Full Pipeline

• Microsoft Foundry resource with fine-tuning enabled
• Python 3.10+ with openai and requests
• Azure CLI (az) authenticated
• A clear task definition: what should the model do differently after fine-tuning?

Phase 1: Define the Task

Answer before touching data or models:

1. What task? (e.g., “translate natural language to Python code”)
2. What does good output look like? Write 5 examples by hand.
3. What does bad output look like? Write 3 anti-examples.
4. How will you measure success? Define evaluation dimensions (see references/grader-design.md).
5. Which base model? Pick 1-3 candidates from the supported model list.

Phase 2: Prepare the Dataset

Option A: You Have Data

1. Convert to SFT JSONL format (see references/dataset-formats.md)
2. Split: 80% train, 10% validation, 10% held-out test
3. Remove or fix low-quality examples

Option B: Synthetic Data

1. Generate using LLM prompts (see workflows/dataset-creation.md)
2. Convert to SFT JSONL with scripts/convert_dataset.py

Option C: Hybrid (Seed + Synthetic)

1. Use existing data as seed, generate synthetic variations
2. Merge, deduplicate, and quality-filter

Checkpoint: You should have training.jsonl, validation.jsonl, and test.jsonl (never used for training).

Phase 3: Establish Baselines

1. Deploy base model (or use existing deployment)
2. Record scores — this is your “zero” that every fine-tune must beat

Phase 4: Choose Training Type

See references/training-types.md for the full decision framework.

Most projects start with SFT. Move to RFT/DPO only if SFT isn’t sufficient.

Phase 5: Upload and Submit Training

Use scripts/submit_training.py or the API directly. See references/hyperparameters.md for starting HP values.

Foundry CLI alternative (no Python):

Phase 6: Monitor and Analyze

1. Wait for completion or use scripts/monitor_training.py
2. Analyze training curves with scripts/check_training.py
3. Read references/training-curves.md to interpret results
4. Check for overfitting — consider deploying an earlier checkpoint if detected

Phase 7: Evaluate Fine-Tuned Model

1. Deploy fine-tuned model (see references/deployment.md for format/SKU)
2. Compare against baseline and previous experiments
3. Delete deployment after evaluation

Phase 8: Iterate

Follow workflows/iterative-training.md:

• Adjust hyperparameters based on training curves
• Try different data subsets or augmentations
• Test different base models
• Track everything in your leaderboard

Phase 9: Ship

When the model convincingly beats baseline:

1. Deploy with production-appropriate capacity
2. Monitor with Application Insights
3. Periodically re-evaluate against test set for regression
4. Retrain as new data becomes available