inference/research
verifiedhigh · 0.94updated 2026-10-05

TEI router/src/prometheus.rs - metric bucket construction for the text-embeddings-inference router

repo

Fields

id
sources/w5e-tei-prometheus-rs
Publisher
huggingface
Kind
repo
Accessed
2026-10-05
notes
Read at main 2026-10-05. 47 lines. Confirms the Suffix("duration") matcher from the prior pass: Matcher::Suffix(String::from("duration")) catches every te_*_duration histogram with one geometric ladder - 35 buckets starting at 1e-5 and multiplied by 1.5 each step. Note the first push happens AFTER the multiply, so the first bound is 1.5e-5, not 1e-5. Other matchers are Full("te_request_input_length") (powers of two filtered to max_input_length), Full("te_batch_next_size") (13 bounds, 2^0..2^12) and Full("te_batch_next_tokens") (21 bounds, 2^0..2^20).

Notes

Read at main 2026-10-05. 47 lines. Confirms the Suffix("duration") matcher from the prior pass: Matcher::Suffix(String::from("duration")) catches every te_*_duration histogram with one geometric ladder - 35 buckets starting at 1e-5 and multiplied by 1.5 each step. Note the first push happens AFTER the multiply, so the first bound is 1.5e-5, not 1e-5. Other matchers are Full("te_request_input_length") (powers of two filtered to max_input_length), Full("te_batch_next_size") (13 bounds, 2^0..2^12) and Full("te_batch_next_tokens") (21 bounds, 2^0..2^20).

Cross-references

This record references no other records.