<feed xmlns='http://www.w3.org/2005/Atom'>
<title>TaSTT-Whisper.git/Designs/BeamSearch.md, branch master</title>
<subtitle>High-performance GPGPU inference of OpenAI's Whisper automatic speech recognition (ASR) model</subtitle>
<id>https://git.yummers.dev/TaSTT-Whisper.git/atom?h=master</id>
<link rel='self' href='https://git.yummers.dev/TaSTT-Whisper.git/atom?h=master'/>
<link rel='alternate' type='text/html' href='https://git.yummers.dev/TaSTT-Whisper.git/'/>
<updated>2023-02-28T02:03:02+00:00</updated>
<entry>
<title>Begin work on beam search decoding</title>
<updated>2023-02-28T02:03:02+00:00</updated>
<author>
<name>yum</name>
<email>yum.food.vr@gmail.com</email>
</author>
<published>2023-02-28T02:03:02+00:00</published>
<link rel='alternate' type='text/html' href='https://git.yummers.dev/TaSTT-Whisper.git/commit/?id=4f967dbdb7972ec52039bd3e3ce3e1e4cbcf6545'/>
<id>urn:sha1:4f967dbdb7972ec52039bd3e3ce3e1e4cbcf6545</id>
<content type='text'>
* ContextImpl.h puts prompts, previous prompts, probabilities, and
  probability IDs into vectors of size 1 or N_BEAMS, depending on
  the decoding strategy.
* Extend sampleBest and friends to return top N tokens, instead of
  just the top 1 token.
</content>
</entry>
</feed>
