Atlas · GenAI 2026
Model Training, Fine-Tuning & Alignment
20 skills · ontology graph below shows relations within this section.
What this domain covers
This edition groups 20 capabilities in Model Training, Fine-Tuning & Alignment across 8 named categories. The inventory contains 16 concepts and 4 tools; 10 skills appeared in at least two of the three original research runs. The remaining entries stay visible with their lower coverage so a reader can distinguish taxonomy scope from research-system agreement.
Current category labels: Alignment · Fine-Tuning · Fine-Tuning Methods · Model Composition · Model Compression · Parameter-Efficient Fine-Tuning · Pre-Training & Adaptation · Training Infrastructure
Frequent learning foundations
- LLM Fine-Tuning supports 6 mapped skills
- RLHF supports 4 mapped skills
- Deep Learning supports 3 mapped skills
- Linear Algebra supports 3 mapped skills
- Transformer Architecture supports 3 mapped skills
Skills in this section
Direct Preference Optimization
Alignment
RLHF
Alignment
Reinforcement Learning from Verifiable Rewards
Alignment
Reward Modeling
Alignment
Catastrophic Forgetting
Fine-Tuning
Fine-Tuning Evaluation
Fine-Tuning
Hugging Face PEFT
Fine-Tuning
Hugging Face TRL
Fine-Tuning
LLM Fine-Tuning
Fine-Tuning
Supervised Fine-Tuning (SFT)
Fine-Tuning Methods
Model Merging
Model Composition
Knowledge Distillation
Model Compression
Model Quantization
Model Compression
LoRA / QLoRA
Parameter-Efficient Fine-Tuning
Continual Pre-Training
Pre-Training & Adaptation
DeepSpeed
Training Infrastructure
Distributed Training
Training Infrastructure
GRPO
Alignment
Unsloth
Fine-Tuning
Federated Learning
Training Infrastructure