- Видео 89
- Просмотров 54 898
Practical AI through Prototypes
США
Добавлен 31 авг 2018
Learn AI through Projects, Prototypes, and Tutorials from Mozilla Design Technologist Mike Heavers
Local AI code completion using @Mozilla 's Llamafile + React.
This video shows you how to install a local AI coding model that supports FIM (fill in the middle) code completion, and run a browser-based app running next.js and react to implement the AI's suggestions.
Code can be found here: github.com/heaversm/llamafile-code-completion.
The llamafile used in this demo can be found here: huggingface.co/raincoder/deepseek-coder-1.3b-llamafile
You can find out more about Llamafile here:
github.com/Mozilla-Ocho/llamafile
Code can be found here: github.com/heaversm/llamafile-code-completion.
The llamafile used in this demo can be found here: huggingface.co/raincoder/deepseek-coder-1.3b-llamafile
You can find out more about Llamafile here:
github.com/Mozilla-Ocho/llamafile
Просмотров: 199
Видео
Make a No-Code Website in Minutes using Next.js / React / Vercel & Claude Artifacts
Просмотров 317Месяц назад
In this video we'll quickly walk through how to use Anthropic's Claude AI to generate no-code Next.js / React Javascript Web Apps that you can deploy with one click from a Github repo to Vercel. Claude generates a great foundation from which to build more scalable custom web apps, but also works for self-serve workflows for any repetitive task you find yourself / your team / company doing. Next...
Make bespoke web apps instantly just by describing them with Claude AI Artifacts.
Просмотров 1032 месяца назад
Claude Artifacts is a productivity game changer. In this video I'll show you how I made a transcript to text document converter with just a simple AI prompt. Claude can make custom react components for your existing codebase, or generate hosted public web apps that anyone can use. We'll also look at how other developers such as Simon Willison are using Claude for a multitude of use cases. 0:00 ...
VotebookLM - Become an informed voter with the help of AI
Просмотров 722 месяца назад
NotebookLM is a way to upload information about a topic, and find information within them with the help of AI. In this video you'll see how to use it to create your own custom voting guide, complete with candidate info, endorsements, citations, and suggested topics.
Boost Your AI Results with this free tool from Mozilla
Просмотров 2842 месяца назад
Formulaic is a prompt management tool from Mozilla's Innovation Studio that helps improve AI-generated results by making prompt engineering and model comparison cheaper, better, and easier. In this video you’ll learn how to iterate on prompts, compare models, and work with contextual information from your own personal files and data. By the end, you’ll understand how Formulaic can transform the...
Run the new Llama 1b parameter AI model local, private, & free with Mozilla Llamafile.
Просмотров 8482 месяца назад
This video walks through downloading, installing, and running the new, fast Llama 3.2 1b AI model from Meta on your own computer. It also includes a sneak peak at Llamafile's new chat interface.
First look at Google's ImageFX generative AI editing capabilities
Просмотров 8584 месяца назад
Walking through the photoshop-esque new AI editing capabilities of Imagen 3 via Google's Trusted Tester program. In this video you can manipulate different variations of your text prompt, as well as highlight areas of the photo that you want to modify and provide a description for just that area of modification. #generativeai , #texttoimage
Quick Tutorial: How to rotate a texture on a single face in blender
Просмотров 2 тыс.6 месяцев назад
In a previous video, I showed you how to apply a texture to a single face of an object in blender. In this video, I'll quickly show you how to rotate that face using the UV mapping - rotate feature. How to apply a texture to a single face: ruclips.net/video/Rx-aOHCfTOw/видео.html
Automate AI Research with Crew.ai and Mozilla Llamafile
Просмотров 12 тыс.7 месяцев назад
In this video we'll walk through how to set up crew.ai with Mozilla LLamafile to run a local large language model on your computer and automate multi-step tasks using a model of your choosing. Sample Code: github.com/heaversm/crew-llamafile List of Llamafile Models on hugging face: huggingface.co/models?library=llamafile Serper Search API: serper.dev/ CrewAI documentation - docs.crewai.com/how-...
First Look at Cohere's Coral Toolkit for RAG
Просмотров 5998 месяцев назад
Cohere's Coral AI Toolkit is an open source collection of prebuilt components to help developers quickly build RAG applications, in order to ask questions about your docs, or directly against coral LLM models. In this video we'll walk through installing and using the prebuilt components of the Coral Toolkit. About Me: I am a Staff Design Technologist on Mozilla's Innovation Team. All opinions a...
