Alibaba Cloud Model Studio:FAQ

Billing related

1. What are the unit prices for model services on Alibaba Cloud Model Studio?

   For the pricing of models, see Billable items and pricing.

2. Are there any subscription services available?

   No. Model Studio currently offers no subscription services.

3. Is the pay-as-you-go bill settled monthly or daily?

   Billing occurs on an hourly basis.

4. How can I check the billing items and details?

   To view your bills, visit Expenses and Costs.

5. How can I apply for an invoice for the amount already spent?

   Go to the Overview of Monthly Bill page. Click Download Invoice in the Operations column.

6. What do I do if a message appears to inform me that the available balance of my account is less than zero when I activate Model Studio?

   Before activating the service, make sure your Alibaba Cloud account balance is not less than zero.

Permission related

1. How can I assigning management permissions for Model Studio to a RAM User?

   To assign management permissions to a RAM user, grant the AliyunSFMFullAccess permission. For more information, see Authorize a RAM user to manage services in Model Studio.

2. Can I use the API key of a RAM user after I delete the RAM user?

   No. The API key becomes invalid once the associated RAM user is deleted.

3. How can I create an API key?

   1. Log on to Alibaba Cloud Model Studio

   2. Hover your cursor over in the upper-right corner of the page and select API-KEY.

      

   3. Click My API Key.

4. Can I activate the Model Studio services as a RAM user?

   You must first use the Alibaba Cloud account to assign the AliyunSFMFullAccess permission to the RAM user. Then, you can use the RAM user to activate the services.

5. Does the AdministratorAccess permission include the AliyunSFMFullAccess permission?

   No, you must grant the RAM user the AliyunSFMFullAccess permission separately to access Model Studio services.

6. I have granted the AliyunSFMFullAccess permission to my RAM user. Why is the RAM user still not able to enter the Model Studio console?

   If a RAM user cannot enter the Model Studio console, you need to use the Alibaba Cloud account to assign a role to the RAM user. For more information, see Authorize a RAM user to manage services in Model Studio.

API/SDK related

1. How to view error code information?

   An API call to Alibaba Cloud Model Studio returns a status code that indicates the result of the call. The following table describes the general status codes that are defined by Alibaba Cloud Model Studio. In addition to the general status codes, models may add custom status codes and messages. You can identify a call result based on the code and status parameters.

