Glossary
Explore key terms in language model fine-tuning, guardrails, agent security, and AI governance.
Fine-tuning basics
- Batch size
- Catastrophic forgetting
- Chat template
- Checkpoint
- Context length
- Continued pre-training
- Data needed for fine-tuning
- Dataset formats
- Epochs and steps
- Evaluate a fine-tuned model
- Fine-tune an LLM
- Fine-tuning
- Fine-tuning time
- Fine-tuning vs prompt engineering
- Fine-tuning vs RAG
- Full fine-tuning
- GPUs needed for fine-tuning
- Gradient accumulation
- Gradient checkpointing
- Gradient clipping
- Instruction tuning
- Learning rate
- LoRA
- LoRA rank and alpha
- LoRA target modules
- LoRA vs full fine-tuning
- Merge a LoRA adapter
- Mixed precision
- Overfitting
- PEFT
- Post-training
- Pre-training
- QLoRA
- Quantization
- Sequence packing
- Synthetic data
- Tokenization
- Warmup and schedules
Training methods
- Async GRPO
- Choose a fine-tuning method
- Classification fine-tuning
- DPO
- Embedding fine-tuning
- GRPO
- Knowledge distillation
- KTO
- Offline GRPO
- Online distillation
- Online GRPO
- PPO
- Preference tuning
- Reinforcement fine-tuning
- Reward model
- RL environments
- RLHF
- Rollout server
- Rollouts
- Self-distillation
- SMPO
- Supervised fine-tuning (SFT)
- Vision-language fine-tuning
Distributed training
Models and tools
- Axolotl
- Fine-tune Command A+
- Fine-tune Gemma 4
- Fine-tune GLM-4.7-Flash
- Fine-tune GPT-OSS
- Fine-tune Laguna 2.1
- Fine-tune LFM2
- Fine-tune Mistral Small 4
- Fine-tune Qwen3
- Fine-tune Qwen3.5 and Qwen3.6
- Fine-tune ZAYA1
- Halo
- Halo license
- Halo vs Megatron
- Halo vs TRL
- Hugging Face Transformers
- Megatron-LM
- Mixture of experts (MoE)
- Open-weight models
- SGLang
- Transformer
- TRL
- Unsloth
- vLLM
AI safety and guardrails
- Add guardrails to an LLM app
- AI alignment
- AI control
- AI guardrails
- AI policy domains
- AI red teaming
- AI safety vs AI security
- Build vs buy guardrails
- Content moderation
- Data leakage
- Evaluating guardrails
- Guard models
- Guardrail actions
- Guardrail false positives
- Guardrail latency
- Guardrail policy
- Guardrails vs evals
- Guardrails vs provider safety filters
- Hallucination
- Indirect prompt injection
- Input and output guardrails
- Jailbreak
- LLM firewall
- Multi-turn attack
- Natural-language policy
- PII redaction
- Prevent prompt injection
- Prompt injection
- RAG security
- Real-time guardrails
- Red team an LLM
- Secret detection
- System prompt leakage
- System prompt vs guardrails
- Writing a guardrail policy