You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Copy file name to clipboardExpand all lines: articles/cognitive-services/language-service/concepts/model-lifecycle.md
+26-20Lines changed: 26 additions & 20 deletions
Display the source diff
Display the rich diff
Original file line number
Diff line number
Diff line change
@@ -9,7 +9,7 @@ ms.service: cognitive-services
9
9
ms.subservice: language-service
10
10
ms.custom: event-tier1-build-2022
11
11
ms.topic: conceptual
12
-
ms.date: 07/27/2022
12
+
ms.date: 08/15/2022
13
13
ms.author: aahi
14
14
---
15
15
@@ -21,21 +21,20 @@ Language service features utilize AI models that are versioned. We update the la
21
21
22
22
### Expiration timeline
23
23
24
-
Our standard (not customized) language service is built upon AI models that we call pre-trained models. We update the language service with new model versions every few months to improve model accuracy, support, and quality.
24
+
Our standard (not customized) language service features are built upon AI models that we call pre-trained models. We update the language service with new model versions every few months to improve model accuracy, support, and quality.
25
25
26
-
Pre-built Model Capabilities: As new models and new functionality become available and older, less accurate models are retired. Unless otherwise noted, retired pre-built models will be automatically updated to the newest model version.
26
+
As new models and functionalities become available, older less accurate models are deprecated. To ensure you are using the latest model version and avoid interruptions to your applications, we highly recommend using the default model-version parameter (`latest`) in your API calls. After their deprecation date, pre-built model versions will no longer be functional and your implementation may be broken.
27
27
28
-
During the model version deprecation period, API calls to the soon-to-be retired model versions will return a warning. After model-version deprecation, API calls to deprecated model-versions will return responses using the newest model version with an additional warning message. So, your implementation will never break, but results might change.
29
-
30
-
The model-version retirement period is defined as: the period of time from a release of a newer model-version for the capability, until a specific older version is deprecated. This period is defined as six months for stable model versions, and three months for previews. For example, a stable model-version `2021-01-01` will be deprecated six months after a successor model-version `2021-07-01` is released, on January 1, 2022. Preview capabilities in preview APIs do not maintain a minimum retirement period and can be deprecated at any time.
28
+
Stable (not preview) model versions are deprecated six months after the release of another stable model version. Features in preview don't maintain a minimum retirement period and may be deprecated at any time.
31
29
32
30
33
31
#### Choose the model-version used on your data
34
32
35
-
By default, API requests will use the latest Generally Available model. You can use an optional parameter to select the version of the model to be used.
33
+
By default, API requests will use the latest Generally Available model. You can use an optional parameter to select the version of the model to be used (not recommended).
36
34
37
35
> [!TIP]
38
36
> If you’re using the SDK for C#, Java, JavaScript or Python, see the reference documentation for information on the appropriate model-version parameter.
37
+
39
38
For synchronous endpoints, use the `model-version` query parameter. For example:
0 commit comments