Phi-3+ContinueDev+Ollama: STOP PAYING for Github's Copilot with this NEW, LOCAL & FREE Alternative

In this video, I'll be telling you that how you can create a free and self hosted local alternative to Github's Copilot & Gemini Code Assist that can run on any computer even without a GPU and is extremely fast. It is based on Microsoft's new opensource LLM PHI-3-mini-3.8b model. This alternative can be used with any opensource LLM such as Mixtral 8x22b, Mixtral 8x7b, GPT-4, Grok-1.5 & Gemini Code Assist.
[Key Takeaways]
💻 Discover how to create a local AI coding assistant using Microsoft's Phi-3 model, a powerful alternative to GitHub's Copilot that's on par with Llama 3 8b and GPT-3.5.
🔧 Learn how to set up a fast and local Copilot alternative that can do everything Copilot can, without relying on third-party APIs or self-hosting large AI models.
💡 Get started with Ollama and the ContinueDev VSCode extension to create a seamless AI-powered coding experience that integrates with your existing workflow.
📊 Take advantage of the Phi-3 model's 128K context version to work with larger code bases and improve your overall coding productivity.
💻 Explore the capabilities of this local AI coding assistant, including code generation, auto-completion, and code explanation, all within your local environment.
👍 Find out why this Copilot alternative is accessible to most computers, making it a great option for developers who want a local AI coding solution without the need for powerful hardware

[@avinav44]

I've been binge-watching your videos. You seriously address issues I've had for a long time. I truly appreciate your content. You should have 1 million subscribers for these videos, man.

[@de2wep]

Thx for very useful information! Great video

[@basilbrush7878]

Wow, impressive. I have a high spec Mac, so I will also try models with more parameters like Mixtral 8x22b using Ollama

[@AICodeKing]

Go for it!

[@salehmir9205]

LLama3 8b is already more powerful than Mixtral 8x22b

[@basilbrush7878]

@salehmir9205 I tested both on CrewAI Agents and additionally WizardLM 8x22b was most impressed with WizardLM 8x22b results. Of course, this is subjective

[@NLPprompter]

yes... it is really hard when naming variables.... it's even more harder to remember them! i need assistant AI for that really! can't wait next one

[@egmluvr]

Damn I wasn't liking that Adam Sandler Spaceman voice at first...but it's relaxing and I was so focused on the video. You killed A.D.D.

[@AICodeKing]

Haha, thanks for the honest feedback

[@egmluvr]

@@AICodeKing Subscribed!

[@kenrock2]

U mean that alien spider right..? Lolz

[@programmertelo]

thanks for very useful information

[@mheetu3909]

Is there a way to feed it the entire project's code? Sometimes code from multiple files is needed to understand the full scope of the problem... while I understand no model has infinite context sizes (except maybe MemGPT?), the versatility of these AI coding assistants is drastically limited by their narrow field of vision.

[@AICodeKing]

You can indeed give it much more data than llama 3 because it has a 128k context window. But, It all depends on the size of your project.

[@mheetu3909]

@@AICodeKing I was asking, more-so, on how... not really if I could. 😕

[@kichirome1553]

Which model I can run with Rx560x for laptops?

[@AICodeKing]

You can run this for sure.

[@abuharis421]

may i know, macbook air m1 8gb are enough for this?

[@AICodeKing]

Yes, should be enough,

[@abuharis421]

Thank youu broo!@@AICodeKing

[@markjoshua8971]

What if i already have a gguf file? Can it run in ollama?

[@AICodeKing]

Yes

[@xbizuk1]

Works for Node.js code development?

[@AICodeKing]

Yes, it does.

[@salehmir9205]

Thank you so much for this. I'm having a good experience with this setup. However, I couldn't get their autocomplete feature working. I did read their docs too. Can you help with that?

[@AICodeKing]

You can go to their "Tab Autocomplete" section in their docs. There you'll find the chnages that needs to be done in config.json in order to make it working.

[@angrycoder]

Can someone tell me from which website did he download the phi3 model?

[@AICodeKing]

Ollama.

[@angrycoder]

Thanks

[@kayondoedward]

the link

[@settlece]

Hay there

[@warlockassim4240]

I just want no ollama running in background i catch continuedev autocomplet . Plz next video do that i looked at continuedev doc and config json api not works for autocomplet . So i installed cody for autocomplet but i prefer to learn autocomplet from continue

[@AICodeKing]

You can go to their "Tab Autocomplete" section in their docs. There you'll find the chnages that needs to be done in config.json in order to make it working.

[@ppbroAI]

for that you need a base model not an instruct model. Deepseek or starcoder