9router/open-sse/translator/response/gemini-to-openai.js
decolua 39278e9613 refactor(open-sse): extract buildUsage helper, dedup token-details (B2)
Add helpers/usageHelper.js for conditional prompt/completion token details.
Apply to gemini/codex/claude response translators; keep each provider's token
math intact. No behavior change; golden + gate: no regression.

Co-authored-by: Cursor <cursoragent@cursor.com>
2026-06-13 17:18:15 +07:00

177 lines
6.5 KiB
JavaScript

import { register } from "../index.js";
import { FORMATS } from "../formats.js";
import { buildChunk } from "../helpers/chunkBuilder.js";
import { buildUsage } from "../helpers/usageHelper.js";
// Build chunk meta for current gemini state
function chunkMeta(state) {
return { id: `chatcmpl-${state.messageId}`, created: Math.floor(Date.now() / 1000), model: state.model };
}
// Convert Gemini response chunk to OpenAI format
export function geminiToOpenAIResponse(chunk, state) {
if (!chunk) return null;
// Handle Antigravity wrapper
const response = chunk.response || chunk;
if (!response || !response.candidates?.[0]) return null;
const results = [];
const candidate = response.candidates[0];
const content = candidate.content;
// Initialize state
if (!state.messageId) {
state.messageId = response.responseId || `msg_${Date.now()}`;
state.model = response.modelVersion || "gemini";
state.functionIndex = 0;
results.push(buildChunk(chunkMeta(state), { role: "assistant" }, null));
}
// Process parts
if (content?.parts) {
for (const part of content.parts) {
const hasThoughtSig = part.thoughtSignature || part.thought_signature;
const isThought = part.thought === true;
// Handle thought signature (thinking mode)
if (hasThoughtSig) {
const hasTextContent = part.text !== undefined && part.text !== "";
const hasFunctionCall = !!part.functionCall;
if (hasTextContent) {
results.push(buildChunk(
chunkMeta(state),
isThought ? { reasoning_content: part.text } : { content: part.text },
null
));
}
if (hasFunctionCall) {
const rawName = part.functionCall.name;
// Restore original tool name from mapping (AG cloaking)
const fcName = state.toolNameMap?.get(rawName) || rawName;
const fcArgs = part.functionCall.args || {};
const toolCallIndex = state.functionIndex++;
const toolCall = {
id: `${fcName}-${Date.now()}-${toolCallIndex}`,
index: toolCallIndex,
type: "function",
function: {
name: fcName,
arguments: JSON.stringify(fcArgs)
}
};
state.toolCalls.set(toolCallIndex, toolCall);
results.push(buildChunk(chunkMeta(state), { tool_calls: [toolCall] }, null));
}
continue;
}
// Text content. Gemini marks model-internal thinking with `thought: true`.
// Some responses include a thoughtSignature, but Google AI Studio/Gemini API
// can also stream thought parts without a signature; those must not be
// surfaced as normal assistant content in OpenAI-compatible clients.
if (part.text !== undefined && part.text !== "") {
results.push(buildChunk(
chunkMeta(state),
isThought ? { reasoning_content: part.text } : { content: part.text },
null
));
}
// Function call
if (part.functionCall) {
const rawName = part.functionCall.name;
// Restore original tool name from mapping (AG cloaking)
const fcName = state.toolNameMap?.get(rawName) || rawName;
const fcArgs = part.functionCall.args || {};
const toolCallIndex = state.functionIndex++;
const toolCall = {
id: `${fcName}-${Date.now()}-${toolCallIndex}`,
index: toolCallIndex,
type: "function",
function: {
name: fcName,
arguments: JSON.stringify(fcArgs)
}
};
state.toolCalls.set(toolCallIndex, toolCall);
results.push(buildChunk(chunkMeta(state), { tool_calls: [toolCall] }, null));
}
// Inline data (images)
const inlineData = part.inlineData || part.inline_data;
if (inlineData?.data) {
const mimeType = inlineData.mimeType || inlineData.mime_type || "image/png";
results.push(buildChunk(
chunkMeta(state),
{
images: [{
type: "image_url",
image_url: { url: `data:${mimeType};base64,${inlineData.data}` }
}]
},
null
));
}
}
}
// Usage metadata - extract before finish reason so we can include it
const usageMeta = response.usageMetadata || chunk.usageMetadata;
if (usageMeta && typeof usageMeta === "object") {
const cachedTokens = typeof usageMeta.cachedContentTokenCount === "number" ? usageMeta.cachedContentTokenCount : 0;
const promptTokenCountRaw = typeof usageMeta.promptTokenCount === "number" ? usageMeta.promptTokenCount : 0;
const thoughtsTokens = typeof usageMeta.thoughtsTokenCount === "number" ? usageMeta.thoughtsTokenCount : 0;
let candidatesTokens = typeof usageMeta.candidatesTokenCount === "number" ? usageMeta.candidatesTokenCount : 0;
const totalTokens = typeof usageMeta.totalTokenCount === "number" ? usageMeta.totalTokenCount : 0;
// prompt_tokens = promptTokenCount (includes cached tokens, matching claude-to-openai.js behavior)
const promptTokens = promptTokenCountRaw;
// Fallback calculation if candidatesTokenCount is 0 but totalTokenCount exists
if (candidatesTokens === 0 && totalTokens > 0) {
candidatesTokens = totalTokens - promptTokenCountRaw - thoughtsTokens;
if (candidatesTokens < 0) candidatesTokens = 0;
}
// completion_tokens = candidatesTokenCount + thoughtsTokenCount (match Go code)
const completionTokens = candidatesTokens + thoughtsTokens;
state.usage = buildUsage({ promptTokens, completionTokens, totalTokens, cachedTokens, reasoningTokens: thoughtsTokens });
}
// Finish reason - include usage in final chunk
if (candidate.finishReason) {
let finishReason = candidate.finishReason.toLowerCase();
if (finishReason === "stop" && state.toolCalls.size > 0) {
finishReason = "tool_calls";
}
const finalChunk = buildChunk(chunkMeta(state), {}, finishReason);
// Include usage in final chunk for downstream translators
if (state.usage) {
finalChunk.usage = state.usage;
}
results.push(finalChunk);
state.finishReason = finishReason;
}
return results.length > 0 ? results : null;
}
// Register
register(FORMATS.GEMINI, FORMATS.OPENAI, null, geminiToOpenAIResponse);
register(FORMATS.GEMINI_CLI, FORMATS.OPENAI, null, geminiToOpenAIResponse);
register(FORMATS.ANTIGRAVITY, FORMATS.OPENAI, null, geminiToOpenAIResponse);
register(FORMATS.VERTEX, FORMATS.OPENAI, null, geminiToOpenAIResponse);