Alibaba Cloud Machine Translation uses API throttling to ensure service stability and fair use of resources. This topic describes the API throttling rules and usage limits for Machine Translation. It also provides recommendations for handling API throttling.
To increase your Queries Per Second (QPS) limit, contact pre-sales support or join our DingTalk group at 23369411.
Scope of throttling
Throttling is enabled for all Machine Translation APIs. Different APIs have different throttling thresholds.
API usage limits
All Machine Translation APIs have usage limits. The types of limits vary by API.
What is the API rate limit for a single user?
The API rate limit for a single user is the maximum number of times an API can be called per second or per minute. This limit applies to an Alibaba Cloud account and all Resource Access Management (RAM) users under that account.
For example, the API rate limit for TranslateGeneral is 50 calls per second. If an Alibaba Cloud account has five RAM users, and all six accounts (the main account plus the five RAM users) call this API, their combined call frequency cannot exceed 50 calls per second.
Machine Translation product limits
For more information about API rate limits, see Quota Center. For more information about API billing, see Billing overview.
Product category | API | API title | Data transfer quotas | Usage limit | Usage notes |
Text translation | General-purpose text translation | 50 QPS | Fewer than 5,000 characters per request | Punctuation marks, spaces, and HTML tags are counted as characters. A Chinese character, an English letter, or a punctuation mark is counted as one character for billing. | |
Professional text translation | 50 QPS | Fewer than 5,000 characters per request | |||
Batch text translation | 10 QPS | Fewer than 1,000 characters per request | |||
CreateAsyncTranslate - Create an asynchronous text translation task | Create an asynchronous text translation task | 5 QPS | Fewer than 1,000,000 characters per request | ||
GetAsyncTranslate - Get the result of an asynchronous text translation task | Get the result of an asynchronous text translation task | 5 QPS | |||
Language detection | 50 QPS | Fewer than 5,000 characters per request | |||
Multimodal translation | Image translation | 5 QPS | File size must be less than 10 MB | Images that fail to be translated are not billed. | |
Certificate translation | 5 QPS | File size must be less than 10 MB | |||
Create a document translation task | 5 QPS | File size must be less than 50 MB | |||
Get the result of a document translation task | 5 QPS | ||||
Cross-border e-commerce multilingual tools | Intelligent product image translation | 3 QPS | File size limit: 10 MB. Pixel limits: The shortest side of the image must be at least 15 pixels, and the longest side must not exceed 8,192 pixels. | ||
Create an asynchronous batch image translation task | 3 QPS | ||||
Call to get the result of an asynchronous image translation task | Get the result of an asynchronous batch image translation task | 3 QPS | |||
Image language diagnosis | 3 QPS | ||||
Title diagnosis | 10 QPS | ||||
Title optimization | 10 QPS | ||||
Intelligent title generation | 10 QPS |
Throttling rules
The following rules apply to API throttling:
Throttling is calculated independently for each API. If an API call reaches the throttling threshold, it does not affect calls to other APIs. It also does not affect the normal use of the same API in other regions.
If an API is throttled during minute T, you can call the API again during minute T+1.
If an API call made using an SDK or the command-line interface (CLI) is throttled, the corresponding operation in the Alibaba Cloud console is also throttled.
Recommendations
When the number of calls to a Machine Translation API reaches the throttling threshold, the system returns a throttling error code and does not process the request. Consider the following recommendations when you call APIs:
Bulk operations
Control the QPS.
Use a backoff strategy. If an API call returns a throttling error code, wait before you send subsequent requests to the same API. During the backoff period, you can use availability probing at a rate of 1 QPS for the same API under the same Alibaba Cloud account.