CtrlK
BlogDocsLog inGet started
Tessl Logo
SkillPublic

finetuning

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).

Quality

90%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide
SecuritybySnyk

Passed

No findings from the security scan

SKILL.md
Quality
Evals
Security

Quality

Content

86%Weight 40%Scale 1-5

Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.

The body is an exemplary lean, actionable overview with concrete commands, rules, and error fixes. Its one real defect is that the Workflows table links to five files that do not exist in the bundle, breaking navigation to the primary step-by-step guides.

Suggestions

Create the missing workflows/ files (quickstart.md, full-pipeline.md, dataset-creation.md, iterative-training.md, diagnose-poor-results.md) or remove the Workflows table until they exist — the broken 'Quick start' link undermines the skill's main entry point.

Add a short numbered pipeline sequence (baseline → validate data → submit → monitor → calibrate → evaluate checkpoints → deploy) directly in SKILL.md so the workflow does not depend solely on the external workflow guides.

DimensionReasoningScore

Conciseness

The body is almost entirely tables and terse directive bullets ('Always baseline first', 'target 25-50% failure rate on the base model'); it never explains what fine-tuning is or pads with concepts Claude already knows. Every section carries operational weight, matching the lean-and-efficient anchor.

5 / 5

Actionability

The Quick Reference gives copy-paste-ready commands with real flags ('python scripts/submit_training.py --model gpt-4.1-mini --training-file train.jsonl --validation-file val.jsonl --type sft'), the Scripts table maps every task to a concrete script, and the Error Handling table pairs specific errors with specific fixes. This meets the fully-executable top anchor.

5 / 5

Workflow Clarity

The Rules encode ordering and validation checkpoints ('Always baseline first', 'Validate data before submitting — run scripts/validate/validate_sft.py', 'Evaluate checkpoints — don't blindly deploy the final one') and the Quick Reference lists tasks in pipeline order, so validation is present for this batch workflow. However, the explicit end-to-end sequence is delegated to workflow guides rather than stated as a numbered sequence in the body, leaving the sequencing implicit — so it sits at 'clear sequence with most checkpoints' rather than the fully explicit top anchor.

4 / 5

Progressive Disclosure

The structure itself is good — well-labeled one-level-deep tables for References and Scripts, and all 12 referenced files and all scripts verified to exist — but all five links in the Workflows table point to a nonexistent workflows/ directory (quickstart.md, full-pipeline.md, dataset-creation.md, iterative-training.md, diagnose-poor-results.md). A third of the navigational paths, including the primary 'Quick start' entry point, dead-end, which is more than the 'minor organization gaps' allowed by the next anchor up.

3 / 5

Total

17

/

20

Passed

Description

92%Weight 40%Scale 1-5

Based on the skill's description, can an agent find and select it at the right time? Clear, specific descriptions lead to better discovery.

A strong description: concrete capabilities, explicit USE FOR / DO NOT USE FOR trigger guidance, and clear boundaries against neighboring skills. The only minor weakness is trigger-term coverage, which misses a few natural synonyms users might say.

DimensionReasoningScore

Specificity

The description names the platform and training types ('Fine-tune models on Microsoft Foundry using SFT (supervised), DPO (preference), or RFT (reinforcement with graders)') and enumerates concrete actions across the whole pipeline ('dataset preparation, training job submission, deployment, and evaluation'), plus distillation and grader calibration. Coverage is comprehensive with no noticeable gaps, matching the anchor for multiple specific concrete actions.

5 / 5

Completeness

Both questions are answered explicitly: the opening sentence states what the skill does, and the 'USE FOR:' clause gives concrete trigger phrases while 'DO NOT USE FOR:' adds explicit negative boundaries with redirects. This clearly matches the top anchor of explicitly answering both what AND when with concrete triggers.

5 / 5

Trigger Term Quality

'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' gives good natural-phrase coverage including synonyms (fine-tune / fine-tuned model). A few common variants are missing (e.g., 'train a model', 'RLHF', 'custom model'), so it falls just short of the comprehensive-synonyms anchor.

4 / 5

Distinctiveness Conflict Risk

The skill occupies a clear niche (fine-tuning on Microsoft Foundry) with distinct trigger terms, and the 'DO NOT USE FOR' clause explicitly disambiguates adjacent skills (deploy-model, agents, prompt-optimizer). Conflict risk is minimal.

5 / 5

Total

19

/

20

Passed

Validation

87%

Checks the skill against the spec for correct structure and formatting. All validation checks must pass before discovery and implementation can be scored.

Validation — 14 / 16 Passed

Validation for skill structure

CriteriaDescriptionResult

relative_links

Relative link issues: 5 missing

Warning

referenced_paths_exist

Referenced path issues: 2 deeper-than-1-level

Warning

Total

14

/

16

Passed

Repository

microsoft/azure-skills

Path

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

Reviewed

Table of Contents

Is this your skill?

If you maintain this skill, you can claim it as your own. Once claimed, you can manage eval scenarios, bundle related skills, attach documentation or rules, and ensure cross-agent compatibility.