Skip to content

Commit 3cb06cb

Browse files
m365-skilling-repo-management[bot]m365-skilling-repo-management[bot]
authored
Merge pull request #28713 from MicrosoftDocs/main
[AutoPublish] main to live - 06/16 22:32 PDT | 06/17 11:02 IST
2 parents 6cb813f + b787699 commit 3cb06cb

22 files changed

Lines changed: 1307 additions & 147 deletions

‎copilot/TOC.yml‎

Lines changed: 167 additions & 135 deletions
Large diffs are not rendered by default.

‎copilot/copilot-tuning-admin-guide.md‎

Lines changed: 106 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,106 @@
1+
---
2+
title: "Microsoft 365 Copilot Tuning Admin Guide (Preview)"
3+
f1.keywords:
4+
ms.author: lauragra
5+
author: lauragra
6+
manager: calvind
7+
ms.date: 06/17/2025
8+
audience: Admin
9+
ms.topic: overview
10+
ms.service: microsoft-365-copilot
11+
ms.localizationpriority: medium
12+
ms.collection:
13+
- m365copilot
14+
- magic-ai-copilot
15+
description: "Find admin guidance for enabling Copilot Tuning in your organization."
16+
---
17+
18+
# Microsoft 365 Copilot Tuning admin guide (preview)
19+
20+
Microsoft 365 Copilot Tuning enables organizations to securely fine-tune large language models (LLMs) using tenant-specific data, and to deploy declarative agents based on those fine-tuned models to Microsoft 365 Copilot. Administrators retain full control over who can fine-tune models, how they're published, and how they're governed throughout their lifecycle.
21+
22+
This article provides information about the governance controls that are available for Copilot Tuning in the Microsoft 365 admin center.
23+
24+
> [!NOTE]
25+
> Copilot Tuning is currently available for Early Access Preview (EAP). For details about how to enroll, see [Prerequisites](#prerequisites).
26+
27+
## Prerequisites
28+
29+
To manage Copilot Tuning governance controls, make sure that you meet the following prerequisites:
30+
31+
- Copilot Tuning Early Access Preview (EAP) enrollment. To enroll in the EAP:
32+
- Your tenant must have at least 5,000 active Microsoft 365 Copilot add-on licenses.
33+
- An AI Admin must accept the EAP terms on behalf of the organization.
34+
> [!NOTE]
35+
> If Copilot Tuning isn't available in your tenant, contact Microsoft support or your account team to request EAP provisioning.
36+
- You must have one of the following roles:
37+
- Global Administrator
38+
- AI Administrator (if defined as a custom role in your organization)
39+
- Copilot extensibility is enabled via **Copilot settings** in the admin center. For more information, see [Manage agents for Microsoft 365 Copilot in the Microsoft 365 admin center](/microsoft-365/admin/manage/manage-copilot-agents-integrated-apps#enable-or-disable-copilot-extensibility).
40+
- If you block new Power Platform connectors by default via Data Loss Prevention (DLP) policies, run one of the following commands to reclassify the connector.
41+
42+
```powershell
43+
$connectorsToReclassify = @([pscustomobject]@{
44+
id = "/providers/Microsoft.PowerApps/apis/shared_tenantcopilot"
45+
name = "Tenant Copilot"
46+
type = "providers/Microsoft.PowerApps/apis"
47+
})
48+
Add-ConnectorsToPolicy -PolicyName {TENANT_DLP_POLICY_GUID} -Connectors $connectorsToReclassify -Classification {'Confidential'}
49+
```
50+
51+
```powershell
52+
$connectorsToReclassify = @([pscustomobject]@{
53+
id = "/providers/Microsoft.PowerApps/apis/shared_tenantcopilot"
54+
name = "Tenant Copilot"
55+
type = "providers/Microsoft.PowerApps/apis"
