File size: 9,395 Bytes
bce29b9 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 |
import type { getProviderHelper } from "../lib/getProviderHelper.js";
import { makeRequestOptions } from "../lib/makeRequestOptions.js";
import type { InferenceTask, Options, RequestArgs } from "../types.js";
import type { EventSourceMessage } from "../vendor/fetch-event-source/parse.js";
import { getLines, getMessages } from "../vendor/fetch-event-source/parse.js";
import { InferenceClientProviderApiError } from "../errors.js";
import type { JsonObject } from "../vendor/type-fest/basic.js";
export interface ResponseWrapper<T> {
data: T;
requestContext: {
url: string;
info: RequestInit;
};
}
function requestArgsToJson(args: RequestArgs): JsonObject {
// Convert the entire args object to a JSON-serializable format
const argsWithData = args as RequestArgs & { data?: Blob | ArrayBuffer };
return JSON.parse(
JSON.stringify({
...argsWithData,
data: argsWithData.data ? "[Blob or ArrayBuffer]" : null,
})
) as JsonObject;
}
/**
* Primitive to make custom calls to the inference provider
*/
export async function innerRequest<T>(
args: RequestArgs,
providerHelper: ReturnType<typeof getProviderHelper>,
options?: Options & {
/** In most cases (unless we pass a endpointUrl) we know the task */
task?: InferenceTask;
/** Is chat completion compatible */
chatCompletion?: boolean;
}
): Promise<ResponseWrapper<T>> {
const { url, info } = await makeRequestOptions(args, providerHelper, options);
const response = await (options?.fetch ?? fetch)(url, info);
const requestContext: ResponseWrapper<T>["requestContext"] = { url, info };
if (options?.retry_on_error !== false && response.status === 503) {
return innerRequest(args, providerHelper, options);
}
if (!response.ok) {
const contentType = response.headers.get("Content-Type");
if (["application/json", "application/problem+json"].some((ct) => contentType?.startsWith(ct))) {
const output = await response.json();
if ([400, 422, 404, 500].includes(response.status) && options?.chatCompletion) {
throw new InferenceClientProviderApiError(
`Provider ${args.provider} does not seem to support chat completion for model ${
args.model
} . Error: ${JSON.stringify(output.error)}`,
{
url,
method: info.method ?? "GET",
headers: info.headers as Record<string, string>,
body: requestArgsToJson(args),
},
{ requestId: response.headers.get("x-request-id") ?? "", status: response.status, body: output }
);
}
if (typeof output.error === "string" || typeof output.detail === "string" || typeof output.message === "string") {
throw new InferenceClientProviderApiError(
`Failed to perform inference: ${output.error ?? output.detail ?? output.message}`,
{
url,
method: info.method ?? "GET",
headers: info.headers as Record<string, string>,
body: requestArgsToJson(args),
},
{ requestId: response.headers.get("x-request-id") ?? "", status: response.status, body: output }
);
} else {
throw new InferenceClientProviderApiError(
`Failed to perform inference: an HTTP error occurred when requesting the provider.`,
{
url,
method: info.method ?? "GET",
headers: info.headers as Record<string, string>,
body: requestArgsToJson(args),
},
{ requestId: response.headers.get("x-request-id") ?? "", status: response.status, body: output }
);
}
}
const message = contentType?.startsWith("text/plain;") ? await response.text() : undefined;
throw new InferenceClientProviderApiError(
`Failed to perform inference: ${message ?? "an HTTP error occurred when requesting the provider"}`,
{
url,
method: info.method ?? "GET",
headers: info.headers as Record<string, string>,
body: requestArgsToJson(args),
},
{ requestId: response.headers.get("x-request-id") ?? "", status: response.status, body: message ?? "" }
);
}
if (response.headers.get("Content-Type")?.startsWith("application/json")) {
const data = (await response.json()) as T;
return { data, requestContext };
}
const blob = (await response.blob()) as T;
return { data: blob as unknown as T, requestContext };
}
/**
* Primitive to make custom inference calls that expect server-sent events, and returns the response through a generator
*/
export async function* innerStreamingRequest<T>(
args: RequestArgs,
providerHelper: ReturnType<typeof getProviderHelper>,
options?: Options & {
/** In most cases (unless we pass a endpointUrl) we know the task */
