vllm-compile
πPluginTorchedHat/ai-marketplace
Expert guidance for vLLM's custom compiler - @support_torch_compile decorator, vllmBackend/Inductor passes, PiecewiseBackend graph splitting, and CudaGraphWrapper CUDA graph management
Part of
TorchedHat/ai-marketplace
Installation
/plugin marketplace add TorchedHat/ai-marketplace/plugin install vllm-compile@ai-marketplaceMore from this repository6
Marketplace
PyTorch torch.compile debugging tools and workflow automation
Tools for creating and managing Claude Code plugins, skills, and agents
Deterministic hook system for automatic skill activation and guardrails with skill-developer documentation
Audit vLLM tests for coincidentally-correct batch invariance assumptions. Explore to find suspicious tests, review to classify via formal decision tree.
MCP server for conversational PyTorch codebase analysis: trace Python->C++->CUDA binding chains, impact analysis, dispatch mapping, call-graph visualization, and test discovery across the PyTorch source.