Links:<ul><li>my little <a href="https://github.com/ggerganov/whisper.cpp/pull/1947">whisper.cpp bug fix</a></li><li><a href="https://github.com/ggerganov/whisper.cpp/issues/1941">why the logits were calculated inconsistently</a></li><li>wav2vec2 on <a href="https://arxiv.org/abs/2006.11477">arxiv</a> and <a href="https://huggingface.co/docs/transformers/en/model_doc/wav2vec2">huggingface</a></li><li>the openai <a href="https://openai.com/research/whisper">whisper</a> asr model announcement</li><li><a href="https://arxiv.org/abs/2204.05424v3">beam search patience</a></li><li>top notch <a href="https://distill.pub/2017/ctc/">guide to CTC</a></li><li>the <a href="https://www.cs.utexas.edu/~eunsol/courses/data/bitter_lesson.pdf">bitter lesson</a> (and in <a href="https://twitter.com/beyang/status/1773021689342767351">meme form</a>)</li></ul>Errata:* I referred in the show to LLMs as encoder-decoder models. Most modern LLMs are decoder-only. * I messed up readability at Google. It means approvability, apparently. 🤷

Links:my little whisper.cpp bug fixwhy the logits were calculated inconsistentlywav2vec2 on arxiv and huggingfacethe openai whisper asr model announcementbeam search patiencetop notch guide to CTCthe bitter lesson (and in meme form)Errata: * I referred in the show to LLMs as encoder-decoder models. Most modern LLMs are decoder-only. * I messed up readability at Google. It means approvability, apparently. 🤷

Links:<ul><li>my little <a href="https://github.com/ggerganov/whisper.cpp/pull/1947" rel="noopener noreferrer">whisper.cpp bug fix</a></li><li><a href="https://github.com/ggerganov/whisper.cpp/issues/1941" rel="noopener noreferrer">why the logits were calculated inconsistently</a></li><li>wav2vec2 on <a href="https://arxiv.org/abs/2006.11477" rel="noopener noreferrer">arxiv</a> and <a href="https://huggingface.co/docs/transformers/en/model_doc/wav2vec2" rel="noopener noreferrer">huggingface</a></li><li>the openai <a href="https://openai.com/research/whisper" rel="noopener noreferrer">whisper</a> asr model announcement</li><li><a href="https://arxiv.org/abs/2204.05424v3" rel="noopener noreferrer">beam search patience</a></li><li>top notch <a href="https://distill.pub/2017/ctc/" rel="noopener noreferrer">guide to CTC</a></li><li>the <a href="https://www.cs.utexas.edu/~eunsol/courses/data/bitter_lesson.pdf" rel="noopener noreferrer">bitter lesson</a> (and in <a href="https://twitter.com/beyang/status/1773021689342767351" rel="noopener noreferrer">meme form</a>)</li></ul>Errata:* I referred in the show to LLMs as encoder-decoder models. Most modern LLMs are decoder-only. * I messed up readability at Google. It means approvability, apparently. 🤷&nbsp;

Iota: whisper.cpp

Substantive, in-depth technical conversations about software engineering (broadly construed).

Share Iota: whisper.cpp

Sign up to save your podcasts

Iota: whisper.cpp

Iota: whisper.cpp