InferLens

Engine-neutral observability for LLM inference engines. Record what vLLM (and soon SGLang) is actually doing internally — scheduling, batching, KV-cache behavior — and explore it as an interactive timeline that answers "why was this request slow?"

Status: pre-alpha. The trace schema and vLLM collector are under active development. Nothing here is stable yet — including the trace format.

Why

Today's tooling for inference engines is bimodal: Grafana dashboards of aggregate metrics (too high-level to explain a single slow request) and kernel-level profiler traces (too low-level to see scheduler semantics). InferLens targets the missing middle layer:

• per-step scheduler state: queue depths, batch composition, preemptions
• KV-cache behavior: usage, prefix-cache hits, evictions
• per-request lifecycle: queued → prefill → decode, with token timings

…captured with a pip-installed plugin (no engine fork), written to a portable trace file, and rendered in a local viewer.

How it will work

vLLM ──(stat-logger plugin + KV events)──►  trace file (.ilens.gz)  ──►  inferlens view
SGLang ──(collector, planned)───────────►       one shared schema        interactive timeline

pip install inferlens
inferlens record -o run.ilens.gz -- vllm serve Qwen/Qwen2.5-1.5B-Instruct   # in development
inferlens info run.ilens.gz
inferlens view run.ilens.gz                                                  # in development

Development

uv sync
uv run pre-commit install
uv run pytest

See CONTRIBUTING.md for coding standards (Google style, DCO sign-off, vLLM-style PR conventions) and docs/TRACE_SPEC.md for the trace format.

License

Apache-2.0