task?: InferenceTask;
/** Is chat completion compatible */
chatCompletion?: boolean;
}
): AsyncGenerator<T> {
const { url, info } = await makeRequestOptions({ ...args, stream: true }, providerHelper, options);
const response = await (options?.fetch ?? fetch)(url, info);
if (options?.retry_on_error !== false && response.status === 503) {
return yield* innerStreamingRequest(args, providerHelper, options);
}
if (!response.ok) {
if (response.headers.get("Content-Type")?.startsWith("application/json")) {
const output = await response.json();
if ([400, 422, 404, 500].includes(response.status) && options?.chatCompletion) {
throw new InferenceClientProviderApiError(
`Provider ${args.provider} does not seem to support chat completion for model ${
args.model
} . Error: ${JSON.stringify(output.error)}`,
{
url,
method: info.method ?? "GET",
headers: info.headers as Record<string, string>,
body: requestArgsToJson(args),
},
{ requestId: response.headers.get("x-request-id") ?? "", status: response.status, body: output }
);
}
if (typeof output.error === "string") {
throw new InferenceClientProviderApiError(
`Failed to perform inference: ${output.error}`,
{
url,
method: info.method ?? "GET",
headers: info.headers as Record<string, string>,
body: requestArgsToJson(args),
},
{ requestId: response.headers.get("x-request-id") ?? "", status: response.status, body: output }
);
}
if (output.error && "message" in output.error && typeof output.error.message === "string") {
/// OpenAI errors
throw new InferenceClientProviderApiError(
`Failed to perform inference: ${output.error.message}`,
{
url,
method: info.method ?? "GET",
headers: info.headers as Record<string, string>,
body: requestArgsToJson(args),
},
{ requestId: response.headers.get("x-request-id") ?? "", status: response.status, body: output }
);
}
// Sambanova errors
if (typeof output.message === "string") {
throw new InferenceClientProviderApiError(
`Failed to perform inference: ${output.message}`,
{
url,
method: info.method ?? "GET",
headers: info.headers as Record<string, string>,
body: requestArgsToJson(args),
},
{ requestId: response.headers.get("x-request-id") ?? "", status: response.status, body: output }
);
}
}
throw new InferenceClientProviderApiError(
`Failed to perform inference: an HTTP error occurred when requesting the provider.`,
{
url,
method: info.method ?? "GET",
headers: info.headers as Record<string, string>,
body: requestArgsToJson(args),
},
{ requestId: response.headers.get("x-request-id") ?? "", status: response.status, body: "" }
);
}
if (!response.headers.get("content-type")?.startsWith("text/event-stream")) {
throw new InferenceClientProviderApiError(
`Failed to perform inference: server does not support event stream content type, it returned ` +
response.headers.get("content-type"),
{
url,
method: info.method ?? "GET",
headers: info.headers as Record<string, string>,
body: requestArgsToJson(args),
},
{ requestId: response.headers.get("x-request-id") ?? "", status: response.status, body: "" }
);
}
if (!response.body) {
return;
}
const reader = response.body.getReader();
let events: EventSourceMessage[] = [];
const onEvent = (event: EventSourceMessage) => {
// accumulate events in array
events.push(event);
};
const onChunk = getLines(
getMessages(
() => {},
() => {},
onEvent
)
);
try {
while (true) {
const { done, value } = await reader.read();
if (done) {
return;
}
onChunk(value);
for (const event of events) {
if (event.data.length > 0) {
if (event.data === "[DONE]") {
return;
}
const data = JSON.parse(event.data);
if (typeof data === "object" && data !== null && "error" in data) {
const errorStr =
typeof data.error === "string"
? data.error
: typeof data.error === "object" &&
data.error &&
"message" in data.error &&
typeof data.error.message === "string"
? data.error.message
: JSON.stringify(data.error);
throw new InferenceClientProviderApiError(
`Failed to perform inference: an occurred while streaming the response: ${errorStr}`,
{
url,
method: info.method ?? "GET",
headers: info.headers as Record<string, string>,
body: requestArgsToJson(args),
},
{ requestId: response.headers.get("x-request-id") ?? "", status: response.status, body: data }
);
}
yield data as T;
}
}
events = [];
}
} finally {
reader.releaseLock();
}
}
|