vLLM per-position accepted-token counter
vllm · counter
Fields
id
metric-exposures/w3m-vllm-vllm-spec-decode-num-accepted-tokens-per-posMetric name
vllm:spec_decode_num_accepted_tokens_per_pos
Metric type
counter
Unit
tokens
Exposure
prometheus
Endpoint
:8000/metrics
Labels
model_nameengineposition
Enabled by default
yes
notes
vllm/v1/spec_decode/metrics.py line 255, main, read 2026-10-04. One series per draft position, so cardinality scales with the number of speculative tokens. Omitted entirely on the diffusion (dLLM) path. Not on by default - only exists with a speculative_config.
Notes
vllm/v1/spec_decode/metrics.py line 255, main, read 2026-10-04. One series per draft position, so cardinality scales with the number of speculative tokens. Omitted entirely on the diffusion (dLLM) path. Not on by default - only exists with a speculative_config.