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.

Janus Pro 1B running 100% locally in-browser on WebGPU

214 pointsby fofoz4 months ago

7 comments

ndr4 months ago
The image generation results are extremely poor, but it's exciting that it does _anything_ in the browser.
评论 #42854514 未加载
评论 #42855192 未加载
评论 #42853194 未加载
amelius4 months ago
The reason why this doesn&#x27;t work on Firefox:<p><a href="https:&#x2F;&#x2F;news.ycombinator.com&#x2F;item?id=41157383">https:&#x2F;&#x2F;news.ycombinator.com&#x2F;item?id=41157383</a>
thefirstname3224 months ago
Hi HN! We’re excited to launch JanusPro-AI, an open-source multimodal model from DeepSeek that unifies text-to-image generation, image understanding, and cross-modal reasoning in a single architecture. Unlike proprietary models, JanusPro is MIT-licensed and optimized for cost-efficiency—our 7B-parameter variant was trained for ~$120k, outperforming DALL-E 3 and Stable Diffusion XL in benchmarks like GenEval (0.80 vs. 0.67) 25.<p>Why JanusPro? Decoupled Visual Encoding: Separates image generation&#x2F;understanding pathways, eliminating role conflicts in visual processing while maintaining a unified backbone 2.<p>Hardware Agnostic: Runs efficiently on consumer GPUs (even AMD cards), with users reporting 30% faster inference vs. NVIDIA equivalents 2.<p>Ethical Safeguards: Open-source license restricts military&#x2F;illegal use, aligning with responsible AI development<p>please checkout the website: <a href="https:&#x2F;&#x2F;januspro-ai.com&#x2F;" rel="nofollow">https:&#x2F;&#x2F;januspro-ai.com&#x2F;</a>
pentagrama4 months ago
Happy to have these models running locally on a browser. However, the results are still quite poor for me. For example: <a href="https:&#x2F;&#x2F;imgur.com&#x2F;a&#x2F;Dn3lxsU" rel="nofollow">https:&#x2F;&#x2F;imgur.com&#x2F;a&#x2F;Dn3lxsU</a>
评论 #42854985 未加载
jedbrooke4 months ago
well it was a long shot anyway but it doesn’t seem to work on mobile. (tried on iOS safari on iPhone 11 pro)<p>a 1B model should be able to run in the RAM constraints of a phone(?) if this is supported soon this would actually be wild. Local LLMs in the palm of your hands
评论 #42853663 未加载
评论 #42854283 未加载
评论 #42857503 未加载
n-gauge4 months ago
I like the local running of this and learning about how it works.<p>Q:These models running in WebGPU all seem to need nodejs installed. I that for just the local &#x27;server side&#x27;, can you not just use a python http server or tomcat for this and wget files?
评论 #42855109 未加载
gavinguang4 months ago
<a href="https:&#x2F;&#x2F;www.janusproai.net&#x2F;" rel="nofollow">https:&#x2F;&#x2F;www.janusproai.net&#x2F;</a> This is janus pro website that can be tried online.