huangzhenpc/sub2api-ht
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

fix(openai-gateway): extract reasoning_effort in raw Chat Completions path

Browse Source 
The forwardAsRawChatCompletions path (used when APIKey accounts target
upstreams that don't support Responses API, e.g. DeepSeek) was missing
reasoning_effort and service_tier extraction, causing all reasoning
effort values to be silently dropped.

Extract both from the raw Chat Completions body before forwarding, and
propagate them through streamRawChatCompletions / bufferRawChatCompletions
to the OpenAIForwardResult.
This commit is contained in:
alfadb
2026-05-02 10:22:16 +08:00
parent adf01ac880
commit 57099a6af6
1 changed files with 29 additions and 17 deletions
Download Patch File Download Diff File Expand all files Collapse all files

46
backend/internal/service/openai_gateway_chat_completions_raw.go
View File

@@ -73,6 +73,10 @@ func (s *OpenAIGatewayService) forwardAsRawChatCompletions(
}
clientStream := gjson.GetBytes(body, "stream").Bool()
// 1b. Extract reasoning effort and service tier from the raw body before any transformation.
reasoningEffort := extractOpenAIReasoningEffortFromBody(body, originalModel)
serviceTier := extractOpenAIServiceTierFromBody(body)
// 2. Resolve model mapping (same as ForwardAsChatCompletions)
billingModel := resolveOpenAIForwardModel(account, originalModel, defaultMappedModel)
upstreamModel := normalizeOpenAIModelForUpstream(account, billingModel)
@@ -206,9 +210,9 @@ func (s *OpenAIGatewayService) forwardAsRawChatCompletions(
// 8. Forward response
if clientStream {
return s.streamRawChatCompletions(c, resp, originalModel, billingModel, upstreamModel, startTime)
return s.streamRawChatCompletions(c, resp, originalModel, billingModel, upstreamModel, reasoningEffort, serviceTier, startTime)
}
return s.bufferRawChatCompletions(c, resp, originalModel, billingModel, upstreamModel, startTime)
return s.bufferRawChatCompletions(c, resp, originalModel, billingModel, upstreamModel, reasoningEffort, serviceTier, startTime)
}
// streamRawChatCompletions 透传上游 CC SSE 流到客户端，并提取 usage（包括
@@ -223,6 +227,8 @@ func (s *OpenAIGatewayService) streamRawChatCompletions(
originalModel string,
billingModel string,
upstreamModel string,
reasoningEffort *string,
serviceTier *string,
startTime time.Time,
) (*OpenAIForwardResult, error) {
requestID := resp.Header.Get("x-request-id")
@@ -288,14 +294,16 @@ func (s *OpenAIGatewayService) streamRawChatCompletions(
}
return &OpenAIForwardResult{
RequestID: requestID,
Usage: usage,
Model: originalModel,
BillingModel: billingModel,
UpstreamModel: upstreamModel,
Stream: true,
Duration: time.Since(startTime),
FirstTokenMs: firstTokenMs,
RequestID: requestID,
Usage: usage,
Model: originalModel,
BillingModel: billingModel,
UpstreamModel: upstreamModel,
ReasoningEffort: reasoningEffort,
ServiceTier: serviceTier,
Stream: true,
Duration: time.Since(startTime),
FirstTokenMs: firstTokenMs,
}, nil
}
@@ -306,6 +314,8 @@ func (s *OpenAIGatewayService) bufferRawChatCompletions(
originalModel string,
billingModel string,
upstreamModel string,
reasoningEffort *string,
serviceTier *string,
startTime time.Time,
) (*OpenAIForwardResult, error) {
requestID := resp.Header.Get("x-request-id")
@@ -340,13 +350,15 @@ func (s *OpenAIGatewayService) bufferRawChatCompletions(
_, _ = c.Writer.Write(respBody)
return &OpenAIForwardResult{
RequestID: requestID,
Usage: usage,
Model: originalModel,
BillingModel: billingModel,
UpstreamModel: upstreamModel,
Stream: false,
Duration: time.Since(startTime),
RequestID: requestID,
Usage: usage,
Model: originalModel,
BillingModel: billingModel,
UpstreamModel: upstreamModel,
ReasoningEffort: reasoningEffort,
ServiceTier: serviceTier,
Stream: false,
Duration: time.Since(startTime),
}, nil
}