github/claude-relay-service
1
0
Fork 0
mirror of https://github.com/Wei-Shaw/claude-relay-service.git synced 2026-01-23 09:38:02 +00:00
Code Issues Packages Projects Releases Wiki Activity

feat: gemini 流式响应

Browse Source
This commit is contained in:
mouyong
2025-08-04 20:34:17 +08:00
parent 2eee902988
commit 54ad8452c3
6 changed files with 100 additions and 1283 deletions
Download Patch File Download Diff File Expand all files Collapse all files

113
src/routes/openaiGeminiRoutes.js
View File

@@ -122,13 +122,16 @@ function convertGeminiResponseToOpenAI(geminiResponse, model, stream = false) {
return geminiResponse;
} else {
// 非流式响应转换
if (geminiResponse.candidates && geminiResponse.candidates.length > 0) {
const candidate = geminiResponse.candidates[0];
// 处理嵌套的 response 结构
const actualResponse = geminiResponse.response || geminiResponse;
if (actualResponse.candidates && actualResponse.candidates.length > 0) {
const candidate = actualResponse.candidates[0];
const content = candidate.content?.parts?.[0]?.text || '';
const finishReason = candidate.finishReason?.toLowerCase() || 'stop';
// 计算 token 使用量
const usage = geminiResponse.usageMetadata || {
const usage = actualResponse.usageMetadata || {
promptTokenCount: 0,
candidatesTokenCount: 0,
totalTokenCount: 0
@@ -189,7 +192,7 @@ router.post('/v1/chat/completions', authenticateApiKey, async (req, res) => {
// 从 URL 路径中提取模型信息（如果存在）
let urlModel = null;
const urlPath = req.body?.config?.url || req.originalUrl || req.url;
const modelMatch = urlPath.match(/\/([^\/]+):(?:stream)?[Gg]enerateContent/);
const modelMatch = urlPath.match(/\/([^/]+):(?:stream)?[Gg]enerateContent/);
if (modelMatch) {
urlModel = modelMatch[1];
logger.debug(`Extracted model from URL: ${urlModel}`);
@@ -204,6 +207,10 @@ router.post('/v1/chat/completions', authenticateApiKey, async (req, res) => {
max_tokens = 4096,
stream = false
} = requestBody;
// 检查URL中是否包含stream标识
const isStreamFromUrl = urlPath && urlPath.includes('streamGenerateContent');
const actualStream = stream || isStreamFromUrl;
// 优先使用 URL 中的模型，其次是请求体中的模型
const model = urlModel || bodyModel;
@@ -292,9 +299,7 @@ router.post('/v1/chat/completions', authenticateApiKey, async (req, res) => {
// 获取OAuth客户端
const client = await geminiAccountService.getOauthClient(account.accessToken, account.refreshToken);
let project = 'verdant-wares-464411-k9';
if (stream) {
if (actualStream) {
// 流式响应
logger.info('StreamGenerateContent request', {
model: model,
@@ -306,7 +311,7 @@ router.post('/v1/chat/completions', authenticateApiKey, async (req, res) => {
client,
{ model, request: geminiRequestBody },
null, // user_prompt_id
project || account.projectId,
account.projectId, // 使用有权限的项目ID
apiKeyData.id, // 使用 API Key ID 作为 session ID
abortController.signal // 传递中止信号
);
@@ -320,9 +325,29 @@ router.post('/v1/chat/completions', authenticateApiKey, async (req, res) => {
// 处理流式响应，转换为 OpenAI 格式
let buffer = '';
// 发送初始的空消息，符合 OpenAI 流式格式
const initialChunk = {
id: `chatcmpl-${Date.now()}`,
object: 'chat.completion.chunk',
created: Math.floor(Date.now() / 1000),
model: model,
choices: [{
index: 0,
delta: { role: 'assistant' },
finish_reason: null
}]
};
res.write(`data: ${JSON.stringify(initialChunk)}\n\n`);
streamResponse.on('data', (chunk) => {
try {
buffer += chunk.toString();
const chunkStr = chunk.toString();
if (!chunkStr.trim()) {
return;
}
buffer += chunkStr;
const lines = buffer.split('\n');
buffer = lines.pop() || ''; // 保留最后一个不完整的行
@@ -341,30 +366,51 @@ router.post('/v1/chat/completions', authenticateApiKey, async (req, res) => {
const data = JSON.parse(jsonData);
// 转换为 OpenAI 流式格式
if (data.candidates && data.candidates.length > 0) {
const candidate = data.candidates[0];
if (data.response?.candidates && data.response.candidates.length > 0) {
const candidate = data.response.candidates[0];
const content = candidate.content?.parts?.[0]?.text || '';
const finishReason = candidate.finishReason?.toLowerCase();
const finishReason = candidate.finishReason;
const openaiChunk = {
id: `chatcmpl-${Date.now()}`,
object: 'chat.completion.chunk',
created: Math.floor(Date.now() / 1000),
model: model,
choices: [{
index: 0,
delta: {
content: content
},
finish_reason: finishReason === 'stop' ? 'stop' : null
}]
};
res.write(`data: ${JSON.stringify(openaiChunk)}\n\n`);
// 如果结束了，发送最终的 [DONE]
if (finishReason === 'stop') {
res.write('data: [DONE]\n\n');
// 只有当有内容或者是结束标记时才发送数据
if (content || finishReason === 'STOP') {
const openaiChunk = {
id: `chatcmpl-${Date.now()}`,
object: 'chat.completion.chunk',
created: Math.floor(Date.now() / 1000),
model: model,
choices: [{
index: 0,
delta: content ? { content: content } : {},
finish_reason: finishReason === 'STOP' ? 'stop' : null
}]
};
res.write(`data: ${JSON.stringify(openaiChunk)}\n\n`);
// 如果结束了，添加 usage 信息并发送最终的 [DONE]
if (finishReason === 'STOP') {
// 如果有 usage 数据，添加到最后一个 chunk
if (data.response.usageMetadata) {
const usageChunk = {
id: `chatcmpl-${Date.now()}`,
object: 'chat.completion.chunk',
created: Math.floor(Date.now() / 1000),
model: model,
choices: [{
index: 0,
delta: {},
finish_reason: 'stop'
}],
usage: {
prompt_tokens: data.response.usageMetadata.promptTokenCount || 0,
completion_tokens: data.response.usageMetadata.candidatesTokenCount || 0,
total_tokens: data.response.usageMetadata.totalTokenCount || 0
}
};
res.write(`data: ${JSON.stringify(usageChunk)}\n\n`);
}
res.write('data: [DONE]\n\n');
}
}
}
} catch (e) {
@@ -402,6 +448,9 @@ router.post('/v1/chat/completions', authenticateApiKey, async (req, res) => {
}
});
} else {
// 如果已经开始发送流数据，发送错误事件
res.write(`data: {"error": {"message": "${error.message || 'Stream error'}"}}\n\n`);
res.write('data: [DONE]\n\n');
res.end();
}
});
@@ -418,7 +467,7 @@ router.post('/v1/chat/completions', authenticateApiKey, async (req, res) => {
client,
{ model, request: geminiRequestBody },
null, // user_prompt_id
account.projectId,
account.projectId, // 使用有权限的项目ID
apiKeyData.id // 使用 API Key ID 作为 session ID
);