# No token counters — fall back to vLLM's avg throughput gauges.
vllm:num_requests_running{model_name="facebook/opt-125m"} 1.0
vllm:num_requests_waiting{model_name="facebook/opt-125m"} 0.0
vllm:kv_cache_usage_perc{model_name="facebook/opt-125m"} 0.05
vllm:avg_prompt_throughput_toks_per_s{model_name="facebook/opt-125m"} 40.0
vllm:avg_generation_throughput_toks_per_s{model_name="facebook/opt-125m"} 120.0
