vllm-tbccl

A vLLM platform integration that carries device-group communication and pipeline activations over torch-tbccl.

Documentation

• Getting started
  • Installing vllm-tbccl
  • Quickstart
• Guides
  • Pipeline parallelism over TBCCL
  • CUDA and Metal pairing
  • Troubleshooting
• Concepts
  • Architecture
  • Security model
• Reference
  • Configuration
  • Compatibility
  • Python API
  • License and third-party software
• Development
  • Testing
  • Building the documentation
  • Release checklist
  • Security policy
  • Changelog
  • Documentation versions
• Architecture decision records
  • ADR 0001: Extend vLLM and vllm-metal without patching them in place
  • ADR template: short title
• Validation
  • vllm-tbccl 0.2.0 validation (draft)
• Related projects