Merge pull request #2798 from RedwindA/feat/GeminiCacheBilling

feat(gemini): support cached token billing
This commit is contained in:
Seefs
2026-02-02 13:21:10 +08:00
committed by GitHub
4 changed files with 12 additions and 5 deletions

View File

@@ -49,6 +49,7 @@ func GeminiTextGenerationHandler(c *gin.Context, info *relaycommon.RelayInfo, re
}
usage.CompletionTokenDetails.ReasoningTokens = geminiResponse.UsageMetadata.ThoughtsTokenCount
usage.PromptTokensDetails.CachedTokens = geminiResponse.UsageMetadata.CachedContentTokenCount
for _, detail := range geminiResponse.UsageMetadata.PromptTokensDetails {
if detail.Modality == "AUDIO" {

View File

@@ -1251,6 +1251,7 @@ func geminiStreamHandler(c *gin.Context, info *relaycommon.RelayInfo, resp *http
usage.CompletionTokens = geminiResponse.UsageMetadata.CandidatesTokenCount + geminiResponse.UsageMetadata.ThoughtsTokenCount
usage.CompletionTokenDetails.ReasoningTokens = geminiResponse.UsageMetadata.ThoughtsTokenCount
usage.TotalTokens = geminiResponse.UsageMetadata.TotalTokenCount
usage.PromptTokensDetails.CachedTokens = geminiResponse.UsageMetadata.CachedContentTokenCount
for _, detail := range geminiResponse.UsageMetadata.PromptTokensDetails {
if detail.Modality == "AUDIO" {
usage.PromptTokensDetails.AudioTokens = detail.TokenCount
@@ -1395,6 +1396,7 @@ func GeminiChatHandler(c *gin.Context, info *relaycommon.RelayInfo, resp *http.R
PromptTokens: geminiResponse.UsageMetadata.PromptTokenCount,
}
usage.CompletionTokenDetails.ReasoningTokens = geminiResponse.UsageMetadata.ThoughtsTokenCount
usage.PromptTokensDetails.CachedTokens = geminiResponse.UsageMetadata.CachedContentTokenCount
for _, detail := range geminiResponse.UsageMetadata.PromptTokensDetails {
if detail.Modality == "AUDIO" {
usage.PromptTokensDetails.AudioTokens = detail.TokenCount
@@ -1447,6 +1449,7 @@ func GeminiChatHandler(c *gin.Context, info *relaycommon.RelayInfo, resp *http.R
}
usage.CompletionTokenDetails.ReasoningTokens = geminiResponse.UsageMetadata.ThoughtsTokenCount
usage.PromptTokensDetails.CachedTokens = geminiResponse.UsageMetadata.CachedContentTokenCount
usage.CompletionTokens = usage.TotalTokens - usage.PromptTokens
for _, detail := range geminiResponse.UsageMetadata.PromptTokensDetails {