Unverified50% confidenceConceptExact time
Classifier-free guidance (CFG) originally came from diffusion generative models; the approach is to compute both conditional and unconditional predictions during sampling, then use a weight w to amplify the difference between them, with the formula ε = ε_uncond + w·(ε_cond − ε_uncond).
1
Sources
50%
Confidence
Long-term
Relevance
10/14/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedCFG performs conditional and unconditional predictions separately during inference and interpolates and amplifies them via the formula output = uncond + guidance_scale × (cond - uncond), at the cost of requiring two forward passes per step.76% similarUnverifiedCFG is a core hyperparameter in diffusion models, rooted in a 2022 research finding, and works by enhancing responsiveness to text descriptions through a weighted mixture of the predictions from the conditional and unconditional denoising paths.72% similarUnverifiedCFG (Classifier-Free Guidance) requires running both conditional and unconditional forward passes simultaneously at each denoising step, doubling the computation per inference step when enabled.70% similarUnverifiedClassifier-free guidance (CFG) was proposed by Jonathan Ho and Tim Salimans in 2022; by randomly dropping text conditioning during training, the model learns both conditional and unconditional generation.65% similarVerifiedSpeculative decoding is a lossless acceleration technique that does not alter the model's output distribution; it uses a draft model with far fewer parameters than the target model to generate candidate tokens, which the target model then verifies in parallel.63% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1024131API
curl https://kongchang.com/api/v1/knowledge/claims/1024131MCP
get_claim(id=1024131)