Location via proxy:   [ UP ]  
[Report a bug]   [Manage cookies]                
×
Jul 29, 2020 · A feedback-based adaptive top-k text decoding algorithm called mirostat that generates text (of any length) with a predetermined value of perplexity.
We introduce mirostat,1 a neural text decoding algorithm that actively controls the generative process to maintain the perplexity of generated text at a certain ...
We provide a new text decoding algorithm that directly controls generated text statistics and hence generates more human-like texts using large language models.
A feedback-based adaptive top-k text decoding algorithm called mirostat is designed that generates text (of any length) with a predetermined target value of ...
People also ask
Jan 17, 2022 · "Mirostat: A Neural Text Decoding Algorithm that Directly Controls Perplexity", Basu et al 2020 (a better GPT sampler than nucleus sampling ...
A feedback-based adaptive top-k text decoding algorithm called mirostat is designed that generates text (of any length) with a predetermined value of ...
Neural text decoding algorithms strongly influence the quality of texts generated using language models, but popular algorithms like top-k, top-p (nucleus), ...
Mirostat: A Neural Text Decoding Algorithm that Directly Controls Perplexity. Govardana Sachitanandam Ramachandran. Salesforce Research. Lav R. Varshney.
A feedback-based adaptive top-k text decoding algorithm called mirostat that generates text (of any length) with a predetermined value of perplexity.