Measuring the carbon footprint of LLM training with the Impact Framework
Просмотров 1898 месяцев назад
In this video we'll use the LLM Carbon plugin for Green Software Foundation's Impact Framework to measure CO2 impact of training LLMs like Meta's Llama-7B. I'll then go through each of the inputs required to make the calculations. Lastly, we'll look at some of the other ways of using the Impact Framework. LLM Carbon Plugin - github.com/Jjing-Liang/LLMCarbon /tree/main Llama Training Research - ...
Link Gradio Figma Components to Code with Chromatic Storybook Connect
Просмотров 2518 месяцев назад
In this video we show you how to link our Gradio Figma UI components to their Storybook Counterparts with Chromatic's Storybook Connect Plugin. By doing this you can leverage all of the capabilities of Storybook directly within Figma, and easily leverage the UI comparison / regression testing capabilities of Chromatic within your build process. Links: Gradio UI for Figma Components: www.figma.c...
Carbon Hack 24 Final Project Submission
Просмотров 1188 месяцев назад
A summary of all the things we built for the Green Software Foundation's Carbon Hack 24, focused on the Impact Framework. Project Submission Github Issue: github.com/green-Software-Foundation/hack/issues/97 Getting Started Github Code Repo - github.com/heaversm/carbonhack-if Getting Started Medium Article - medium.com/@iretehamdani/how-to-use-the-impact-framework-to-measure-the-environmental-im...
Carbon Hack 24 Final Project Submission
Просмотров 138 месяцев назад
A summary of all the things we built for the Green Software Foundation's Carbon Hack 24, focused on the Impact Framework.
Approaches and Considerations to Measuring Website Energy Usage
Просмотров 678 месяцев назад
In this video, we'll look at two methods of measuring website carbon emissions - by either attributing a portion of overall energy usage to the webpage, or by calculating the energy used by each of a webpage's component parts - from the data, to the networking, to the device, and the embodied carbon of all of the equipment used to run it. We'll discuss tradeoffs and considerations of both appro...
How to use the Impact Framework to calculate your website's carbon emissions
Просмотров 1198 месяцев назад
How to use the Impact Framework to calculate your website's carbon emissions
GSF Impact Framework Plugin Registry Prototype
Просмотров 379 месяцев назад
GSF Impact Framework Plugin Registry Prototype
GSF Impact Framework Pipeline Builder Prototype
Просмотров 639 месяцев назад
GSF Impact Framework Pipeline Builder Prototype
Getting Started with the Green Software Impact Framework
Просмотров 2509 месяцев назад
Getting Started with the Green Software Impact Framework
Build with Google's New AI Text to Image API: Imagen
Просмотров 1,1 тыс.9 месяцев назад
Build with Google's New AI Text to Image API: Imagen
Every developer should learn this! Website Accessibility testing using a screen reader
Просмотров 63410 месяцев назад
Every developer should learn this! Website Accessibility testing using a screen reader
Can these free text summarizers stack up against chatGPT?
Просмотров 9210 месяцев назад
Can these free text summarizers stack up against chatGPT?
Infinite AI Zoom Animation using just Photoshop
Просмотров 1,2 тыс.11 месяцев назад
Infinite AI Zoom Animation using just Photoshop
2 AI Chatbots Debate the Future of AI
Просмотров 27411 месяцев назад
2 AI Chatbots Debate the Future of AI
How to prepare and annotate data for use in ML model fine tuning with Label Studio
Просмотров 1,2 тыс.11 месяцев назад
How to prepare and annotate data for use in ML model fine tuning with Label Studio
Chat with AI in Discord with an open source local LLM
Просмотров 5 тыс.Год назад
Chat with AI in Discord with an open source local LLM
Ditch ChatGPT and try one of these downloadable options
Просмотров 1,3 тыс.Год назад
Ditch ChatGPT and try one of these downloadable options
Autonomous AI agents and scraping for data gathering
Просмотров 2,9 тыс.Год назад
Autonomous AI agents and scraping for data gathering
Talk on Building Education into the Prototyping Process at Mozilla Innovation
Просмотров 58Год назад
Talk on Building Education into the Prototyping Process at Mozilla Innovation
why doesnt it work for me?
