.
vllm-projectvllm-semantic-router-harness
Resolve primary skills and run harness validation in the vLLM Semantic Router repository.
Frequently Asked Questions About vLLM
FAQPage SchemaWhat specific tasks can I perform using vLLM's infrastructure?▼
You can deploy diffusion and text-to-speech models, calibrate semantic routing policies, and manage model adaptation for NPU hardware. The framework supports multi-GPU parallelism, Cache-DiT acceleration, and end-to-end signal coordination across router configurations and deployment profiles.
Which technical personas benefit from these capabilities?▼
Infrastructure engineers, machine learning platform developers, and site reliability engineers benefit from these capabilities. These personas use the framework to manage model serving lifecycles, synchronize cross-stack configuration changes, and validate performance across distributed compute clusters.
What are the prerequisites for deploying models on this infrastructure?▼
Deployment requires a configured environment supporting CUDA or Ascend NPU runtimes. Users must define model manifests, align router-side service modules, and ensure Kubernetes CRDs are updated to match the specific deployment profile and memory requirements of the target model.