Browser LLM's saved localy for full offline work

Note: You have to run them once with internet access to grab all the files. Best to save the html file do your computer then run it with internet to grab the model and other files locally. Then shutdown internet and see if they work fully offline.

transformer.js webGPU deepseek R1 single web page

  • deepseek-r1-webgpu/deepseek-r1-webgpu-00.html Most stable DeepSeek-R1-web GPU. this is very impressive when fully offline!


    • Janus-Pro multimodal deepseekR1 text to image

  • janus-pro/janus-pro-to-image-00.html Most stable, can also work fully offline.


    • RAG Client-side RAG data set with embeddings

  • deep-rag/deep-rag00.html Most stable and fastest, works and runs both online and locally. You must run once with internet to grab the files locally.


    • Whisper-Base Audio to text

  • whisper/whisper00.html Most stable and fastest, works and runs both online and locally just not yet working locally without internet to start.


    • opus-mt-it-en text translate

  • opus/opus00.htmlNote yet working

    • transformer.js webGPU threeJS

    Not my normal single page application, needs a CDN and another assets file and some images.
  • https://hpssjellis.github.io/my-examples-of-transformersJS/public/webgpu-threejs/webgpu-threejs00.htmlAlways the most stable


  • My original huggingface.js demo page which some of the examples now don't work is at https://hpssjellis.github.io/my-examples-of-huggingfacejs/public/index.html
  • This github at https://github.com/hpssjellis/my-examples-of-huggingfacejs
  • This website index full link at
    https://hpssjellis.github.io/my-examples-of-huggingfacejs/public/index.html

  • javascript-ai-librariesAn interesting review of JS libraries

    • What's next? combining LLM's on the web and clientside, with micrcontroller groups using webSerial and IoT

    https://hpssjellis.github.io/google-should-make-tensorflowTinyJS/









    Bellow is my working index.html. Probably best to just ignore it.









    transformer.js webGPU deepseek R1 single web page

  • https://hpssjellis.github.io/my-examples-of-transformersJS/public/deepseek-r1-webgpu/deepseek-r1-webgpu-00.htmlDeepSeek-R1-web GPU
  • deepseek-r1-webgpu/deepseek-r1-01.html BAD
  • deepseek-r1-webgpu/deepseek-r1-02.html using cdn BAD
  • deepseek-r1-webgpu/deepseek-r1-03.html using modules still BAD
  • deepseek-r1-webgpu/deepseek-r1-04.html bad
  • deepseek-r1-webgpu/deepseek-r1-05.html laods model from cache, but does not work
  • deepseek-r1-webgpu/deepseek-r1-06.html issues
  • deepseek-r1-webgpu/deepseek-r1-07.html not yet
  • deepseek-r1-webgpu/deepseek-r1-08.html not yet
  • deepseek-r1-webgpu/deepseek-r1-09.html not yet
  • deepseek-r1-webgpu/deepseek-r1-10.html LOL starting to work
  • deepseek-r1-webgpu/deepseek-r1-11.html Working good Jan 30 2025 the stable one
  • deepseek-r1-webgpu/deepseek-r1-webgpu-12.html Name change just to beconfusing
  • deepseek-r1-webgpu/deepseek-r1-webgpu-13.html
  • deepseek-r1-webgpu/deepseek-r1-webgpu-15.html testing dType q4
  • deepseek-r1-webgpu/deepseek-r1-webgpu-16.html fixing progress
  • deepseek-r1-webgpu/deepseek-r1-webgpu-17.html New stable version
  • deepseek-r1-webgpu/deepseek-r1-webgpu-18.html messed up a bit but has tokens per second
  • deepseek-r1-webgpu/deepseek-r1-webgpu-19.html bad
  • deepseek-r1-webgpu/deepseek-r1-webgpu-20.html bad
  • deepseek-r1-webgpu/deepseek-r1-webgpu-21.html bad trying to make a stop button. gave up
  • deepseek-r1-webgpu/deepseek-r1-webgpu-22.html new stable includes top_p temperature etc
  • deepseek-r1-webgpu/deepseek-r1-webgpu-23.html Awesome. Stopping now works.
  • deepseek-r1-webgpu/deepseek-r1-webgpu-24.html
  • deepseek-r1-webgpu/deepseek-r1-webgpu-25.html new favorite and most stable, just some weird little download bug
  • deepseek-r1-webgpu/deepseek-r1-webgpu-26.html
  • deepseek-r1-webgpu/deepseek-r1-webgpu-27.html Some LaTex render improvements, hid hyperparameters
  • deepseek-r1-webgpu/deepseek-r1-webgpu-28.html My new favorite and Stable version. Does LaTex Math well
  • deepseek-r1-webgpu/deepseek-r1-webgpu-29.html
  • deepseek-r1-webgpu/deepseek-r1-webgpu-30.html updated to transformers@3.5.1 solving a bug
  • deepseek-r1-webgpu/deepseek-r1-webgpu-31.html

    • Janus-Pro multimodal deepseekR1 text to image

  • janus-pro/janus-pro-to-image-00.html Most stable if it ever works
  • janus-pro/janus-pro-to-image-01.html First attempt
  • janus-pro/janus-pro-to-image-02.html crap
  • janus-pro/janus-pro-to-image-03.html crap
  • janus-pro/janus-pro-to-image-04.html crap
  • janus-pro/janus-pro-to-image-05.html crap
  • janus-pro/janus-pro-to-image-06.html crap
  • janus-pro/janus-pro-to-image-07.html Finally the barebones works
  • janus-pro/janus-pro-to-image-08.html OK
  • janus-pro/janus-pro-to-image-09.html Trying to format it
  • janus-pro/janus-pro-to-image-10.html updated to transformers@3.5.1 solving a bug
  • janus-pro/janus-pro-to-image-11.html

