Model server contract

The contract defines a common name, instrument type, unit, and meaning for each model-serving metric. The table below lists every signal currently defined in pkg/contract/signals.go, with the native metric names used by the bundled engine profiles.

Contract metric	Type	Unit	Semantics	vLLM	SGLang
gen_ai.server.request.duration	Histogram	s (seconds)	Request duration until the last byte or last output token.	vllm:e2e_request_latency_seconds	sglang:e2e_request_latency_seconds
gen_ai.server.time_to_first_token	Histogram	s (seconds)	Time to generate the first token for a successful response, including queue and prefill time.	vllm:time_to_first_token_seconds	sglang:time_to_first_token_seconds
gen_ai.server.time_per_output_token	Histogram	s (seconds)	Average time per output token after the first token, recorded per successful request.	vllm:request_time_per_output_token_seconds	sglang:request_time_per_output_token_seconds
gen_ai.server.request.queued	Gauge	{request}	Number of requests admitted by the server but not yet scheduled into a batch.	vllm:num_requests_waiting	sglang:num_queue_reqs
gen_ai.server.request.running	Gauge	{request}	Number of requests executing forward passes.	vllm:num_requests_running	sglang:num_running_reqs