Week 08 — MLOps, Fine-Tuning & Model Publishing

Train, fine-tune, package, publish, and monitor models: data/versioning, MLflow, Vertex AI, HuggingFace, quantization, and responsible model release.

Topics

#	Topic	Focus
1	Cloud Storage for ML	GCS buckets, versioned checkpoints, datasets
2	BigQuery ML	Training in SQL, Vertex AI integration
3	MLflow	Runs, experiments, registry
4	Fine-Tuning Strategy	Fine-tune vs RAG vs prompting; data cleaning
5	HuggingFace Ecosystem	Datasets, transformers, PEFT, TRL, accelerate
6	Fine-Tuning Techniques	LoRA, QLoRA, DPO, RLHF/RLAIF concepts
7	Quantization	GGUF, AWQ, GPTQ — speed vs accuracy
8	Gemma 4 Fine-Tuning	Unsloth, evaluation, upload
9	GCP ML Pipeline	Train → eval → register → deploy → monitor
10	Model Publishing & Model Cards	Licensing, responsible AI, model cards

Hands-on Lab (Labs + Capstones)

• CAPSTONE: Production Fine-Tuned Model + Full MLOps
• Lab: Full GCP pipeline (GCS → BigQuery ML → Vertex AI → Cloud Run)

Learning outcomes (dummy)

• Explain the core concepts in this week’s toolchain.
• Implement a small working prototype.
• Measure or validate results with at least one simple check.

Content status

All pages are placeholders right now. We’ll replace them with real notes later.