   HTTP status code	Code	Message (Specific messages may vary in different scenarios)	Description
   400	InvalidParameter	Required parameter(s) missing or invalid, please check the request parameters.	Request parameters are invalid.
   400	InvalidParameter	Either \"prompt\" or \"messages\" must exist and cannot both be none	The prompt and messages parameters cannot be empty at the same time. The error may be caused by wrong format. For example, If you use non-OpenAI-compatible HTTP to call services, the messages parameter must be inside the input, not in parallel with the model parameter. { "model": "Specified Model", "input": { "messages": [ {"role": "user", "content": "Who are you"}, ] }, "parameters": { "result_format": "message" } }
   400	InvalidParameter	File [id:file-fe-***********] format is not supported.	The Qwen-Long model can process only pure text files, including TXT, DOCX, PDF, PDF, EPUB, MOBI, and MD. The model does not support images or scanned documents. If you want to extract text from images for analysis and summary, you can use Qwen-VL.
   400	DataInspectionFailed	Input or output data may contain inappropriate content. Input data may contain inappropriate content. Output data may contain inappropriate content.	The input or output data may contain inappropriate content and is blocked during content moderation.
   400	BadRequest.EmptyInput	Required input parameter missing from request.	The input cannot be empty.
   400	BadRequest.EmptyParameters	Required parameter "parameters" missing from request.	Request parameters cannot be empty.
   400	BadRequest.EmptyModel	Required parameter "model" missing from request.	The request model cannot be empty.
   400	InvalidURL	Invalid URL provided in your request.	The request URL is incorrect.
   400	Arrearage	Access denied, please make sure your account is in good standing.	Possible cause 1: Your Alibaba Cloud account has overdue payment. Go to Expenses and Costs to check overdue payments. Possible cause 2: If the error occurs when you call the Batch interface, you have not activated the Alibaba Cloud Model Studio service. Go to Models to activate the service for free and use an API Key of Model Studio to call the interface.
   400	UnsupportedOperation	The operation is unsupported on the referee object.	The referee object does not support this operation. This message may vary based on the actual scenario.
   400	FlowNotPublished	Flow has not published yet, please publish flow and try again.	The flow is not published. Publish the flow and try again.
   400	InvalidSchema	Database schema is invalid for text2sql.	Enter the schema of your database.
   400	InvalidSchemaFormat	Database schema format is invalid for text2sql.	The schema format of the input data table is invalid.
   400	FaqRuleBlocked	Input or output data is blocked by faq rule.	The input or output is blocked by FAQ rules.
   400	CustomRoleBlocked	Input or output data may contain inappropriate content with custom rule.	The input or output is blocked by custom rules.
   400	InternalError.Algo	Missing Content-Length of multimodal url.	The response header information for the URL request lacks the Content-Length field. If the problem persists, try other image links instead. View Content-Length Open your browser (such as Chrome or Firefox). Open the Developer Tools (usually by pressing F12 or right-clicking and selecting Inspect). Switch to the Network tab. Enter the image URL into the address bar and access it. Locate the corresponding request and check the Headers section. In the Response Headers part, find the Content-Length field.
   401	InvalidApiKey	Invalid API-key provided.	Possible cause 1: The input API key is invalid. Possible cause 2: You are using https://dashscope.aliyuncs.com as the base URL. Change it to https://dashscope-intl.aliyuncs.com and try again.
   403	AccessDenied	Access denied.	You are not authorized to access this API. For example, you are not in the invitational preview. Go to Models and apply for the desired model first.
   403	Workspace.AccessDenied	Workspace access denied.	You are not authorized to access the applications or models in this workspace. Check whether you specified the valid workspace ID. If you want to call a model in a sub-workspace, see Call a model in a sub-workspace. You can also use the API Key of the Alibaba Cloud account. The account has permissions of all workspaces.
   403	Model.AccessDenied	Model access denied.	The Resource Access Management (RAM) user is not authorized to access models in this workspace. Perform the following steps: Check whether you have specified the ID of a sub-workspace. RAM users cannot access applications or models in the default workspace. Check whether the RAM user is authorized to access the sub-workspace. Check whether the sub-workspace is authorized to call the model. For more information, see the Prerequisites section of the Call a model in a sub-workspace topic.
   403	AccessDenied.Unpurchased	Access to model denied. Please make sure you are eligible for using the model.	You may have not activated Alibaba Cloud Model Studio. You need to register or log in to your account and then go to Models to activate model service.
   404	WorkSpaceNotFound	WorkSpace can not be found.	The workspace that you specified does not exist.
   404	ModelNotFound	Model can not be found.	The model that you specified does not exist.
   404	ModelNotFound	The model xx does not exist or you do not have access to it.	You have not activated the Alibaba Cloud Model Studio service. Go to Models to activate model service.
   408	RequestTimeOut	Request timed out, please try again later. Response timeout!	The request timed out. You can try again. The request may time out because the response is too long. You can try the streaming output mode when making requests.
   413	BadRequest.TooLarge	Payload Too Large.	The gateway at the access layer returns an error that the request body is too large. If the error is returned by the Microservice Engine (MSE) gateway, no code is returned and the message cannot be customized. If the error is returned by the RESTful gateway, a code is returned.
   415	BadRequest.InputDownloadFailed	Failed to download the input file: xxx.	Failed to download the input file, which may be due to download timeout, download failure, or file size exceeding the limit. The message may include additional details.
   415	BadRequest.UnsupportedFileFormat	Input file format is not supported.	The format of the input file is not supported.
   429	Throttling	Requests throttling triggered.	The API call triggers throttling.
   429	Throttling.RateQuota	Requests rate limit exceeded, please try again later.	The frequency of calls triggers throttling, such as the number of requests per second.
   429	Throttling.AllocationQuota	Allocated quota exceeded, please increase your quota limit.	The number of calls triggers throttling, such as the number of tokens generated per minute.
   429	LimitRequests	You exceeded your current requests list	The request throttling is triggered. You can make requests again after falling down below the throttling threshold.
   429	Throttling.AllocationQuota	Free allocated quota exceeded.	Free quota is exhausted and billable access of the model is not enabled.
   429	PrepaidBillOverdue	The prepaid bill is overdue.	Subscription of this workspace is overdue.
   429	PostpaidBillOverdue	The postpaid bill is overdue.	The model inference service is overdue.
   429	CommodityNotPurchased	Commodity has not purchased yet.	The service is not activated in this workspace.
   500	InternalError	An internal error has occured, please try again later or contact service support.	An internal error occurred.
   500	InternalError.Algo	An internal error has occured during execution, please try again later or contact service support.	An internal algorithm error occurred.
   500	SystemError	An system error has occured, please try again later.	A system error occurred.
   500	InternalError.Timeout	An internal timeout error has occured during execution, please try again later or contact service support.	An asynchronous task waits for 3 hours after it was submitted from the gateway to the algorithm service layer. If no result is returned, the task times out.
   500	RewriteFailed	Failed to rewrite content for prompt.	Prompt rewriting failed.
   500	RetrivalFailed	Failed to retrieve data from documents.	Document retrieval failed.
   500	AppProcessFailed	Failed to proceed application request.	Application flow processing failed.
   500	ModelServiceFailed	Failed to request model service.	Model service call failed.
   500	InvokePluginFailed	Failed to invoke plugin.	Plug-in call failed.
   503	ModelUnavailable	Model is unavailable, please try again later.	The model is temporarily unavailable for service.
   	NetworkError	Can not find api-key.	The environment variable configuration does not take effect, You can try restarting the client or IDE.