56+
})
57+
Add-ConnectorsToPolicy -PolicyName {TENANT_DLP_POLICY_GUID} -Connectors $connectorsToReclassify -Classification {'General'}
58+
```
59+
60+
For more information, see [Add-ConnectorsToPolicy](/powershell/module/microsoft.powerapps.administration.powershell/add-connectorstopolicy).
61+
62+
63+
## What is Copilot Tuning?
64+
65+
Copilot Tuning is a self-serve, secure Microsoft 365 offering for fine-tuning LLMs using your organization's own data. It allows for a self-service workflow where model makers select datasets and Copilot Studio seamlessly handles data preparation, model training, and evaluation. This is suited for low-code domain adaptation tasks with minimal engineering overhead.
66+
67+
After a model is fine-tuned, users can create and deploy declarative agents based on the model to Microsoft 365 Copilot. These agents surface in Word, Teams, Outlook, and other Microsoft 365 apps to perform tasks such as legal clause generation, incident report summarization, or contract drafting.
68+
69+
For more information, see [Copilot Tuning overview](copilot-tuning-overview.md).
70+
71+
## Enable Copilot Tuning
72+
73+
To activate the Copilot Tuning service and scope its availability:
74+
75+
1. Go to the [Microsoft 365 admin center](https://admin.microsoft.com) and sign in with your Global Administrator or AI Administrator account.
76+
2. Go to **Copilot** > **Copilot settings**.
77+
3. Choose **Copilot Tuning**, and choose **Accept** to accept the [EAP terms](https://go.microsoft.com/fwlink/?linkid=2312708).
78+
79+
> [!NOTE]
80+
> The **Copilot Tuning** setting is only available if your tenant meets the criteria described in the [Prerequisites](#prerequisites) section.
81+
82+
4. Choose **Specific users in your organization** and add the users in your organization who can create task-specific models. For more information, see [Manage Copilot Tuning settings](#manage-copilot-tuning-settings).
83+
84+
## Manage Copilot Tuning settings
85+
86+
Fine-grained access controls ensure that only authorized users within the tenant can initiate fine-tuning or publishing models. You can manage this access by specifying users and security groups for these controls.
87+
88+
### Configure model makers
89+
90+
Designate users who are authorized to use Copilot Tuning to fine-tune task-specific models. These model makers are generally subject matter experts (SMEs) within a particular part of the organization like marketing, finance, or legal. The users that you designate as model makers are then able to access the Copilot Tuning workflow in Copilot Studio.
91+
92+
You can designate up to 10 users to be model makers. If you need more than 10 model makers in your organization, reach out to Copilot Tuning support or your Microsoft account team.
93+
94+
When you add a user to Copilot Tuning, they receive an email with instructions to get started. You can also use the **Copilot Tuning** control in the admin center to remove users from Copilot Tuning.
95+
96+
### Manage tuned task-specific models
97+
98+
You can review and manage the set of models that model makers create and publish via the **Copilot Tuning** control in the admin center.
99+
100+
Choose any published model to review the associated security groups and update them as needed.
101+
102+
You can also remove a published model by choosing **Remove this model**.
103+
104+
## Related content
105+
106+
- [Copilot Tuning overview](copilot-tuning-overview.md)

