---
title: "Foundry fine-tuning: review, setup & limitations · Undominated.ai"
canonical: https://undominated.ai/skills/microsoft-finetuning/
description: "Pins Microsoft's Foundry fine-tuning skill, which treats cleanup as deletion and says to validate training data before a job is submitted."
---

# Foundry fine-tuning: review, setup & limitations · Undominated.ai

> Pins Microsoft's Foundry fine-tuning skill, which treats cleanup as deletion and says to validate training data before a job is submitted.

[← Explore all skills](/skills/)

MODEL FINE-TUNING / Microsoft

# Foundry fine-tuning

Pins Microsoft's Foundry fine-tuning skill, which treats cleanup as deletion and says to validate training data before a job is submitted.

 [See setup guidance ↓](#setup)[Original source ↗](https://github.com/microsoft/skills/tree/fc164ba1d7a0c10377dbcc1c9ac07253b9329fdb/.github/plugins/azure-skills/skills/microsoft-foundry/finetuning)

SOURCE REVIEW

 Reviewed 2026-10-15
 Evidence 5 linked sources
 Publisher Microsoft
 Licence [MIT ↗](https://github.com/microsoft/skills/blob/fc164ba1d7a0c10377dbcc1c9ac07253b9329fdb/LICENSE)
 Revision fc164ba1d7a0
 [Read what was—and wasn’t—checked ↓](#review)

“Fine-tune models on Microsoft Foundry using SFT (supervised), DPO (preference), or RFT (reinforcement with graders). Covers dataset preparation, training job submission, deployment, and evaluation. USE FOR: fine-tune, SFT, DPO, RFT, training data, grader, distillation, fine-tuned model, training job, large file upload, calibrate grader, deploy fine-tuned model, evaluate fine-tuned model. DO NOT USE FOR: general model deployment without fine-tuning (use deploy-model), agent creation (use agents), prompt optimization without training (use prompt-optimizer).”

 [Microsoft · upstream description ↗](https://github.com/microsoft/skills/blob/fc164ba1d7a0c10377dbcc1c9ac07253b9329fdb/.github/plugins/azure-skills/skills/microsoft-foundry/finetuning/SKILL.md) Our analysis follows below.

01 / THE REASONING

## Why this made the selection.

 - The file says cleanup.py deletes old files and deployments, and that a DeploymentNotReady race is fixed by deleting and recreating the deployment.
- It says to validate SFT data before submitting a job, not to deploy the final checkpoint blindly, and not to use this skill when the deployment is not a fine-tune.

### A good fit for

 - Submitting an SFT, DPO, or RFT job only after the named data validator has been run
- Evaluate the base model before training and compare checkpoints before choosing a fine-tuned deployment.

### Weigh up before choosing

 - This review did not execute the resource.
- `/plugin install azure@skills` installs the azure plugin group named in the repository README, not this file alone.
- The azure plugin README prints `/plugin install azure@azure-skills` after marketplace `microsoft/azure-skills`, which is not the same command.
- The azure plugin README says to install the plugin only on trusted devices because it downloads its MCP server with npx.
- The azure plugin README collects telemetry unless AZURE_MCP_COLLECT_TELEMETRY=false.
- The skill metadata version is 0.0.0-placeholder, so this pin is not a numbered release.
- workflows/, references/, and scripts/ were not opened. The file does not itself state an az login step.
- The source review is pinned to revision fc164ba1d7a0c10377dbcc1c9ac07253b9329fdb, but the unversioned plugin install command selects the marketplace release available when installed; its code and dependencies may differ from this reviewed snapshot.

02 / THE REVIEW RECORD

## What we actually inspected.

Source review has boundaries. A clear record is more useful than a “safe” badge.

### Material inspected

 - .github/plugins/azure-skills/skills/microsoft-foundry/finetuning/SKILL.md
- README.md
- .github/plugins/azure-skills/README.md
- LICENSE

### Our findings

 - The cleanup script is described as deleting old fine-tuning files and deployments.
- When a deployment stays not ready after ARM success, the file says to delete that deployment and create it again.
- A content-safety check can block deployment when the training data is dense with PII.
- The rules say to validate SFT data before submit and not to deploy the final checkpoint without evaluating it.
- The file refuses general model deployment, agent creation, and prompt-only optimization.

### Not established by this review

 - The resource was not executed and its tools were not called.
- No training job was submitted and no file or deployment was deleted.
- The referenced workflows, validators, and scripts were not opened.
- The azure plugin was not installed.
- Not executed: it was not installed or run.

The review applies to the material and revision named here. A newer upstream release can change its behavior.

03 / PUT IT TO WORK

## Add a skill to your workflow.

[Upstream setup instructions ↗](https://github.com/microsoft/skills/blob/fc164ba1d7a0c10377dbcc1c9ac07253b9329fdb/README.md)

DOCUMENTED COMMAND

 /plugin install azure@skills Copy command ↗

Copying does not execute this command. It may retrieve a newer version than the reviewed source.

