Release notes from tool-eval-bench
Articles about Release notes from tool-eval-bench
Release v2.8.0 · SeraphimSerapis/tool-eval-bench
Highlights Hard Mode grows to 28 scenarios, for 97 total with --hardmode. TC-89 tests compensation after partial success, TC-90 to TC-92 cover governance, and TC-93 to TC-97 cover document workflows. The default run still contains 69 scenarios. Reports now break Hard Mode down by capability.
Release v2.7.0 · SeraphimSerapis/tool-eval-bench
Added NInfer backend detection: tool-eval-bench now recognizes the NInfer inference engine and labels it correctly in reports (backend: ninfer, Engine: NInfer) instead of mis-detecting it as llama.cpp. NInfer serves /v1/models with owned_by == "ninfer", which is now probed before the generic /health fallback that previously produced the false llama.cpp label. ninfer is also an accepted value for --backend.