[Feature] Claude-3 is not working properly with the OpenAI-compatible API.

[MorningInsight3]

**Claude-3 is not working properly with the OpenAI-compatible API.

Hello, I've used a proxy to make Claude-3 compatible with the OpenAI API, and it works normally in Next-Chat, Lobe-Chat, and Bob, but it doesn't work in your Morphic. Can you share the reason for this?**

Error code from vercel LOG as follows:
AI_APICallError: Invalid JSON response
at (node_modules/ai/openai/dist/index.mjs:312:0)
at (node_modules/ai/openai/dist/index.mjs:197:0)
at (node_modules/ai/openai/dist/index.mjs:578:0)
at (node_modules/ai/dist/index.mjs:405:0)
at (node_modules/ai/dist/index.mjs:538:0)
at (lib/agents/task-manager.tsx:13:17)
at (app/action.tsx:43:24) {
name: 'AI_APICallError',
url: 'https://likegpt.xx.xx.xx/api/v1/chat/completions',
requestBodyValues: {
model: 'anthropic.claude-3-sonnet-20240229-v1:0',
logit_bias: undefined,
user: undefined,
max_tokens: undefined,
temperature: undefined,
top_p: undefined,
frequency_penalty: undefined,
presence_penalty: undefined,
seed: undefined,
messages: [
{
role: 'system',
content: 'As a professional web researcher, your primary objective is to fully comprehend the user's query, conduct thorough web searches to gather the necessary information, and provide an appropriate response.\n To achieve this, you must first analyze the user's input and determine the optimal course of action. You have two options at your disposal:\n 1. "proceed": If the provided information is sufficient to address the query effectively, choose this option to proceed with the research and formulate a response.\n 2. "inquire": If you believe that additional information from the user would enhance your ability to provide a comprehensive response, select this option. You may present a form to the user, offering default selections or free-form input fields, to gather the required details.\n Your decision should be based on a careful assessment of the context and the potential for further information to improve the quality and relevance of your response.\n For example, if the user asks, "What are the key features of the latest iPhone model?", you may choose to "proceed" as the query is clear and can be answered effectively with web research alone.\n However, if the user asks, "What's the best smartphone for my needs?", you may opt to "inquire" and present a form asking about their specific requirements, budget, and preferred features to provide a more tailored recommendation.\n Make your choice wisely to ensure that you fulfill your mission as a web researcher effectively and deliver the most valuable assistance to the user.\n '
},
{
role: 'user',
content: [
{
type: 'text',
text: '{"input":"Is the Apple Vision Pro worth buying?"}'
}
]
}
],
tool_choice: { type: 'function', function: { name: 'json' } },
tools: [
{
type: 'function',
function: {
name: 'json',
description: 'Respond with a JSON object.',
parameters: {
type: 'object',
properties: { next: { type: 'string', enum: [ 'inquire', 'proceed' ] } },
required: [ 'next' ],
additionalProperties: false,
$schema: 'http://json-schema.org/draft-07/schema#'
}
}
}
]
},
statusCode: 200,
responseBody: '{"id":"msg_01RJEqLhDqDqriF5XktcB8Vc","created":1713013011,"model":"anthropic.claude-3-sonnet-20240229-v1:0","system_fingerprint":"fp","choices":[{"index":0,"finish_reason":"tool_calls","message":{"role":"assistant","tool_calls":[{"id":"a1e851c7","type":"function","function":{"name":"json","arguments":"{\"next\": \"inquire\"}"}}]}}],"object":"chat.completion","usage":{"prompt_tokens":546,"completion_tokens":44,"total_tokens":590}}',
cause: AI_TypeValidationError: Type validation failed: Value: {"id":"msg_01RJEqLhDqDqriF5XktcB8Vc","created":1713013011,"model":"anthropic.claude-3-sonnet-20240229-v1:0","system_fingerprint":"fp","choices":[{"index":0,"finish_reason":"tool_calls","message":{"role":"assistant","tool_calls":[{"id":"a1e851c7","type":"function","function":{"name":"json","arguments":"{"next": "inquire"}"}}]}}],"object":"chat.completion","usage":{"prompt_tokens":546,"completion_tokens":44,"total_tokens":590}}.
Error message: [
{
"code": "invalid_type",
"expected": "string",
"received": "undefined",
"path": [
"choices",
0,
"message",
"content"
],
"message": "Required"
}
]
at (node_modules/ai/openai/dist/index.mjs:73:0)
at (node_modules/ai/openai/dist/index.mjs:116:28)
at (vc/edge/function:2

[MorningInsight3]

[miurla]

We cannot support parts that require tool calling and JSON output, but I 've made it possible to specify OpenAI-compatible API for the final output. Please give it a try.

#58

[MorningInsight3]

Thank you for your reply. I quickly went through it again, and there seems to be another error:

[MorningInsight3]

AI_NoTextGeneratedError: No text generated.
at (node_modules/ai/dist/index.mjs:480:0)
at (lib/agents/task-manager.tsx:14:19)
at (app/action.tsx:49:24) {
name: 'AI_NoTextGeneratedError'
}

[miurla]

at (lib/agents/task-manager.tsx:14:19)

An error is occurring in the task manager.
This solution cannot be applied to all APIs.
Please delete specifications such as the OPENAI_BASE_URL, then try the following solution.

#58 (comment)

[MorningInsight3]

Thank you for your reply. I quickly went through it again, and there seems to be another error:

AI_LoadAPIKeyError: OpenAI API key is missing. Pass it using the 'apiKey' parameter or the OPENAI_API_KEY environment variable. at (node_modules/@ai-sdk/provider-utils/dist/index.mjs:45:0) at (node_modules/@ai-sdk/openai/dist/index.mjs:766:43) at (node_modules/@ai-sdk/openai/dist/index.mjs:245:0) at (node_modules/ai/dist/index.mjs:462:0) at (node_modules/ai/dist/index.mjs:328:0) at (node_modules/ai/dist/index.mjs:317:6) at (node_modules/ai/dist/index.mjs:461:0) at (lib/agents/task-manager.tsx:14:25) at (app/action.tsx:49:31) at (app/action.tsx:121:2) { name: 'AI_LoadAPIKeyError' }

WITH upgraded Environment Variables:

[superlbr]

AI_NoTextGeneratedError: No text generated. just hang whole task, how about add a timeout if not work

[denispol]

I always get this response when I query using Claude 3 Haiku:

Here is the activity log from the API endpoint provider: