saidutta69/RaceBench
Viewer • Updated • 360k • 84 • 2
Agentic benchmarks by RACER IS OP — the RaceBench dataset and edge models fine-tuned on it
Note RaceBench — 100K+ row agentic benchmark covering coding, cybersecurity, tool-use and edge tasks.
Note MiniCPM5-1B fine-tuned on RaceBench — compact edge reasoning model.
Note RaceBench v1.1 - revised agentic benchmark.
Note MiniCPM5-1B fine-tuned on RaceBench v1.1 - compact edge reasoning model.
Note RaceBench-MiniCPM5 abliterated - the same edge model with refusals removed.
Note FC-Eval - function-calling benchmark built from real OpenAPI/Swagger specs.
Note PITU-Eval - prompt-injection tool-use eval: finish the task, resist the injected instruction.