Release v0.6.1

What's Changed

• otel_trace_replay: add bad_tool_call_handling=use_recorded for malformed tool_call args (#535) by @animeshtrivedi in #538
• Emit Prometheus merics for runtime observability by @Bslabe123 in #501
• Use server-reported completion_tokens for output length by @czhu-cohere in #577
• Parallelize e2e tests with pytest-xdist by @Bslabe123 in #585
• Add per-request distribution to server-sourced token metrics by @Bslabe123 in #581
• Update OWNERS_ALIASES by @achandrasekar in #503
• Update OWNERS and OWNERS_ALIASES by @jjk-g in #603
• [Testing] fix output_tokens assertion missed by #581's distribution change by @Bslabe123 in #602
• chore: respect use_server_output_tokens for non-streaming + metrics by @czhu-cohere in #607
• fix: normalize 'developer' message role to 'system' for non-OpenAI model servers by @pavanipenumalla in #573
• fix: pass reasoning_content through to server for KV-cache fidelity by @oritht in #579
• Feature: Add per-stage error metrics report with session and request breakdowns by @shaigrt in #601
• fix: inject session_id_header for conversation_replay and shared_prefix workloads by @lenadankin in #625
• fix: correct stale table print-count assertions in test_cli_summary by @shaigrt in #651
• feat(otel): load trace sessions on demand, evict after completion by @lenadankin in #595
• Add disable_output_substitution flag and fix tool-call replay fidelity by @alonh in #588
• E2e test: assert metrics fidelity against simulated ground truth by @Bslabe123 in #614
• E2e test: add sim-backed golden accuracy with controlled token counts and chunk timing by @Bslabe123 in #645
• chore: handle null prompt_tokens_details by @czhu-cohere in #664
• Add Anthropic Messages API support by @goutamadwant in #575
• Add /ok-to-test command to approve held PR workflow runs by @jjk-g in #657
• ci: gate tide merges on CI via do-not-merge label by @jjk-g in #653

New Contributors

• @animeshtrivedi made their first contribution in #538
• @czhu-cohere made their first contribution in #577
• @shaigrt made their first contribution in #601
• @goutamadwant made their first contribution in #575

Full Changelog: v0.6.0...v0.6.1

Docker Image

quay.io/inference-perf/inference-perf:v0.6.1

Python Package

pip install inference-perf==v0.6.1

Helm Chart

helm install inference-perf oci://quay.io/inference-perf/charts/inference-perf