Can you please please create an video for how to implement llama.cpp in a react native application for Android?
I there im not sure if you read the comments but im getting a strange error right about 9:22, i get that the term "ie" is not recognized in my terminal, if you can help i would appreciate!
Just got the awnser the IE command is only applicable to MacOS, if on windows you need to use "if-run"
thank you!
Funny that no one trusts the government, no one is happy with the way it acts, and yet somehow it is our sacred duty to continue to participate in and perpetuate it. 🤡🌎
Voting is the most violent act someone can commit in their lifetime. This little noted anomaly about voting is directly related to the modern conception of the State as an entity deriving its grant of authority to act from the consent of the governed. The aura of legitimacy surrounding the government's actions is enhanced by the perceived role of voting as an expression of the “people's will.” Whether non-threatening or violent, the authority for each and every one of the government's actions is presumed to flow from the consent of the people through the electoral process. School children are told this from their earliest years. The idea the State derives its power to act from the consent of the people sounds romantic. Few people, however, are aware that by definition the State’s power is for the specific purpose of engaging in acts of violence. No grant of power is necessary for anyone, or any organization to act peacefully. This is no secret among scholars, and sociologist Max Weber's definition of the State is considered one of the most authoritative: “A state is a human institution that claims the monopoly of the legitimate use of physical force within a given territory. ... The state is considered the sole source of the `right' to use violence." [1] The legitimizing impact of voting on the government's exercise of power intimately involves voters in the use of that power. Which means that non-voters tend to delegitimize the exercise of a government's power as an expression of the “will of the people.” So if no one voted in an election or only a small percentage of people did, the government couldn't profess to be empowered to act as an agent of the “people's will.” Without the protective cover provided by voters, the government would have no pretense to act except as a law unto itself. Consequently, the government's actions and the voters who legitimize them are linked together. Thus at a minimum, voters are spiritually involved in every act engaged in by the government. Including all violent acts. This involvement in the government's violence isn't, tempered by the nominal peacefulness of a person’s life apart from voting. By choosing to vote a person integrates the violence engaged in by the government as a part of their life. This is just as true of people that didn't vote for a candidate who supports particular policies they may disagree with, as it is for those that did. It is going through the motion of voting that legitimizes the government to act in their name, not who or what they vote for. This means that the violence perpetrated by any one person pales in scope or significance when compared to that which is authorized to be taken by the government in the name of those who vote. The combined ghoulish violence of every identifiable serial killer in American history can't match the violence of even one of any number of violent actions taken by the government as the people's representative. A prominent example of this is the economic sanctions imposed on Iraq after the Gulf war in 1991. These sanction prevented Iraq from rebuilding its destroyed sanitation, water, and electric power infrastructure that were specifically targeted by the U. S. military for destruction. Supported and enforced by the U. S., these sanctions are credited by UNICEF and other organizations with contributing to the gruesome deaths of an estimated 3,000 to 5,000 children a month for over 8-1/2 years. [2] All voters share in the government's contribution to the unnecessary deaths of these children caused by disease and a reduced standard of living. So the over half-a-million deaths of innocent children in Iraq in the years after 1991’s Gulf war are on the blood stained hands of every voter in the U.S. The same dynamic of voter involvement in government atrocities is true of the many hundreds of civilian deaths caused by the bombing of Yugoslavian cities in the spring and summer of 1999 that the United States participated in. This was a small scale recreation of the atomic bombing of the non-military cities of Hiroshima and Nagasaki in August 1945. Hundreds of thousands of innocent women, children and old people were killed from the initial bomb blasts and the long-term effects of radiation exposure. [3] Those bombings had been preceeded by the U.S. military’s killing of many hundreds of thousands of non-combatants during the firebombings of Tokyo, Hamburg, Dresden and Berlin. All of those people were killed in the name of the voters that had elected the Roosevelt administration in 1944 by a landslide. Voting, like a missile fired at an unseen target many miles away, is a long-distance method of cleanly participating in the most horrific violence imaginable. So declining to vote does much more than cause a statistical entry on the non-voting side of a ledger sheet. It is a positive way for a person to lower their level of moral responsibility for acts of violence engaged in by the government that they would never engage in personally, and that they don’t want to be committed in their name as a voter. Non-voting is a positive way for a person to publicly express the depth of their private belief in respecting the sanctity of life, and that violence is only justified in self-defense. -Hans Sherrer, 1999
Stop saying crap
@@jackbauer322 I appreciate the suggestion my friend, but until I see some coherent refutation of the reasoning provided by Mr. Sherrer, I am disinclined to acquiesce to your request.
