beam search
AssessTechniques
A search method that keeps the best-scoring partial candidates at each step.
Why it's here
Placed in Assess: 2 article(s) of evidence from 1 source(s), led by model releases, with 2 in the last 30 days. Confidence 32%.
Evidence (2)
- 7Hacker News·10/8/2026model_releaseWhistle: 16.9 MB On-Device Speech-to-Text Model
Whistle is a compact speech recognition model designed to run entirely on-device on CPUs with no dependencies. It supports transcription, word-level timestamps, and speech embeddings for up to 30 seconds of audio in seven languages, with audio never leaving the device.
- 4Hacker News·9/22/2026researchUsing gzip as a language model
The article explores whether the gzip compressor can be repurposed for language modeling by priming it with a corpus and scoring candidate continuations by compressed length. It shows that gzip can produce partially coherent Shakespeare-like text using beam search, illustrating the compression-prediction link rather than a practical replacement for neural language models.