Developer Guide

This section is for developers who want to contribute to vLLM Ascend or understand its internal architecture.

Contribution

• Contribution Guide — How to contribute to vLLM Ascend
• Testing — Write and run unit, E2E, and nightly tests
• Doc Writing — Documentation contribution guide
• Multi-Node Test — Multi-node testing guide
• Nightly CI Test — Nightly CI testing
• E2E CI Test — E2E CI testing

Design Documents

Explore the design documents covering patch architecture, CPU binding, model runner internals, disaggregated prefill, EPLB, ACL Graph, KV Cache Pool, custom operators, context parallel, quantization, and NPUGraph.

Evaluation

• Using EvalScope — Model evaluation with EvalScope
• Using lm_eval — Model evaluation with lm_eval
• Using AISBench — Model evaluation with AISBench
• Using OpenCompass — Model evaluation with OpenCompass

Performance and Debug

• Performance Benchmark — Benchmarking guide
• Optimization and Tuning — Performance optimization
• Service Profiling Guide — Service profiling
• msprobe Guide — Debugging with msprobe