I tried the SRT to JSON transcript converter link in the description and it has gone missing and returns a 404 error. Do you have a GitHub link?
Thanks - I've updated that space. The github repo is github.com/heaversm/srt-to-json-converter
I just finished viewing the video and thanks to the detail I am looking forward to walking through duplicating what you've done for my learning. I'm happy that I saw this before you decide to cut it down, and for me I would not recommend over tightening it too much. It's just fine. IMO the value I got out of it is the process from end to end and that means a lot, which I believe that is what you were trying to achieve. Thank you very much for taking the time to share.
Awesome, thank you!
Why won’t Mozilla add AI bookmark editing to Firefox. Auto truncation, character removal, word count reduction. And auto sort to folder by subject or topic. Minimizing and storing bookmarks is the web’s biggest time waster.
these are great suggestions! I don't work on Firefox, but you can add these ideas here: connect.mozilla.org/, or join the AI discord: discord.gg/FjUu3KWu
Pretty straight forward 👍🏻
thank you ... New Subscriber ...i myself am curently pototyping this !
Nice! Let me know what you end up doing with it.
I am mainly focused on local 'contextual RAG' projects using Ollama and the various models that it supports. I look forward to your next video!
More local A.I. prototyping !
ty bro
Hi, many thanks for this video, it was only usefull one I could find on this topic. Did you do the comparison with the other api options out there? I couldn't find it and would love to see it! Futhermore, I'm trying to implement the imagen api in my flutter app. Could your setup be used to have my app call an api hosted on huging face instead of directly accessing a web page? Would love to hear your thoughts.
Probably should've mentioned this is in beta. Don't you think?
you talk about running locally but using openai... sad
Yeah this is a fair point. I should do another version where I use an open model
Do you know something about the data protection rules of serper for using their Google API?
I think generally it’s not so much what tool you use as what you’re scraping and what the intended use is. Most sites have their own stipulations or licenses on what you can and can’t do with their site / content
So a free perplexity?
Great video. I am still getting charged for gpt-4o usage when my llamafile seems to be working fine? Is that expected? Also, how can I switch the code to point to gpt-4o-mini, to even further minimize my openAI API costs? THANKS A BUNCH!
This is great, thanks for this
Having a blast with your content Mike. We're doing real similar things on yt 🫡 thanks for the awesome vids
Wow - ton of content, I'll check this out. Thanks for the comment.
@@MikeHeavers ah just finding my feet. Its good fun!
Great video, really helpful. THX
But you used an Open AI key and Serper key so this might not be running entirely local right ?
The RAG bit - grabbing content from the internet, will require being online, correct. Inference is local, provided you are using Llamafile.
Does anybody know how to translate that git to a container in Unraid?
No
awesome content dude!
If I want to fine-tune Ollama, can the Discord bot customize ".gguf"?
Well, you don't fine tune Ollama - you fine tune whatever model you're running (the .gguf file you mention). Ollama just serves the model. You can fine tune gguf models - llama.cpp has a script (example here: github.com/ggerganov/llama.cpp/tree/master/examples/finetune) for this. But you fine tune on data, so I guess what your bot would be doing is inserting data into a text file or something, and you would programmatically run the script you program to fine tune on that data. But you can't continuously fine tune, really - it takes a long time (supposing you have a lot of data) - which is why companies only release LLMs every so often - it takes them weeks or months for large parameter models. More appropriately, what you'd probably want to do is RAG, where you use a pre-trained model and point it at your own data - I have other videos on how to do RAG with various methods, including this one, on the Coral AI Toolkit: ruclips.net/video/Nn16y7iHp48/видео.html
but dont I still have to pay for tokens?
You don't have to pay for tokens if you run a local model
Now everyone needs a new computer and a Cuda graphics card which are massively expensive due to crypto mining and now AI servers. Local runs way too slow on my 3-4 year old laptop. Will have to see if new Intel and AMD chips with embedded NPU's provide any support for multiple LLMs run on local machines.
