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
Starting with version `2024-07-31-preview` and later you can receive **10 hours** of free model training. Billing charges are calculated for model trainings that exceed 10 hours. You can choose to spend all of 10 free hours on a single build with a large set of data, or utilize it across multiple builds by adjusting the maximum duration value for the `build` operation by specifying `maxTrainingHours`:
279
+
280
+
Starting with version `2024-07-31-preview`, you can train your custom neural model for longer durations than 30 minutes. Previous versions have been capped at 30 minutes per training instance, with a total of 20 free training instances per month. Now with `2024-07-31-preview`, you can receive **10 hours** of free model training, and train a model for as long as 10 hours. If you would like to train a model for longer than 10 hours, billing charges are calculated for model trainings that exceed 10 hours. You can choose to spend all of 10 free hours on a single build with a large set of data, or utilize it across multiple builds by adjusting the maximum duration value for the `build` operation by specifying `maxTrainingHours` as below:
281
281
282
282
```bash
283
283
@@ -288,7 +288,10 @@ POST /documentModels:build
288
288
}
289
289
```
290
290
291
-
Build time varies. Billing is calculated for the actual time spent (excluding time in queue), with a minimum of 30 minutes per training job. The elapsed time is converted to V100 equivalent training hours and reported as part of the model.
291
+
> [!NOTE]
292
+
> For Document Intelligence versions `v3.1 (2023-07-31)` and `v3.0 (2022-08-31)`, custom neural model's paid training is not enabled. For the two older versions, you will get a maximum of 30 minutes training duration per model. If you would like to train more than 20 model instances, you can request for increase in the training limit.
293
+
294
+
Each training hour is the amount of compute a single V100 GPU can perform in an hour. As each build takes different amount of time, billing is calculated for the actual time spent (excluding time in queue), with a minimum of 30 minutes per training job. The elapsed time is converted to V100 equivalent training hours and reported as part of the model.
292
295
293
296
```bash
294
297
@@ -305,10 +308,30 @@ This billing structure enables you to train larger data sets for longer duration
305
308
306
309
:::moniker-end
307
310
311
+
:::moniker range="doc-intel-3.1.0"
312
+
313
+
## Billing
314
+
315
+
For Document Intelligence versions `v3.1 (2023-07-31)` and `v3.0 (2022-08-31)`, you will get a maximum of 30 minutes training duration per model, and a maximum of 20 trainings forfree per month. If you would like to train more than 20 model instances, you can request for increasein the training limit.
316
+
317
+
If you are interested in training models for longer durations than 30 minutes, we support **paid training**for our newest version, `v4.0 (2024-07-31)`. Using the latest version, you can train your model for a longer duration to process larger documents.
318
+
319
+
:::moniker-end
320
+
321
+
:::moniker range="doc-intel-3.0.0"
322
+
323
+
## Billing
324
+
325
+
For Document Intelligence versions `v3.1 (2023-07-31)` and `v3.0 (2022-08-31)`, you will get a maximum of 30 minutes training duration per model, and a maximum of 20 trainings forfree per month. If you would like to train more than 20 model instances, you can request for increasein the training limit.
326
+
327
+
If you are interested in training models for longer durations than 30 minutes, we support **paid training**for our newest version, `v4.0 (2024-07-31)`. Using the latest version, you can train your model for a longer duration to process larger documents.
328
+
329
+
:::moniker-end
330
+
308
331
## Next steps
309
332
310
333
Learn to create and compose custom models:
311
334
312
335
> [!div class="nextstepaction"]
313
336
> [**Build a custom model**](how-to-guides/build-a-custom-model.md)
Copy file name to clipboardExpand all lines: articles/ai-services/document-intelligence/concept-read.md
+1Lines changed: 1 addition & 0 deletions
Display the source diff
Display the rich diff
Original file line number
Diff line number
Diff line change
@@ -121,6 +121,7 @@ The searchable PDF capability enables you to convert an analog PDF, such as scan
121
121
>
122
122
> * Currently, the searchable PDF capability is only supported by Read OCR model `prebuilt-read`. When using this feature, please specify the `modelId` as `prebuilt-read`, as other model types will return error for this preview version.
123
123
> * Searchable PDF is included with the 2024-07-31-preview `prebuilt-read` model with no additional cost for generating a searchable PDF output.
124
+
> * Searchable PDF currently only supports PDF files as input. Support for other file types, such as image files, will be available later.
|**Min number of samples per class * Classifier**| 5 | 5 (default value) |
182
+
| Adjustable | No | No |
183
+
184
+
::: moniker-end
185
+
186
+
::: moniker range="=doc-intel-3.1.0"
114
187
115
188
## Custom model usage
116
189
@@ -169,12 +242,15 @@ Document Intelligence billing is calculated monthly based on the model type and
169
242
::: moniker range=">=doc-intel-2.1.0"
170
243
171
244
> <sup>1</sup> For **Free (F0)** pricing tier see also monthly allowances at the [pricing page](https://azure.microsoft.com/pricing/details/form-recognizer/).</br>
172
-
> <sup>2</sup> See [best practices](#example-of-a-workload-pattern-best-practice), and [adjustment instructions(#create-and-submit-support-request).</br>
245
+
> <sup>2</sup> See [best practices](#example-of-a-workload-pattern-best-practice), and [adjustment instructions](#create-and-submit-support-request).</br>
173
246
> <sup>3</sup> Neural models training count is reset every calendar month. Open a support request to increase the monthly training limit.
174
247
::: moniker-end
175
248
::: moniker range=">=doc-intel-3.0.0"
176
249
> <sup>4</sup> This limit applies to all documents found in your training dataset folder prior to any labeling-related updates.
177
250
::: moniker-end
251
+
::: moniker range=">=doc-intel-4.0.0"
252
+
> <sup>5</sup> This limit applies for `v 4.0 (2024-07-31)` custom neural models only. Starting from `v 4.0`, we support training larger documents for longer durations (up to 10 hours for free, and incurring charges after). For more information, please refer to [custom nerual model page](concept-custom-neural.md).
253
+
::: moniker-end
178
254
179
255
## Detailed description, Quota adjustment, and best practices
0 commit comments