ai-fallback - npm Package Compare versions

Comparing version 0.0.6 to 0.1.0

dist/index.js

 export function createFallback(settings) {
     return new FallbackModel(settings);
 }
-// Default error retry logic
+const retryableStatusCodes = [
+    401,
+    403,
+    408,
+    409,
+    413,
+    429,
+    500, // server error (and above)
+];
 function defaultShouldRetryThisError(error) {
     let statusCode = error === null || error === void 0 ? void 0 : error['statusCode'];
     if (statusCode &&
-        (statusCode === 408 || // request timeout
-            statusCode === 409 || // conflict
-            statusCode === 429 || // too many requests
-            statusCode >= 500) // server error)
-    ) {
+        (retryableStatusCodes.includes(statusCode) || statusCode > 500)) {
         return true;
@@ -14,0 +18,0 @@ }

dist/index.test.js

@@ -12,2 +12,3 @@ var __asyncValues = (this && this.__asyncValues) || function (o) {
 import { createOpenAI } from '@ai-sdk/openai';
+import { createGroq } from '@ai-sdk/groq';
 import { createAnthropic } from '@ai-sdk/anthropic';
@@ -53,2 +54,22 @@ import { generateText, streamText, streamObject } from 'ai';
 });
+test('groq switches model on error, switches to third model', async () => {
+    const model = createFallback({
+        models: [
+            createGroq({ apiKey: 'wrong-key' })('gpt-3.5-turbo'),
+            createOpenAI({ apiKey: 'wrong-key' })('gpt-3.5-turbo'),
+            anthropic('claude-3-haiku-20240307'),
+        ],
+    });
+    model.currentModelIndex = 0;
+    const result = await generateText({
+        model,
+        system: 'You only respond hello',
+        messages: [{ role: 'user', content: 'Say hello' }],
+    });
+    // After error with OpenAI, should have switched to Anthropic
+    expect(model.currentModelIndex).toBe(2);
+    expect(model.modelId).toMatchInlineSnapshot(`"claude-3-haiku-20240307"`);
+    expect(result.text).toBeTruthy();
+    model.currentModelIndex = 0;
+});
 test('streamText works', async () => {
@@ -55,0 +76,0 @@ var _a, e_1, _b, _c;

package.json

 {
   "name": "ai-fallback",
-  "version": "0.0.6",
+  "version": "0.1.0",
   "description": "Automatically switch AI SDK model provider when one of the providers has downtime",
@@ -24,2 +24,3 @@ "type": "module",
     "@ai-sdk/anthropic": "^1.0.5",
+    "@ai-sdk/groq": "^1.0.9",
     "@ai-sdk/openai": "^1.0.8",
@@ -26,0 +27,0 @@ "ai": "^4.0.14",

src/index.test.ts

@@ -5,2 +5,3 @@ import { test, expect } from 'vitest'
 import { createOpenAI } from '@ai-sdk/openai'
+import { createGroq } from '@ai-sdk/groq'
 import { createAnthropic } from '@ai-sdk/anthropic'
@@ -56,2 +57,26 @@ import { generateText, streamText, streamObject } from 'ai'
 
+test('groq switches model on error, switches to third model', async () => {
+    const model = createFallback({
+        models: [
+            createGroq({ apiKey: 'wrong-key' })('gpt-3.5-turbo'),
+            createOpenAI({ apiKey: 'wrong-key' })('gpt-3.5-turbo'),
+            anthropic('claude-3-haiku-20240307'),
+        ],
+    })
+
+    model.currentModelIndex = 0
+
+    const result = await generateText({
+        model,
+        system: 'You only respond hello',
+        messages: [{ role: 'user', content: 'Say hello' }],
+    })
+
+    // After error with OpenAI, should have switched to Anthropic
+    expect(model.currentModelIndex).toBe(2)
+    expect(model.modelId).toMatchInlineSnapshot(`"claude-3-haiku-20240307"`)
+    expect(result.text).toBeTruthy()
+    model.currentModelIndex = 0
+})
+
 test('streamText works', async () => {
@@ -58,0 +83,0 @@ const model = createFallback({

src/index.ts

@@ -28,11 +28,18 @@ import {
 
-// Default error retry logic
+const retryableStatusCodes = [
+    401, // wrong API key
+    403, // permission error, like cannot access model or from a non accessible region
+    408, // request timeout
+    409, // conflict
+    413, // payload too large
+    429, // too many requests/rate limits
+    500, // server error (and above)
+]
+
 function defaultShouldRetryThisError(error: Error): boolean {
     let statusCode = error?.['statusCode']
+
     if (
         statusCode &&
-        (statusCode === 408 || // request timeout
-            statusCode === 409 || // conflict
-            statusCode === 429 || // too many requests
-            statusCode >= 500) // server error)
+        (retryableStatusCodes.includes(statusCode) || statusCode > 500)
     ) {
@@ -144,3 +151,2 @@ return true
                 lastError = error as Error
-
                 // Only retry if it's a server/capacity error
@@ -147,0 +153,0 @@ const shouldRetry =

No alert changes

Improved metrics

Total package byte prevSize

57419

4.74%

Lines of code

1016

5.28%

Worsened metrics

Dev dependency count

5

25%

No dependency changes