[Doc][Metrics] Fix spec-decode PromQL examples to use the exposed names (#55016)
Signed-off-by: Michele Campi <michele.campi@outlook.com>
Michele Campi committed
Sep 14, 2026 at 23:15 UTC
d392ac836e17e87f5fec6594ef5ee5b56db72a91
1 file changed
+3
-3
vllm/v1/spec_decode/metrics.py
+3
-3
@@ -187,12 +187,12 @@ class SpecDecodingProm:
187
188
1 + (
189
rate(vllm:spec_decode_num_accepted_tokens_total[$interval]) /
190
- rate(vllm:spec_decode_num_drafts[$interval]))
190
+ rate(vllm:spec_decode_num_drafts_total[$interval]))
191
192
A per-position acceptance rate vector can be computed using
193
194
- vllm:spec_decode_num_accepted_tokens_per_pos[$interval] /
195
- vllm:spec_decode_num_drafts[$interval]
194
+ vllm:spec_decode_num_accepted_tokens_per_pos_total[$interval] /
195
+ vllm:spec_decode_num_drafts_total[$interval]
196
"""
197
198
_counter_cls = prometheus_client.Counter