2. How to install the SDK?

   Model Studio supports SDKs for Java and Python, see Install the SDK.

3. Does the Assistant API support function calls where users need to call two local functions in sequence?

   Assistant API does not support calling two functions separately. But you can manually create two assistants and allocate the returns of the two assistants separately.

4. Does the Assistant API have memory-related capabilities?

   No. Memory is a relatively complex feature, and implementing it in the Assistant API is complex.

Product related

1. How to activate Alibaba Cloud Model Studio services?

   Log in to your Alibaba Cloud account and go to the Alibaba Cloud Model Studio console. You can click to activate model services on the home page or the Models page as needed. For more information, see: Activate Alibaba Cloud Model Studio.

2. How to deactivate Alibaba Cloud Model Studio services?

   Once activated, Alibaba Cloud Model Studio services cannot be deactivated. To prevent further API calls, simply delete the API-KEY on the API-KEY page within the console.

3. How to experience model services?

   Experience the model services on the Playground. .

4. What is the difference between Alibaba Cloud Model Studio and Qwen (Tongyi Qianwen)?

   Alibaba Cloud Model Studio is a comprehensive platform offering a range of models, including the Qwen series.

5. How can I ensure data isolation across different lines of business?

   Assign distinct workspace permissions to different RAM users by using the Alibaba Cloud account to isolate data across workspaces. For more information, see Use workspaces.

6. Does Alibaba Cloud Model Studio retain my data when I call a model?

   • When you call a model through its APIs, Alibaba Cloud Model Studio does not retain your conversation history. It only de-identifies and logs the status of the call. The user prompt and the generated response are deleted after each call.

   • When you call a model through the Assistant API provided by Alibaba Cloud Model Studio, your conversation history is retained to improve user experience. Currently, no expiration time is set for the retained data.

   • When your application in Alibaba Cloud Model Studio calls a model:

     Your conversation history is cached for 60 minutes. Alibaba Cloud Model Studio offers theNumber of Rounds with Context parameter, which controls the number of conversation rounds retained in the cache. If a new conversation occurs within the 60-minute period, the expiration time of the cache is extended by 60 minutes. If no conversation takes place within this period, the cached conversation history is deleted.

