common.ts
5.8 KB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
import { NextRequest, NextResponse } from "next/server";
import { getServerSideConfig } from "../config/server";
import { OPENAI_BASE_URL, ServiceProvider } from "../constant";
import { cloudflareAIGatewayUrl } from "../utils/cloudflare";
import { getModelProvider, isModelNotavailableInServer } from "../utils/model";
const serverConfig = getServerSideConfig();
export async function requestOpenai(req: NextRequest) {
const controller = new AbortController();
const isAzure = req.nextUrl.pathname.includes("azure/deployments");
var authValue,
authHeaderName = "";
if (isAzure) {
authValue =
req.headers
.get("Authorization")
?.trim()
.replaceAll("Bearer ", "")
.trim() ?? "";
authHeaderName = "api-key";
} else {
authValue = req.headers.get("Authorization") ?? "";
authHeaderName = "Authorization";
}
let path = `${req.nextUrl.pathname}`.replaceAll("/api/openai/", "");
let baseUrl =
(isAzure ? serverConfig.azureUrl : serverConfig.baseUrl) || OPENAI_BASE_URL;
if (!baseUrl.startsWith("http")) {
baseUrl = `https://${baseUrl}`;
}
if (baseUrl.endsWith("/")) {
baseUrl = baseUrl.slice(0, -1);
}
console.log("[Proxy] ", path);
console.log("[Base Url]", baseUrl);
const timeoutId = setTimeout(
() => {
controller.abort();
},
10 * 60 * 1000,
);
if (isAzure) {
const azureApiVersion =
req?.nextUrl?.searchParams?.get("api-version") ||
serverConfig.azureApiVersion;
baseUrl = baseUrl.split("/deployments").shift() as string;
path = `${req.nextUrl.pathname.replaceAll(
"/api/azure/",
"",
)}?api-version=${azureApiVersion}`;
// Forward compatibility:
// if display_name(deployment_name) not set, and '{deploy-id}' in AZURE_URL
// then using default '{deploy-id}'
if (serverConfig.customModels && serverConfig.azureUrl) {
const modelName = path.split("/")[1];
let realDeployName = "";
serverConfig.customModels
.split(",")
.filter((v) => !!v && !v.startsWith("-") && v.includes(modelName))
.forEach((m) => {
const [fullName, displayName] = m.split("=");
const [_, providerName] = getModelProvider(fullName);
if (providerName === "azure" && !displayName) {
const [_, deployId] = (serverConfig?.azureUrl ?? "").split(
"deployments/",
);
if (deployId) {
realDeployName = deployId;
}
}
});
if (realDeployName) {
console.log("[Replace with DeployId", realDeployName);
path = path.replaceAll(modelName, realDeployName);
}
}
}
const fetchUrl = cloudflareAIGatewayUrl(`${baseUrl}/${path}`);
console.log("fetchUrl", fetchUrl);
const fetchOptions: RequestInit = {
headers: {
"Content-Type": "application/json",
"Cache-Control": "no-store",
[authHeaderName]: authValue,
...(serverConfig.openaiOrgId && {
"OpenAI-Organization": serverConfig.openaiOrgId,
}),
},
method: req.method,
body: req.body,
// to fix #2485: https://stackoverflow.com/questions/55920957/cloudflare-worker-typeerror-one-time-use-body
redirect: "manual",
// @ts-ignore
duplex: "half",
signal: controller.signal,
};
// #1815 try to refuse gpt4 request
if (serverConfig.customModels && req.body) {
try {
const clonedBody = await req.text();
fetchOptions.body = clonedBody;
const jsonBody = JSON.parse(clonedBody) as { model?: string };
// not undefined and is false
if (
isModelNotavailableInServer(
serverConfig.customModels,
jsonBody?.model as string,
[
ServiceProvider.OpenAI,
ServiceProvider.Azure,
jsonBody?.model as string, // support provider-unspecified model
],
)
) {
return NextResponse.json(
{
error: true,
message: `you are not allowed to use ${jsonBody?.model} model`,
},
{
status: 403,
},
);
}
} catch (e) {
console.error("[OpenAI] gpt4 filter", e);
}
}
try {
const res = await fetch(fetchUrl, fetchOptions);
// Extract the OpenAI-Organization header from the response
const openaiOrganizationHeader = res.headers.get("OpenAI-Organization");
// Check if serverConfig.openaiOrgId is defined and not an empty string
if (serverConfig.openaiOrgId && serverConfig.openaiOrgId.trim() !== "") {
// If openaiOrganizationHeader is present, log it; otherwise, log that the header is not present
console.log("[Org ID]", openaiOrganizationHeader);
} else {
console.log("[Org ID] is not set up.");
}
// to prevent browser prompt for credentials
const newHeaders = new Headers(res.headers);
newHeaders.delete("www-authenticate");
// to disable nginx buffering
newHeaders.set("X-Accel-Buffering", "no");
// Conditionally delete the OpenAI-Organization header from the response if [Org ID] is undefined or empty (not setup in ENV)
// Also, this is to prevent the header from being sent to the client
if (!serverConfig.openaiOrgId || serverConfig.openaiOrgId.trim() === "") {
newHeaders.delete("OpenAI-Organization");
}
// The latest version of the OpenAI API forced the content-encoding to be "br" in json response
// So if the streaming is disabled, we need to remove the content-encoding header
// Because Vercel uses gzip to compress the response, if we don't remove the content-encoding header
// The browser will try to decode the response with brotli and fail
newHeaders.delete("content-encoding");
return new Response(res.body, {
status: res.status,
statusText: res.statusText,
headers: newHeaders,
});
} finally {
clearTimeout(timeoutId);
}
}