TE
TechEcho
Home24h TopNewestBestAskShowJobs
GitHubTwitter
Home

TechEcho

A tech news platform built with Next.js, providing global tech news and discussions.

GitHubTwitter

Home

HomeNewestBestAskShowJobs

Resources

HackerNews APIOriginal HackerNewsNext.js

© 2025 TechEcho. All rights reserved.

Talk = GPT-2 and Whisper and WASM

189 pointsby tomtheover 2 years ago

12 comments

atum47over 2 years ago
<p><pre><code> &gt; whisper: number of tokens: 2, &#x27;Hello?&#x27; &gt; gpt-2: I want to have you on my lap. </code></pre> this GPT-2 better chill
tomtheover 2 years ago
This would of course be even more fun with ChatGPT, but it is a nice and funny demo of their whisper.cpp library. The second video is worth watching: <a href="https:&#x2F;&#x2F;user-images.githubusercontent.com&#x2F;1991296&#x2F;202914175-115793b1-d32e-4aaa-a45b-59e313707ff6.mp4" rel="nofollow">https:&#x2F;&#x2F;user-images.githubusercontent.com&#x2F;1991296&#x2F;202914175-...</a>
评论 #33894259 未加载
评论 #33893863 未加载
swyxover 2 years ago
<p><pre><code> The total data that the page will have to load on startup (probably using Fetch API) is: - 74 MB for the Whisper tiny.en model - 240 MB for the GPT-2 small model - Web Speech API is built-in in modern browsers </code></pre> cool but im now wondering what it would take to bring this down enough to put this in real apps? anyone talking about this?
评论 #33895467 未加载
评论 #33894867 未加载
评论 #33896976 未加载
评论 #33899815 未加载
评论 #33895691 未加载
评论 #33899363 未加载
评论 #33896807 未加载
Terrettaover 2 years ago
Listening to that demo, it&#x27;s incredible how far we&#x27;ve come!<p>Or, not.<p>Racter was <i>commercially</i> released for Mac in December 1985:<p><i>Racter strings together words according to &quot;syntax directives&quot;, and the illusion of coherence is increased by repeated re-use of text variables. This gives the appearance that Racter can actually have a conversation with the user that makes some sense, unlike Eliza, which just spits back what you type at it. Of course, such a program has not been written to perfection yet, but Racter comes somewhat close.</i><p><i>Since some of the syntactical mistakes that Racter tends to make cannot be avoided, the decision was made to market the game in a humorous vein, which the marketing department at Mindscape dubbed &quot;tongue-in-chip software&quot; and &quot;artificial insanity&quot;.</i><p><a href="https:&#x2F;&#x2F;www.mobygames.com&#x2F;game&#x2F;macintosh&#x2F;racter" rel="nofollow">https:&#x2F;&#x2F;www.mobygames.com&#x2F;game&#x2F;macintosh&#x2F;racter</a><p><a href="https:&#x2F;&#x2F;www.myabandonware.com&#x2F;game&#x2F;racter-4m&#x2F;play-4m" rel="nofollow">https:&#x2F;&#x2F;www.myabandonware.com&#x2F;game&#x2F;racter-4m&#x2F;play-4m</a><p>It&#x27;s only amazing that chatGPT backed by GPT-3 is the <i>first thing since then</i> to do enough better that <i>everyone</i> is engaged.<p>I owned that in 1985, and having studied AI&#x2F;ML previously I&#x27;ve been (and remain something of) an AGI skeptic. But now in 2022, I finally think <i>“this changes everything”</i> ... not because it&#x27;s AI, but because it&#x27;s making the application of matching probabilistic patterns across mass knowledge practical and useful for everyday work, particularly as a structured synthesis assistant.
评论 #33896878 未加载
评论 #33896106 未加载
Rickvstover 2 years ago
I implemented whisper + chatgpt + pyttsx3 and it worked. But then suddenly the chatgpt wrapper that I found on github stopped working.<p>edit: whisper is awesome
评论 #33894956 未加载
评论 #33895935 未加载
rahimnathwaniover 2 years ago
I&#x27;m curious how they chose between:<p>A) ggml <a href="https:&#x2F;&#x2F;github.com&#x2F;ggerganov&#x2F;ggml&#x2F;tree&#x2F;master&#x2F;examples&#x2F;gpt-2" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;ggerganov&#x2F;ggml&#x2F;tree&#x2F;master&#x2F;examples&#x2F;gpt-2</a><p>B) Fabrice Bellard&#x27;s GPT2C <a href="https:&#x2F;&#x2F;bellard.org&#x2F;libnc&#x2F;gpt2tc.html" rel="nofollow">https:&#x2F;&#x2F;bellard.org&#x2F;libnc&#x2F;gpt2tc.html</a>
评论 #33899016 未加载
iandanforthover 2 years ago
Technically this seems to work, and mad props to the author for getting to this point. On my computer (MacBook Pro) it&#x27;s very slow but there are enough visual hints that it&#x27;s thinking to make the wait ok. I have plenty of complaints about the output but most of that is GPT-2&#x27;s problem.
boredemployeeover 2 years ago
offtopic but what are the real limitations of gpt2 vs gpt3? (i know that gpt2 is free)
评论 #33895716 未加载
评论 #33895653 未加载
bilaterover 2 years ago
I&#x27;ve been thinking of doing something like this but hooked up with ChatGPT&#x2F;GPT-3-daviinci003. Obviously model will not load in the browser but we cna call the API. Could be a neat way to interact with the bot.
simonwover 2 years ago
Anyone found a sentence that GPT-2 returns a good response for? My experiments have been not great so far.<p>(LOVE this demo.)
hanozover 2 years ago
What are some good things to try? I can&#x27;t get any sense out of it at all so far.
评论 #33899333 未加载
thundergolferover 2 years ago
This guy&#x27;s doing really great work recently. Keep it up, Georgi!