   • Alibaba Cloud is committed to data privacy and does not use your data to train models. All data transmitted during application development or model training is encrypted to ensure security.

Model center

1. How are the parameters of the models stored?

   You can download a model in ModelScope. The model file includes a JSON file detailing the structure of the model. You can use open-source Python components to parse the file and view the vector data. The vector data may be complex, but it can help you understand the storage process.

2. How many languages does the Qwen models support?

   The Qwen models support 14 languages, including Chinese, English, Arabic, Spanish, French, Portuguese, German, Italian, Russian, Japanese, Korean, Vietnamese, Thai, and Indonesian.

3. Can the models accept structured data sources like MySQL and Hive?

   No. But development is in progress with a focus on enabling connectivity to RDS services.

4. Is the text generation speed of models like Qwen2 and Qwen-Max consistent for all users, or can the speed be adjusted?

   The generation speed varies and depends on the available online resources and the overall user demand.

5. After triggering model throttling, how long should I wait before retrying?

   The wait time after triggering model throttling depends on the specific throttling parameters. For instance, if the limit is 120 queries per minute (QPM), and you execute 2 requests every second. Submitting another request within 0.2 seconds will cause throttling, and you would need to wait for the remaining 0.8 seconds.

Application center

1. What types of plug-ins are officially supported currently?

   Official plug-ins include: Python code interpreter, calculator, and image generation. Note that you may need to apply before you can use some plug-ins.

2. Am I charged for using custom plug-ins?

   You are not charged for using custom plug-ins. However, you are charged for prompt optimization, application calls, and test window usage during agent API configuration.

3. How does the assistant understand and use the parameters of my custom plug-ins or functions when I use the Assistant API?

   Custom API plug-ins are designed to be understood by the assistant, which follows the established protocol to pass parameters to the model. For functions, the model learns from the input parameter information and returns comprehensive results.

4. What is the main difference between an agent and the Assistant API?

   The assistant API offers various classes for straightforward tuning.

5. In which fields has Retrieval-Augmented Generation (RAG) been applied?

   RAG demonstrates significant potential and value across various fields, including Q&A systems, dialog systems, text summarization, knowledge graph construction and reasoning, educational and training programs, customer service, news and content generation, as well as intelligent search and recommendation systems.

6. How can I optimize a RAG agent application if the model response is inaccurate?

   To provide feedback on model responses, click Feedback below the response content, select the issue type, and click Submit. Alternatively, copy the requestID and submit a ticket to provide feedback.

7. If the AI output contains **xxxxx**, how can I order the model to use bold text instead?

   The syntax ** ** is used for bold text in the markdown format. When displaying content generated by the model on the frontend, parse and render the markdown syntax accordingly.

8. Can I set the application response to incremental when using the API?

   Configure the following parameters:

   stream=True, # Set the output mode to stream;

   incremental_output=True # Enable incremental stream output;

9. Is the order of RAG parallel or serial with a sequence?

   Retrieval is performed in parallel according to user-configured knowledge bases, followed by the selection of the top N results based on their scores.

10. When Model Studio calls a custom plug-in, can the model pass through the header? The header is shown to be passed through, but according to the return from the server, it is not actually passed through.

    Custom headers are not supported. Only authorization headers are passed through.

Data management

1. What to do if the following is returned: Only PDF, doc, and docx file formats are supported for upload. Please try again. Error code: 140010

   Ensure that the file extension for PDFs is in lowercase, specifically 'pdf'.

2. Why is the MD5 parameter mandatory for the file upload interface?

   The MD5 parameter is required to verify the integrity of the uploaded file.

3. Can I use API to delete or manage of document tags?

   Data Center does not have the document tag feature. Instead, you can create knowledge indexes for differentiation. No API operation to delete or manage tags are provided.

4. I have imported 100 entries of structured data, but only 20 entries are visible, why is that?

   Verify if there are any empty rows within your table data. According to our product policy, data entries following an empty row are not recognized. An initial empty row will result in the file being considered empty.