‎copilot/copilot-tuning-doc-generation.md‎

Lines changed: 122 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,122 @@
1+
---
2+
title: Configure Copilot Tuning for document generation
3+
author: jasonjoh
4+
ms.author: jasonjoh
5+
manager: calvind
6+
ms.audience: ITPro
7+
ms.reviewer: jwolk
8+
ms.date: 06/17/2025
9+
ms.topic: how-to
10+
ms.localizationpriority: medium
11+
description: Learn how to use Copilot Tuning to build an AI model for document generation based on organizational knowledge.
12+
---
13+
14+
# Configure Copilot Tuning for document generation
15+
16+
<!-- cSpell:ignore calvind jwolk -->
17+
18+
Initial drafts of new documents can be generated using a custom document generation model trained with Copilot Tuning. Copilot Tuning document generation models use an original input document and specified changes to create a draft of output file.
19+
20+
Document generation works best when the drafting process follows recognizable patterns and requires incorporating consistent modifications across similar types of documents. This approach streamlines the creation of first drafts, reducing manual editing time and promoting consistency across your documents.
21+
22+
[!INCLUDE [copilot-tuning-preview](includes/copilot-tuning-preview.md)]
23+
24+
Some example use cases include:
25+
26+
- Human resources – Generate new job listings matching your organization's tone and structure based on existing job listing templates and information about new jobs.
27+
- Legal – Draft recurring contracts quickly and consistently based on previous contracts and new or updated contract terms.
28+
- Compliance – Create new compliance forms based on an approved form template and information tailored to new clients, deals, or jurisdictions.
29+
- Procurement – Create draft purchase orders using previous purchase orders and information about new purchases.
30+
- Documentation – Draft new product or feature docs based on existing documentation templates and information about new products or features.
31+
32+
## Prerequisites
33+
34+
- You must have permission to use Copilot Tuning in Copilot Studio. <!-- TODO: Link to permission doc here if it exists -->
35+
- A collection of original documents and corresponding final draft documents that are stored in SharePoint.
36+
- A collection of change logs or specifications stored in SharePoint.
37+
- A structured version of required changes to provide in the supplementary field within Copilot Tuning.
38+
- A minimum of 20 well-aligned pairs of reference documents to target pairs that reflect a representative range of changes you expect the system to handle.
39+
40+
> [!IMPORTANT]
41+
> Document generation supports working with the following file formats: .doc, .docx, .html, .md, or .pdf. Copilot Tuning only uses information found in text. Copilot Tuning document generation doesn't use information in images, tables, or unstructured web content in your documents.
42+
43+
## Create a document generation model
44+
45+
The following are the high-level steps to configure a custom document generation model using Copilot Tuning.
46+
47+
> [!div class="checklist"]
48+
>
49+
> - [Prepare a mapping file](#prepare-a-mapping-file) to identify pairings of original files and final draft files in the training data
50+
> - [Customize the model](#customize-the-model)
51+
> - [Select security groups](#review-copilot-tuning-acl-analysis)
52+
> - [Label training data](#label-training-data)
53+
> - [Train the model and evaluate results](#train-and-evaluate-results)
54+
> - [Publish the model](#publish-the-model)
55+
56+
### Prepare a mapping file
57+
58+
Your knowledge source should have at least 20 example pairs of original files and corresponding final (draft) files. In this step, you prepare a CSV file that provides at least 20 examples of original files to final (draft) documents. Copilot Tuning uses these examples to fine-tune the generation logic, helping the model learn how your organization typically edits or adapts documents.
59+
60+
Create a file named **mapping.csv** and store it in the root directory of your knowledge source. This file should have two columns:
61+
62+
- The first column is named `precedent` and contains the path to an original file in the data source.
63+
- The second column is named `target` and contains the path to the final draft file in the data source that was created using the original file as a basis.
64+
65+
```CSV
66+
precedent,target
67+
"https://contoso.sharepoint.com/sites/ProductSpecs/Shared%20Documents/Mark-8-FAQ.docx", "https://contoso.sharepoint.com/sites/ProductSpecs/Shared%20Documents/mark-8-faq.md"
68+
```
69+
70+
### Customize the model
71+
72+
To customize the model:
73+
74+
1. Open [Copilot Studio](https://copilotstudio.microsoft.com) in your browser. In the left-hand navigation bar, select the ellipses (**...**), and then select **Copilot Tuning**.
75+
1. Select **Create new**.
76+
1. Enter a name for your model and provide a description of what the model does. Ensure that the model name and description are user-friendly and clearly describe the purpose of the model.
77+
1. Select **Add knowledge** to select the SharePoint location that contains your training data.
78+
1. Under **Permissions** you can enter security groups or email addresses of the individuals who should be able to use this model. Copilot Tuning ensures that any knowledge that isn't accessible by these selected security groups is filtered out. If you don't explicitly give permissions, the model is only accessible to you.
