DeepSeek-Coder-V2 + ContinueDev + ShellGPT : The FREE & BEST LOCAL COPILOT EVER! (works without GPU)

In this video, I'll be creating an amazing Github Copilot Alternative with the new DeepSeek-Coder-V2 model. To do this I'll also be using the ContinueDev VSCode/NeoVim extension. I'll also be using the ShellGPT program to create a shell copilot as well. I'll also be utilizing the Free 5M Tokens API given by Deepseek. That means you can use it as a copilot for free and you don't need any GPUs to use it. This new model and copilot also beats GPT-4O, Claude-3, CodeQwen, Mixtral 8x22b, Mixtral 8x7b, GPT-4, Grok-1.5 & Gemini Code Assist.
-----
Key Takeaways:
🌟 DeepSeek Coder V2 Launched: DeepSeek recently launched two coding-specific models: a 16 billion parameter model and a 232 billion parameter model. These models outperform Codestral, GPT-4 Turbo, and Claude in most benchmarks, offering great alternatives for local Copilot setups.
🚀 Free API and Affordable Pricing: DeepSeek is offering 5 million free tokens for API usage, making it accessible for those without GPUs. The API pricing is also very affordable, allowing continuous usage for coding tasks without breaking the bank.
🔧 Setup Local Shell Copilot: Learn how to set up a local shell Copilot using the ShellGPT extension and DeepSeek's API. This setup mimics GitHub Copilot's functionality, providing efficient shell script suggestions and advanced code generation.
🖥️ VS Code Integration with DeepSeek: Integrate DeepSeek Coder with VS Code using the ContinueDev extension. This integration allows seamless code generation and chat interface functionality, enhancing your coding experience with powerful autocompletion and code suggestions.
💡 Utilize Both Models for Different Tasks: Use the locally hosted 16 billion parameter model for autocompletion and the 232 billion parameter model through the API for chat and advanced code generation. This dual setup optimizes performance for various coding needs.
📈 Superior Performance and Compatibility: DeepSeek Coder V2 supports over 300 programming languages and is licensed for commercial use, making it a versatile and powerful tool for developers. Its performance surpasses many local models, ensuring high-quality code generation.
🔥 Step-by-Step Installation Guide: Follow the detailed instructions to install ShellGPT, LiteLLM, and configure the DeepSeek API for local and API-based usage. This guide ensures you get the most out of DeepSeek Coder V2 quickly and efficiently.
-------
Timestamps:
00:00 - Introduction
00:04 - About Deepseek-Coder-V2
00:43 - Free 5 Million Tokens for API Usage
01:08 - Plan for usage as Copilot
01:26 - ShellGPT
01:49 - ContinueDev
01:55 - Installation (DeepSeek API)
02:22 - Installation (ShellGPT)
03:40 - Installation (ContinueDev)
05:49 - Local / Ollama Installation
07:15 - Ending

[@D4MN_SON_UR_GMA_FINE]

Awesome thank you I’m putting you on repeat and it is getting downloaded right now. Thank you.

[@mohammedel-cheikh6682]

Thank you for the videos. Wondering if you have used these tools for any of your automation projects. If you have, would love to watch an in-depth video as your tutorials are great :)

[@D4MN_SON_UR_GMA_FINE]

Awesome thank you

[@vickyrajeev9821]

Thank You Brother

[@StefanEdlich]

Can you make a video on how to make local inference (like with Ollama) using the graphics card GPU?

[@AdvanceGame-eo6yx]

❤❤❤❤❤

[@MaxF88]

Nice one! Any smaller models which is good with Python and javascript, suitable for 4GB VRAM?

[@sunrisestudios5206]

Can it be used for building a website/ can it host a website locally and keep coding and improving it or can it only write the code

[@D4MN_SON_UR_GMA_FINE]

I do have a question. Can I use the dolphin version from the GitHub repository on my llama instead of just the regular one that way it’s not censored

[@AICodeKing]

Yes, you can

[@MeinDeutschkurs]

What do you think about comparing multi language open source ai embedding models?

[@AICodeKing]

Will consider.

[@EladBarness]

awesome video that you! What about using it in a Devin-like experience, what’s the best option?

[@AICodeKing]

Aider scores the best in their benchmark for a devin like experience

[@cwirus99]

How does it compare to Claude 3.5?

[@Terry-m3c]

Installed it locally on ollama but the generated text is unrecognized. Tried to set the enocding to utf 8 but still same thing. Does anybody know what might be the problem ?

[@AICodeKing]

Try adding "reply output only in english"

[@Terry-m3c]

@@AICodeKing after changing the encoding to utf - 8 and adding "reply output only in english" I got success. Thank you!

[@ppbroAI]

Is supossed better than codestral, BUT codestral output decent json, can autocomplete (in the same model) amd it is able to use tools. Would be useful if you can compare those features ty.

[@StephenSmith304]

So it actually works for me to use deep seek coder v2 for auto complete with continue. I just put the title/provider/etc in the tabAutocompleteModel section. Unfortunately no matter what model I use for autocomplete it seems to time out a lot, but on the plus side deepseekcoder v2 runs a lot faster on my CPU than codeqwen7b despite having more params (16b) and so far has been giving me much better code. So when autocomplete does decide to work it's nice that I can use deepseek there too.

[@ppbroAI]

@@StephenSmith304 llm studip give quick completions, idk ollama

[@diakorudd7268]

I try to use lightningAI like you are doing in this video but its not allowing me to create the ~/.config/shell_gpt/.sgptrc . How did you work around this ? Thanks for uploading your gemstones of videos.

[@AICodeKing]

You'll need to first run the SGPT command then it will be automatically created.

[@diakorudd7268]

PermissionError: [Errno 13] Permission denied: '/teamspace/studios/this_studio/.config/shell_gpt/.sgptrc'

[@AICodeKing]

Try running it with Sudo or run "sudo chmod 777 /teamspace/studios/this_studio/.config/shell_gpt/.sgptrc"

[@diakorudd7268]

@@AICodeKing I got it to work in github codespaces but its still not working in lightning because its not allowing me to change permission of .config/ folder or any subdirectory .config/*

[@jaradaty88]

I hope they fix continue dev auto complete with api providers such as deepseek

[@Noaman2022]

Bro can i use deepseek on ollama local and connect to vacode to run localy without paid api ?

[@AICodeKing]

Yes, definitely

[@megafoxatron3rd521]

i am not seeing the 5m free tokken

[@AICodeKing]

They have removed that offer now

[@vhphan19]

How come I don't see 5M free tokens on my deepseek? 😞

[@AICodeKing]

Yes, seems like they have removed the Free Token Claim Option now. Although, Aider's site still stays that they are offering it (if you look at their free options). I had signed up on the day of DeepSeek-Coder launch and it was there and I still have 96% tokens left. So, I guess you're late now.

[@josersleal]

how is it free?

[@woofwoofification]

Is there a way to make it know your local code?

[@AICodeKing]

Yes you can add your context file to it

[@gani2an1]

Does it have memory?

[@AICodeKing]

The chat interface does have memory.

[@tranquiloteov]

Seems they dont give 5 millons free token anymore. I just created an account, and i have 0 free tokens T_T

[@user-ej7jj8ou6r]

Same here... not free... no trial

[@AICodeKing]

Yes, seems like they have removed the Free Token Claim Option now. Although, Aider's site still stays that they are offering it (if you look at their free options). I had signed up on the day of DeepSeek-Coder launch and it was there and I still have 96% tokens left. So, I guess you're late now.

[@Msyo_Jaber]

this is so usofel i larn alot but a ves issue