v0.3.0

What's Changed

• refactor: remove sim-dependent tests and adapters by @hlin99 in #269
• docs: unified structure — LICENSE, README, CONTRIBUTING, bot/ by @hlin99 in #270
• docs: upgrade bot policies to proxy-level strict standard by @hlin99 in #271

Full Changelog: v0.2.0...v0.3.0

v0.2.0

xPyD-bench v0.2.0

Breaking Change: Dummy Server → xpyd-sim Adapter

The built-in dummy server has been replaced by a thin adapter to xpyd-sim. Bench now uses the real simulator instead of maintaining its own 1000+ line implementation.

• New dependency: xpyd-sim>=0.4.0
• New module: xpyd_bench/sim_adapter/ — clean re-export from xpyd-sim
• Backward compat: from xpyd_bench.dummy.server import ServerConfig, create_app still works via shim
• Net -963 lines of duplicated code removed

Other Changes (from v0.1.0)

• Flat layout: src/xpyd_bench/ → xpyd_bench/
• Dummy server cleanup: removed 6 non-vLLM features
• vLLM response compatibility: stop_reason, service_tier fields
• DESIGN_PRINCIPLES.md: vLLM Boundary Rule + API Compatibility Levels

Testing

All 1830 tests pass. Zero tests deleted.

v0.1.0

xPyD-bench v0.1.0 — Initial Release

Benchmarking & PD ratio planning tool for xPyD-proxy.

Features

• Full vLLM bench CLI compatibility
• OpenAI API endpoint support: /v1/completions, /v1/chat/completions, /v1/embeddings
• Configurable request rate patterns (constant, burst, ramp, poisson, custom)
• Rich metrics: TTFT, TPOT, ITL, P50/P90/P95/P99 latency, throughput
• Multiple output formats: JSON, CSV, Markdown, HTML, JUnit XML, Prometheus
• Benchmark comparison & regression detection
• Concurrency sweep, multi-endpoint, multi-model comparison
• SLA validation, warmup profiling, adaptive concurrency
• Distributed benchmark coordination
• 100+ milestone features (M1-M97 complete)

Structure

• Flat layout: xpyd_bench/ at repo root
• Dummy server follows vLLM API boundary rule
• 1830 tests passing