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.

WebGPU GPT Model Demo

154 pointsby b_mc2about 2 years ago
Tweet by creator: <a href="https:&#x2F;&#x2F;twitter.com&#x2F;willdepue&#x2F;status&#x2F;1649147091573432321" rel="nofollow">https:&#x2F;&#x2F;twitter.com&#x2F;willdepue&#x2F;status&#x2F;1649147091573432321</a>

11 comments

DustinBrettabout 2 years ago
It indeed works and loads quick. I am more interested currently in the Vicuna 7B example from <a href="https:&#x2F;&#x2F;mlc.ai&#x2F;web-llm&#x2F;" rel="nofollow">https:&#x2F;&#x2F;mlc.ai&#x2F;web-llm&#x2F;</a><p>Also instead of just &quot;Update Chrome to v113&quot; the domain owner could sign up for an origin trial <a href="https:&#x2F;&#x2F;developer.chrome.com&#x2F;origintrials&#x2F;#&#x2F;view_trial&#x2F;118219490218475521" rel="nofollow">https:&#x2F;&#x2F;developer.chrome.com&#x2F;origintrials&#x2F;#&#x2F;view_trial&#x2F;11821...</a>
评论 #35649332 未加载
FL33TW00Dabout 2 years ago
My 250M parameter model runs in 50ms&#x2F;token ;)<p>Releasing April 26th when Chrome 113 hits stable. Open source NPM library you can add to any project.<p>Preview here: <a href="https:&#x2F;&#x2F;twitter.com&#x2F;fleetwood___&#x2F;status&#x2F;1646608499126816799?s=20" rel="nofollow">https:&#x2F;&#x2F;twitter.com&#x2F;fleetwood___&#x2F;status&#x2F;1646608499126816799?...</a>
评论 #35650796 未加载
doodlesdevabout 2 years ago
<p><pre><code> &gt; WebGPU is supported in your browser! &gt; Uncaught (in promise) DOMException: WebGPU is not yet available in Release or Beta builds. </code></pre> Anyone using Chromium care to chime in?<p>If no one chimes in I might set up a Chromium browser up just to take a look at this, seems pretty cool.
评论 #35649125 未加载
评论 #35650165 未加载
ianpurtonabout 2 years ago
Question. I can see in the code the WGSL that&#x27;s needed to implement inference on the GPU. <a href="https:&#x2F;&#x2F;github.com&#x2F;0hq&#x2F;WebGPT&#x2F;blob&#x2F;main&#x2F;kernels.js">https:&#x2F;&#x2F;github.com&#x2F;0hq&#x2F;WebGPT&#x2F;blob&#x2F;main&#x2F;kernels.js</a><p>Could this code also be used to train models or only for inference?<p>What I&#x27;m getting at, is could I take the WGSL and using rust wgpu create a mini ChatGPT that runs on all GPU&#x27;s?
评论 #35654118 未加载
评论 #35651578 未加载
MuffinFlavoredabout 2 years ago
&gt; At the moment, WebGPT averages ~300ms per token on GPT-2 124M running on a 2020 M1 Mac with Chrome Canary.<p>How do ChatGPT on GPT-3.5 &#x2F; GPT-4 compare?
评论 #35650138 未加载
评论 #35649638 未加载
评论 #35650000 未加载
luizfelbertiabout 2 years ago
It&#x27;s really a shame that there is no 8-bit float support in the WebGPU spec. Even though few cards support it natively, it&#x27;d still massively benefit ML workloads.<p>Another annoying constraint but specific to wgpu (Rust&#x27;s implementation of WebGPU) is that it does not support f16 yet (which IS in the spec), only through SPIR-V passthrough...
tormehabout 2 years ago
Any way to run this kind of thing outside the browser? Chrome hasn&#x27;t enabled WebGPU on Linux yet.
评论 #35652219 未加载
kristianpabout 2 years ago
Not available in Firefox yet: <a href="https:&#x2F;&#x2F;bugzilla.mozilla.org&#x2F;show_bug.cgi?id=1602129" rel="nofollow">https:&#x2F;&#x2F;bugzilla.mozilla.org&#x2F;show_bug.cgi?id=1602129</a>
junrushao1994about 2 years ago
Is there any plan to support larger models than GPT-2?
eurekinabout 2 years ago
Omg, no pytorch&#x2F;wsl&#x2F;conda hiccups... This could save me some sleepless nights
samuelduranteabout 2 years ago
I didn&#x27;t understand why I need WebGPU to use WebGPT...