79+
1. Set the task type to **Document generation**.
80+
1. Provide **Model Instructions** about your scenario. Instructions ensure Copilot Tuning is using the terminology your organization uses and evaluates your models output using the criteria one of your users would use to evaluate the document. It's important to accurately answer these questions because this information cues the system how to interpret the underlying data.
81+
1. Select **Prepare labeling data**.
82+
83+
Copilot Tuning begins analyzing the ACLs of the training documents. Based on the size of your dataset, this step might take up to 24 hours. Copilot Tuning sends you an email when it's ready for you to go to the next step in the model creation process. In the meanwhile, you can check the status of the model from the Copilot Tuning landing page.
84+
85+
### Review Copilot Tuning ACL analysis
86+
87+
Copilot Tuning analyzes the document access control lists (ACLs) of the training documents and then filters out any knowledge that isn't accessible to the security groups selected in the previous step. Copilot Tuning might suggest more security groups based on its analysis to ensure access to the training data.
88+
89+
Based on the size of your dataset, this step might take up to 24 hours. Copilot Tuning sends you an email when it's ready for you to go to the next step in the model creation process. In the meanwhile, you can check the status of the model from the Copilot Tuning landing page.
90+
91+
### Label training data
92+
93+
Preparing high-quality datasets for fine-tuning can be time-consuming and often requires multiple iterations and input from data scientists. Copilot Tuning simplifies this process by automatically identifying which of your provided document pairs are most effective for fine-tuning. This learning process is iterative, and at times, Copilot Tuning might send an email asking you to label files to improve its understanding of what constitutes a high-quality target. When you receive an email indicating that data is ready for labeling, follow these steps to complete the process.
94+
95+
1. Open [Copilot Studio](https://copilotstudio.microsoft.com) in your browser. In the left-hand navigation bar, select the ellipses (**...**), and then select **Copilot Tuning**.
96+
1. Select the model that is ready for labeling to open the labeling form.
97+
1. The labeling form provides sample documents that Copilot Tuning identified as example final draft documents. Examine each document and label whether or not you think the document is a good example draft document. Once you label all of the documents, submit the labeling form.
98+
99+
Copilot Tuning continues preparing the training data based on your input. Based on the size of your dataset, this step might take up to 24 hours. You can return to the Copilot Tuning landing page and see the status of your model configuration. Your model configuration might ask you to label more data, might be processing the data in your knowledge source, or it might be ready to fine-tune the model. Copilot Tuning sends you an email when it's ready for you to go to the next step in the model creation process.
100+
101+
### Train and evaluate results
102+
103+
When sufficient data is prepared by Copilot Tuning, you can proceed with training the model. Fine-tuning of your model occurs in Azure AI Foundry. After the model is fine-tuned, Copilot Tuning generates results that you can review to evaluate if the model is of sufficient quality to deploy.
104+
105+
1. Open [Copilot Studio](https://copilotstudio.microsoft.com) in your browser. In the left-hand navigation bar, select the ellipses (**...**), and then select **Copilot Tuning**.
106+
1. Select the model that is ready for training to open the model configuration dialog form.
107+
1. Select **Start model training** to begin fine-tuning of your custom model.
108+
109+
Based on the size of your dataset, the fine-tuning step might take up to 24 hours. You can return to the Copilot Tuning landing page and see the status of your model.
110+
111+
If the results don't meet your expectations, you can go to the previous steps in the flow to add more data sources, adjust the model instructions, or provide more samples in your mapping file. Retrain the model until document generation results meet expectations.
112+
113+
### Publish the model
114+
115+
When the model generates results that meet your quality standards, you can publish the model through the Copilot Tuning interface. Publish the model by selecting the **Publish model** button. The model is deployed to an isolated and secured environment. It's only accessible to the security groups or users that were selected during model customization.
116+
117+
## Related content
118+
119+
- [Microsoft 365 Copilot Tuning overview (preview)](copilot-tuning-overview.md)
120+
- [Microsoft 365 Copilot Tuning FAQ](copilot-tuning-faq.yml)
121+
- [Create declarative agents with Copilot Studio agent builder and Microsoft 365 Copilot Tuning models](/microsoft-365-copilot/extensibility/copilot-studio-agent-builder-tuned-models).
122+
- [Troubleshooting document generation models](copilot-tuning-troubleshooting-doc-generation.md).

0 commit comments

Comments
 (0)