ext/new-api
2
0
Fork 0
mirror of https://github.com/QuantumNous/new-api.git synced 2026-09-01 19:41:57 +00:00
Code Activity

Merge f208026f08 into 2b6f1dfefb

Browse Source
This commit is contained in:
7. Sun
2026-08-31 01:55:41 +08:00
committed by GitHub
parent 2b6f1dfefb f208026f08
commit 7061cef406
1 changed files with 14 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files
service/text_quota.go
+14 -4
View File
@@ -247,10 +247,20 @@ func calculateTextQuotaSummary(ctx *gin.Context, relayInfo *relaycommon.RelayInf
summary.IsClaudeUsageSemantic = summary.UsageSemantic == "anthropic"
if usage == nil {
usage = &dto.Usage{
PromptTokens: relayInfo.GetEstimatePromptTokens(),
CompletionTokens: 0,
TotalTokens: relayInfo.GetEstimatePromptTokens(),
// When a stream ends abnormally (client disconnect, upstream timeout, etc.)
// before any completion tokens are produced, do not charge prompt tokens.
// The user received no output; billing with estimated prompt tokens is
// incorrect. The zero-TotalTokens guard below will set Quota=0.
// For non-stream requests (or normally-ended streams) with missing usage,
// fall back to estimated prompt tokens as before.
if relayInfo.IsStream && !relayInfo.StreamStatus.IsNormalEnd() && relayInfo.SendResponseCount == 0 {
usage = &dto.Usage{}
} else {
usage = &dto.Usage{
PromptTokens: relayInfo.GetEstimatePromptTokens(),
CompletionTokens: 0,
TotalTokens: relayInfo.GetEstimatePromptTokens(),
}
}
}