Training
Published training-regime pages covering pretraining, post-training, distillation, quantization-aware training, and related optimization methods.
- Diffusion Training Objective
The pretraining-style regime where a denoiser learns to predict noise or a clean-direction signal from corrupted examples across a noise schedule.
- Direct Preference Optimization
A preference-optimization training regime that teaches a model from preferred versus rejected answer pairs without fitting a separate reward model and PPO loop first.
- Distillation
A teacher-student training regime where a smaller model learns from supervision produced by a larger or stronger teacher model.
- Dropout
A training-time regularization method that randomly disables a subset of activations during learning and uses the full network, with appropriate scaling, at inference.
- FP4 Quantization-Aware Training
A DeepSeek-V4 training step that prepares the model for very low precision by exposing it to quantization effects during training.
- Group Relative Policy Optimization
A groupwise alignment training regime that samples multiple candidate answers for the same prompt and updates the model from how those answers rank relative to one another.
- Instruction Tuning
A supervised post-training regime that teaches a model to follow user instructions by training on prompts paired with desired answers.
- Mid-Training
Continued training on a pretrained base checkpoint before final post-training, using targeted data or objectives to extend capability without yet shaping final assistant behavior.
- On-Policy Distillation
A DeepSeek-V4 post-training routine where the student is refined on trajectories produced under its current policy rather than only on frozen offline traces.
- Post-Training
The training stages after base pretraining that reshape how a language model behaves in use, including instruction following, preference alignment, and safety shaping before deployment.
- Pretraining
The large base-model training stage where a transformer learns broad language patterns from massive token sequences before later alignment or deployment shaping.
- Proximal Policy Optimization
A post-training regime that updates a model with clipped reinforcement-learning steps so RLHF behavior improves without each update moving too far at once.
- Reinforcement Learning from Human Feedback
A post-training alignment workflow that collects human preference signals, fits a reward model, and optimizes the language-model policy toward preferred behavior.
- Reinforcement Learning with Verifiable Rewards
A post-training reinforcement learning regime where rewards come from externally checkable task outcomes rather than human preference labels alone.
- Specialist Training
A DeepSeek-V4 post-training pattern that sharpens separate capability areas before they are merged back into the final stack.
- Supervised Fine-Tuning
A post-training regime that adapts a capable base model by training on labeled input-output examples.