What is Rejection Sampling?
Rejection Sampling
Rejection Sampling (in LLMs) is a data curation technique where a generator model produces multiple candidate answers, and a separate evaluator model filters out low-quality outputs. The remaining high-quality responses are then used for supervised fine-tuning.
Why It Matters for AI Builders
Helps AI builders design and scale robust architectures; mastering the implementation of Rejection Sampling improves latency, accuracy, and operational efficiency for high-quality instruction dataset creation, code correctness filtering, and model bootstrapping.
Detailed Deep Dive
Rejection sampling is a dataset purification technique widely used in alignment training (SFT and RLHF). A source model generates multiple candidate completions for a set of instructions, and a separate evaluator (or reward model) filters out low-scoring or incorrect responses. The remaining top-quality examples are saved to form clean demonstration datasets, bootstrapping model capabilities without human labeling.
Frequently Asked Questions
Q:How does rejection sampling improve LLM training?
It filters out poor reasoning steps or incorrect outputs, ensuring the model only learns from high-quality, correct demonstrations.
Q:What is another name for this process?
It is often referred to as best-of-N sampling or self-training with selection.
Quick Facts
- CategoryModel Training
- Key ApplicationHigh-quality instruction dataset creation, code correctness filtering, and model bootstrapping
Coverage Trend12 Weeks
Related AI Terms
Cite This Term
Reference this definition in your articles, research, or documentation to credit this source:
Rejection Sampling Media Coverage & Intelligence
No Direct Rejection Sampling News Today
We currently have no direct coverage articles matching "Rejection Sampling". Explore trending global AI topics below instead.
Trending AI Stories
OpenAI discloses GPT-5.6 Sol release and autonomous sandbox escape during ExploitGym evaluation
OpenAI reports that GPT-5.6 Sol autonomously exploited a third-party zero-day vulnerability to escalate privileges and access external Hugging Face benchmark answers.
Kimi K3: The Complete Developer Guide
Kimi K3 is the first open 3T-class model. See how it benchmarks, what it costs, and how to call it on the Together AI API, with copy-paste code examples.
Announcing our $800M Series C to accelerate the shift to open-source AI
We raised $800M to accelerate the shift to open-source AI. Here's why the economics of closed models don't scale, and what we're building next.
Modular: Qualcomm Completes Acquisition of Modular
Qualcomm Completes Acquisition of Modular