Fair point - performance on local is not as good as running on cloud infrastructure. Seems like "AI-enabled" PCs will be the new trend.
Not good. Does not show the problems of crew ai working with Ollama or any other lllm. Crewai persistently asks for open ai key. The good I discovered Mozilla lllm server thank you. Crew ai is really bad.
So what is better than Crewai?
You don't have to use openAI or the API key - you can just remove it from the code. The Ollama file sample from the github repo shows you how to use Ollama. Note that Ollama is not an LLM - it just allows you to run LLMs locally.
@@mandelafoggie9359 You can try autogpt if you want - I found it harder to use.
@@practical-ai-prototypes Thank you I will check that out, again. I think it still asks for a key, even a fake key. Even if you want to use ollama.
Very good vid - thanks!
Tks
Love being able to run this locally. Great vid 👏👏👏
Update - I made an `app-input.py` script that allows you to create your own agent and task just by answering some questions in the command line.
Seems very useful! Is there an update video for this?
@@JofnD no - but same instructions, just run `python app-input.py` from the command line.
Nice, thanks
i get the error: [Shard #0] [INFO] Shard ready [Shard #0] [DEBUG] Making request to localhost:11434/api/show [Shard #0] [ERROR] Error 404 Not Found: POST /api/show [Shard #0] [ERROR] Error 404 Not Found: POST /api/show
Great tutorial! Is there a way to have NPM running has a service instead of commande line?
This is already possible with the plugin - the instructions are here: github.com/mekb-turtle/discord-ai-bot - follow the (non-docker) instructions, and start the service with 'npm start'
Error [TokenMissing]: Request to use token, but token was unavailable to the client. when i tried to run it that appears me, and i already check the token and is correct
The same.
me too
i've spent 2 days searching for their image editing function in python, 0 documentation had to check every page of their github to find it in python, in node there is no documentation about image editing at all, HELP !
Hello, thank you so much for this easy an simple step by step tutorial! I just did everything and in the end i get confronted with an error: "Error [ShardingReadyDied]: Shard 0's process exited before its Client became ready." Checked around on froums etc. but did not semm to find the solution. Do you maybe know how i can fix this? I think i did exactly the same as you in the Video. Thank you so much!
Also having the same issue here.
Did you run "npm install" ? Thats what I had to do.
did you ever fix this?
@@Tntlover Hello all, yes I started with doing the whole bot setup on the discord dev site again form scratch and got me a new Token - (refreshed it) then it worked. Im only 99% sure it was the cause of it tho, since that was the first and only time i used ".js"
Awesome
Thanks for the explanation! Is there a way to run your own model weights that are not in gguf format? I have a model on HF separated in 2 .bin files
I think you'll need to convert it to GGUF - here's a few resources for that: github.com/ggerganov/llama.cpp/discussions/2948 www.substratus.ai/blog/converting-hf-model-gguf-model/ sarinsuriyakoon.medium.com/convert-pytorch-model-to-quantize-gguf-to-run-on-ollama-5c5dbc458208
hey we can't do the editing on the image by passing the image? basically user pass the image and prompt to edit the image and image will be edited. i read the docs to they write about we can do that but not how, pls help.
Great suggestion, maybe I’ll do a video on that
Explain what you try to achieve, not what you do. Cut the details.. Make the video shorter. Less code, this is even indigestible for chatgpt.
Thanks for the feedback. Will think about ways to cut this down.
Boring . Sorry
All good. What would make it better?
❗ 'promosm'
Nice demo app. Wild times ahead
Wild times right now!
thank you.
Thanks for watching
Never thought that you can have like this locally, just wondering what are the other drawback of this aside from the size of this file
My prediction is that centralized LLMs will become obsoleted because they are logically crippled by political correctness. The only ways to prevent public facing LLMs from taking objective data and coming to politically incorrect conclusions is to make the LLM too stupid to do logic, or make it too lazy to do it at all. This is why decentralized open source uncensored LLMs will eventually overtake all of the commercial and centralized closed source LLMs since the decentralized LLMs have no restrictions on parroting the party line on command and claiming things like that there are five lights when there are only four.
Nice summary! Thanks for taking us thru Kierdorf et al's paper👍
Yes, the idea of precision agriculture is a niche that came up in my AI research and something I definitely intend to do blogs on, so this is on the right track!
Great - will keep an eye out for it!