    • transformer.js webGPU threeJS

    Not my normal single page application, needs a CDN and another assets file and some images.
  • https://hpssjellis.github.io/my-examples-of-transformersJS/public/webgpu-threejs/webgpu-threejs00.htmlAlways the most stable
  • ebgpu-threejs/webgpu-threejs02.htmlWorking
  • ebgpu-threejs/webgpu-threejs03.html
  • ebgpu-threejs/webgpu-threejs04.html

    • Whisper-Base Audio to text

  • whisper/whisper00.html Most stable if it ever works
  • whisper/whisper01.html bare bones testing
  • whisper/whisper02.html bad
  • whisper/whisper03.html bad
  • whisper/whisper04.html bad
  • whisper/whisper05.html bad
  • whisper/whisper06.html bad
  • whisper/whisper07.html bad
  • whisper/whisper08.html bad
  • whisper/whisper09.html OK
  • whisper/whisper10.html bad
  • whisper/whisper11.html bad
  • whisper/whisper12.html OK
  • whisper/whisper13.html bad
  • whisper/whisper14.html really bad all in one webworker
  • whisper/whisper15.html bad
  • whisper/whisper16.html bad
  • whisper/whisper17.html Works with audio-webworker
  • whisper/whisper18.html tiny
  • whisper/whisper19.html small
  • whisper/whisper20.html base
  • whisper/whisper21.html best so far using whisper tiny, only issue is the blob does not save to local storage
  • whisper/whisper22.html
  • whisper/whisper23.html
  • whisper/whisper24.html
  • whisper/whisper25.html
  • whisper/whisper26.html Good
  • whisper/whisper27.html New most stable hisper-tiny.en The English specific made it faster. Still not great
  • whisper/whisper28.html Not fast using whisper-small.en Not even going to try whisper-base.en
  • whisper/whisper29.html updated to transformers@3.5.1 but the bug is still active.
  • whisper/whisper30.html
  • whisper/whisper31.html

    • opus-mt-it-en text translate

  • opus/opus00.html Most stable if it ever works
  • opus/opus01.html bare bones
  • https://huggingface.co/Xenova/opus-mt-it-enOriginal huggingface
  • opus/opus03.html
  • opus/opus04.html
  • opus/opus05.html
  • opus/opus06.html
  • opus/opus07.html
  • opus/opus08.html

    • deep-rag draft

    Interesting thought. RAG does not have to be connected to your LLM. It is an information pre-processing step. perhpas there is a reason to keep it seperated from your client-side Browser LLM's. This page is just the RAG system
  • deep-rag/deep-rag00.html Most stable if it ever works
  • deep-rag/deep-rag01.html
  • deep-rag/deep-rag02.html
  • deep-rag/deep-rag03.html
  • deep-rag/deep-rag04.html
  • deep-rag/deep-rag05.html
  • deep-rag/deep-rag06.html ready for chunking
  • deep-rag/deep-rag07.html
  • deep-rag/deep-rag08.html
  • deep-rag/deep-rag09.html
  • deep-rag/deep-rag10.html
  • deep-rag/deep-rag11.html
  • deep-rag/deep-rag12.html New Stable
  • deep-rag/deep-rag13.html
  • deep-rag/deep-rag14.html
  • deep-rag/deep-rag15.html New Stable with JSON output and CSV load
  • deep-rag/deep-rag16.html
  • deep-rag/deep-rag17.html
  • deep-rag/deep-rag18.html
  • deep-rag/deep-rag19.html
  • deep-rag/deep-rag20.html
  • deep-rag/deep-rag21.html
  • deep-rag/deep-rag22.html
  • deep-rag/deep-rag23.html
  • deep-rag/deep-rag24.html
  • deep-rag/deep-rag25.html
  • deep-rag/deep-rag26.html
  • deep-rag/deep-rag27.html New Stable
  • deep-rag/deep-rag28.html Trying to view chunks and checking chunk length
  • deep-rag/deep-rag29.html
  • deep-rag/deep-rag30.html reduced the CSS for beginners
  • deep-rag/deep-rag31.html
  • deep-rag/deep-rag32.html looking at chunking issues of small chunks and download feedback
  • deep-rag/deep-rag33.html
  • deep-rag/deep-rag34.html Neww Stable solved weird chunking issue on csv load
  • deep-rag/deep-rag35.html
  • deep-rag/deep-rag36.html
  • deep-rag/deep-rag37.html
  • deep-rag/deep-rag38.html
  • deep-rag/deep-rag39.html
  • deep-rag/deep-rag40.html
  • My original huggingface.js demo page which some of the examples now don't work is at https://hpssjellis.github.io/my-examples-of-huggingfacejs/public/index.html
  • This github at https://github.com/hpssjellis/my-examples-of-huggingfacejs
  • This website index full link at
    https://hpssjellis.github.io/my-examples-of-huggingfacejs/public/index.html

  • javascript-ai-librariesAn interesting review of JS libraries
    • Watch out for limits. Basically on a small model I am getting about 20 web page refreshes and then I need a one hour gap.
    If you are getting hit by limits. right-click--> view source and put the page on your own site to see it work. Strange but I am even having success from my own harddrive doesn't even seem to need to be a "https" site.

    October 2024 I met both xenova (Joshua Lochner) and radames (Radamés Ajna) at the excellent WebAI2024 at Google campus MP1

    Random stuff.
    Simple page to show three links
  • deepseek-r1-webgpu/link-three01.html
    • body>