From 07290e18da93fcbec494c6fe8743ff7c7e4e8ad3 Mon Sep 17 00:00:00 2001 From: Luis Pater Date: Thu, 12 Nov 2026 16:08:54 +0800 Subject: [PATCH] fix(translator): map upstream MAX_TOKENS to OpenAI finish_reason length - Map upstream MAX_TOKENS to finish_reason "length" in the Antigravity and Gemini OpenAI Chat Completions translators (streaming and non-streaming), since "max_tokens" is not a valid OpenAI finish_reason and strict clients reject it. - Keep native_finish_reason as the lowercased upstream value and preserve tool_calls precedence and content_filter handling, updating and adding regression test cases for each affected path. Closes: #7463 Co-authored-by: Dylan <1990016@gmail.com> Co-authored-by: lzt404 <2596933790@qq.com> --- .../chat-completions/antigravity_openai_response.go | 2 +- .../antigravity_openai_response_test.go | 13 +++++++++---- .../chat-completions/gemini_openai_response.go | 6 ++++-- .../chat-completions/gemini_openai_response_test.go | 6 ++++++ 4 files changed, 20 insertions(+), 7 deletions(-) diff --git a/internal/translator/antigravity/openai/chat-completions/antigravity_openai_response.go b/internal/translator/antigravity/openai/chat-completions/antigravity_openai_response.go index a0f5559d4f..f27d8aa103 100644 --- a/internal/translator/antigravity/openai/chat-completions/antigravity_openai_response.go +++ b/internal/translator/antigravity/openai/chat-completions/antigravity_openai_response.go @@ -287,7 +287,7 @@ func resolveOpenAIFinishReason(params *convertCliResponseToOpenAIChatParams) (fi case params.SawToolCall: finishReason = "tool_calls" case params.UpstreamFinishReason == "MAX_TOKENS": - finishReason = "max_tokens" + finishReason = "length" case isAntigravityContentFilterFinishReason(params.UpstreamFinishReason): finishReason = "content_filter" default: diff --git a/internal/translator/antigravity/openai/chat-completions/antigravity_openai_response_test.go b/internal/translator/antigravity/openai/chat-completions/antigravity_openai_response_test.go index 855b569df0..bf80f68b1d 100644 --- a/internal/translator/antigravity/openai/chat-completions/antigravity_openai_response_test.go +++ b/internal/translator/antigravity/openai/chat-completions/antigravity_openai_response_test.go @@ -77,10 +77,15 @@ func TestFinishReasonMaxTokens(t *testing.T) { chunk2 := []byte(`{"response":{"candidates":[{"finishReason":"MAX_TOKENS"}],"usageMetadata":{"promptTokenCount":10,"candidatesTokenCount":100,"totalTokenCount":110}}}`) result2 := ConvertAntigravityResponseToOpenAI(ctx, "model", nil, nil, chunk2, ¶m) - // Verify finish_reason is "max_tokens" + // Verify finish_reason is "length" (the OpenAI value for an output token limit) fr := ojson.GetBytesNoCopy(result2[0], "choices.0.finish_reason").String() - if fr != "max_tokens" { - t.Errorf("Expected finish_reason 'max_tokens', got: %s", fr) + if fr != "length" { + t.Errorf("Expected finish_reason 'length', got: %s", fr) + } + // Verify native_finish_reason preserves the upstream value + nfr := ojson.GetBytesNoCopy(result2[0], "choices.0.native_finish_reason").String() + if nfr != "max_tokens" { + t.Errorf("Expected native_finish_reason 'max_tokens', got: %s", nfr) } } @@ -96,7 +101,7 @@ func TestToolCallTakesPriorityOverMaxTokens(t *testing.T) { chunk2 := []byte(`{"response":{"candidates":[{"finishReason":"MAX_TOKENS"}],"usageMetadata":{"promptTokenCount":10,"candidatesTokenCount":100,"totalTokenCount":110}}}`) result2 := ConvertAntigravityResponseToOpenAI(ctx, "model", nil, nil, chunk2, ¶m) - // Verify finish_reason is "tool_calls" (takes priority over max_tokens) + // Verify finish_reason is "tool_calls" (takes priority over upstream MAX_TOKENS) fr := ojson.GetBytesNoCopy(result2[0], "choices.0.finish_reason").String() if fr != "tool_calls" { t.Errorf("Expected finish_reason 'tool_calls', got: %s", fr) diff --git a/internal/translator/gemini/openai/chat-completions/gemini_openai_response.go b/internal/translator/gemini/openai/chat-completions/gemini_openai_response.go index 24063d6bdb..7c503fa25c 100644 --- a/internal/translator/gemini/openai/chat-completions/gemini_openai_response.go +++ b/internal/translator/gemini/openai/chat-completions/gemini_openai_response.go @@ -513,7 +513,9 @@ func ConvertGeminiResponseToOpenAINonStream(_ context.Context, _ string, origina // Set finish reason. if finishReasonResult := candidate.Get("finishReason"); finishReasonResult.Exists() { finishReason := strings.ToLower(finishReasonResult.String()) - if isGeminiContentFilterFinishReason(finishReasonResult.String()) { + if strings.EqualFold(strings.TrimSpace(finishReasonResult.String()), "MAX_TOKENS") { + finishReason = "length" + } else if isGeminiContentFilterFinishReason(finishReasonResult.String()) { finishReason = "content_filter" } _ = choiceDoc.Set("finish_reason", finishReason) @@ -639,7 +641,7 @@ func resolveGeminiOpenAIFinishReason(upstreamFinishReason string, sawToolCall bo case sawToolCall: finishReason = "tool_calls" case upstreamFinishReason == "MAX_TOKENS": - finishReason = "max_tokens" + finishReason = "length" case isGeminiContentFilterFinishReason(upstreamFinishReason): finishReason = "content_filter" default: diff --git a/internal/translator/gemini/openai/chat-completions/gemini_openai_response_test.go b/internal/translator/gemini/openai/chat-completions/gemini_openai_response_test.go index 211f72df39..da0a6239ba 100644 --- a/internal/translator/gemini/openai/chat-completions/gemini_openai_response_test.go +++ b/internal/translator/gemini/openai/chat-completions/gemini_openai_response_test.go @@ -198,6 +198,12 @@ func TestConvertGeminiResponseToOpenAI_NonStopFinishReasonsMapToContentFilter(t wantFinishReason: "content_filter", wantNativeReason: "prohibited_content", }, + { + name: "MAX_TOKENS", + upstreamReason: "MAX_TOKENS", + wantFinishReason: "length", + wantNativeReason: "max_tokens", + }, } for _, tc := range cases {