xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2026-05-31 03:27:04 +08:00
Code Issues Packages Projects Releases Wiki Activity

[V1][Minor] Enhance SpecDecoding Metrics Log in V1 (#15902)

Browse Source 
Signed-off-by: Woosuk Kwon <woosuk.kwon@berkeley.edu>
This commit is contained in:
Woosuk Kwon 2025-04-01 23:38:02 -07:00 committed by GitHub
parent 2039c6305b
commit 274d8e8818
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 10 additions and 7 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

17
vllm/v1/spec_decode/metrics.py
View File

@ -47,13 +47,16 @@ class SpecDecodingMetrics:
num_draft_tokens = np.sum(self.num_draft_tokens) num_draft_tokens = np.sum(self.num_draft_tokens)
num_accepted_tokens = np.sum(self.num_accepted_tokens) num_accepted_tokens = np.sum(self.num_accepted_tokens)
draft_acceptance_rate = (num_accepted_tokens / num_draft_tokens draft_acceptance_rate = (num_accepted_tokens / num_draft_tokens *
if num_draft_tokens > 0 else float("nan")) 100 if num_draft_tokens > 0 else float("nan"))
logger.info( logger.info(
"Speculative metrics: " "SpecDecoding metrics: "
"Draft acceptance rate: %.3f, " "Draft acceptance rate: %.1f%%, "
"Number of accepted tokens: %d, " "Accepted: %d tokens, "
"Number of draft tokens: %d, ", draft_acceptance_rate, "Drafted: %d tokens",
num_accepted_tokens, num_draft_tokens) draft_acceptance_rate,
num_accepted_tokens,
num_draft_tokens,
)
self.reset() self.reset()