 - The repository README shows `/plugin marketplace add microsoft/skills` and then `/plugin install azure@skills` inside Copilot CLI. That installs the azure plugin group, not this one file.
- The azure-skills plugin README prints a different pair: `/plugin marketplace add microsoft/azure-skills` and `/plugin install azure@azure-skills`.
- The repository README also shows `npx skills add microsoft/skills`, which opens a wizard for the whole repository.

### Before you start

 - An Azure account
- Azure CLI login, or the service-principal variables or managed identity named in the azure plugin README
- Node.js 22 or later if the plugin's npx Azure MCP server is used
- Python if the skill's listed scripts are run; this review did not run them

### Compatibility

Claude Code · GitHub Copilot · GitHub Copilot CLI

### Implementation

python

### Local skill text; Foundry jobs change only if an operator runs the commands

 - Read the skill text
- Authenticate to Azure only if the operator uses the login methods named in the azure plugin README
- Submit a training job or delete fine-tuning files and deployments only if the operator runs those commands
- Installing the full plugin can download and execute Azure MCP/npm hook code and send usage telemetry; review local npm configuration and the telemetry setting before enabling that plugin path.

### Cost model

MIT; no Azure or Python command was run

04 / FOLLOW THE EVIDENCE

## The source trail.

Our notes are separate from the original resource. Check upstream before adopting a new version.

 - [finetuning SKILL.md ↗](https://github.com/microsoft/skills/blob/fc164ba1d7a0c10377dbcc1c9ac07253b9329fdb/.github/plugins/azure-skills/skills/microsoft-foundry/finetuning/SKILL.md) Checked 2026-10-14 https://github.com/microsoft/skills/blob/fc164ba1d7a0c10377dbcc1c9ac07253b9329fdb/.github/plugins/azure-skills/skills/microsoft-foundry/finetuning/SKILL.md Supports: cleanup deletion, delete-and-recreate, content-safety block, and pre-submit validation
- [Azure plugin README ↗](https://github.com/microsoft/skills/blob/fc164ba1d7a0c10377dbcc1c9ac07253b9329fdb/.github/plugins/azure-skills/README.md) Checked 2026-10-14 https://github.com/microsoft/skills/blob/fc164ba1d7a0c10377dbcc1c9ac07253b9329fdb/.github/plugins/azure-skills/README.md Supports: Claude Code and GitHub Copilot CLI setup, az login, and a different plugin install line
- [Repository README ↗](https://github.com/microsoft/skills/blob/fc164ba1d7a0c10377dbcc1c9ac07253b9329fdb/README.md) Checked 2026-10-14 https://github.com/microsoft/skills/blob/fc164ba1d7a0c10377dbcc1c9ac07253b9329fdb/README.md Supports: /plugin install azure@skills and the MIT statement
- [Repository LICENSE ↗](https://github.com/microsoft/skills/blob/fc164ba1d7a0c10377dbcc1c9ac07253b9329fdb/LICENSE) Checked 2026-10-14 https://github.com/microsoft/skills/blob/fc164ba1d7a0c10377dbcc1c9ac07253b9329fdb/LICENSE Supports: MIT License
- [Pinned source sections reviewed in continuation W47 on 2026-10-15 ↗](https://raw.githubusercontent.com/microsoft/skills/fc164ba1d7a0c10377dbcc1c9ac07253b9329fdb/.github/plugins/azure-skills/skills/microsoft-foundry/finetuning/SKILL.md) Checked 2026-10-15 https://raw.githubusercontent.com/microsoft/skills/fc164ba1d7a0c10377dbcc1c9ac07253b9329fdb/.github/plugins/azure-skills/skills/microsoft-foundry/finetuning/SKILL.md Supports: upstreamDescription, bestFor, review.reviewedAt, review.notTested

KEEP COMPARING

## Other approaches to consider.

Related by category or shared topics. These are alternatives to inspect, not a measured quality order.

 [### Foundry customize deployment ↗ Pins Microsoft's Foundry customize-deployment skill, which blocks the create until quota data is valid and the operator approves the summary.](/skills/microsoft-customize/)[### Foundry deploy model ↗ Pins Microsoft's Foundry deploy-model router, which will not create a deployment until the target project is shown and confirmed.](/skills/microsoft-deploy-model/)[### Hugging Face LLM Trainer ↗ Prepares and submits managed training jobs with dataset checks, monitoring and explicit persistence of trained artifacts.](/skills/huggingface-huggingface-llm-trainer/)

 [AI Tools ↗](/tools/)[Skills ↗](/skills/)[Agents ↗](/agents/)[MCP Servers ↗](/mcp-servers/)[Workflows ↗](/workflows/)

## Continue your investigation

 - [Choose a toolkit](/tools/)
- [Inspect execution hosts](/agents/)
- [Check tool access](/mcp-servers/)
- [Check project dependencies](/check-repo/)
