Thank you Dave for these great tutorials, as a non-native English speaker you speak English so evident that I can comprehend I'm stiil learning HTML and CSS Dave with your tutorials, your tutorials are so invaluable 🌻❤
I tried this and it's amazing. The only drawback is the response is really slow. Do you have any idea about that? I'm running on a MacBook Pro 2019 16inch Intel with 32GB RAM, 2.6GHZ 6-core CPU
I, also, have this issue. Sometimes it never responds and I have to cancel the question. Not having a good experience. How do I uninstall the 4GB LLM? Going back to ChatGPT4.
With the advent of *Apple Intelligence*, I am thinking there may be a way to integrate your code editor into the model like this. If and when its possible PLEASE make a video on that. 😄
Everybody has an opinion on which Linux distro is best, and all of us are 100% right. :) I use PopOS as a daily driver and suggest you give it a try. Zorin and Mint are also often suggested for somebody just coming over from Windows. Ubuntu is a good choice as well. Depending on your hardware, you should be able to boot from a USB stick and try any of these before you commit to an installation.
@@Peacemaker.404 If you can boot from USB for a test drive then it's easier to try them all and pick the one that best meets your needs. Unless you are doing something that requires some specific software that requires Windows, you will probably find them to all work for you. Then you can pick the one that looks/feels the best.
I really appreciate your videos! Could you recommend a large language model for coding that I can run locally without worrying about my computer slowing down?
You definitely deserve your entry to the Nirvana! (The thing is: you bring so much enlightenment to our world that if you do so, you will be missed a lot! Hopefully you volunteer to reincarnate on this planet to bring your light!)
What hardware are you running this with? I am running a 11th gen i7 with 32gb ram and SSD and unfortunately codeqwen is running at an absolute snails pace to point that it is unusable 😢
Same on macbook pro M1Pro 32g RAM - unusable. Few promps in terminal - and CPU goes to 100% load. Will try to move ollama with codeqwen to the server with GPU
thanks Dave, your always as ever on the money so to speak. Local LLMs may well be the next thing and privacy is king so this is well worth the time and investment. Thanks agai, your a ⭐cheers
@@DaveGrayTeachesCode yep, I've got deepseek-coder-v2 running already, thanks to your prompting me to take a look on now windows up to now I've been running on WSL so in order to get this to work I had to first stop ollama in WSL with syttemctl commands to stop and disable but that aside, running in windows now which is terrific, its even working with dart / flutter which is amazing
hey Dev, your tutorials are awsome and provide deep incites. i tried this with qwen2 modal but its not working as expected. i have setup everthing, but its very slow on my system, so better to go with chatgpt. but i wonder is there any lightweight application that we can use for auto code completion similar to copilot.
Cheers. Haven’t tried CodeQwen yet. I prefer LM Studio to Ollama if nothing else because it keeps the downloaded file in a for. That’s usable by other programs. Where as Ollama just hides the download intoweirdly named files.
I installed at 3:56 so not sure what you mean? I do mention that I previously had it installed so I did not get the splash screen again. I talk about what you should see.
@@DaveGrayTeachesCode Thanks for the reply, its a screen that tells users to install llama3 and run it as well as starcoder2. Easy spot to get lost. thanks!
Hello, I can see that you have python and js lessons on your channel, but what language has more opportunities to start job at least for free or for food:/? I learned python and Django before, but dropped it because people said that it's too slow, and i'm practicing MERN now, of course it feels a bit harder than Django. And i'm a bit confused now, was it right choice or wrong to drop python web
Thanks for the feedback! I do want to compare more of these solutions. I really like this one because it is local and you choose your own model - which allows you to upgrade as models improve.
I don't think you can target a framework like that unless someone specifically creates an LLM for it. Just go with the rankings for coding like I do with EvalPlus in this video.
Thank you this is an amazing video! 🎉 I have a question concerning it's ability or not to understand the full context of a codebase/project/folder or is it limited to selected code/limited number of open files
from continue extension docs: Completions don't know about my code We are working on this! Right now Continue uses the Language Server Protocol to add definitions to the prompt, as well as using similarity search over recently edited files. We will be improving the accuracy of this system greatly over the next few weeks.
Which model you choose and how many tokens it supports will impact that, but yes, continue supports that. Reference: docs.continue.dev/walkthroughs/codebase-embeddings
I am using MacBook m1 and did everything as mentioned in the video. It seems like vscode is lagging after applying all the configuration. Also, auto suggestion is not working for me. Is there anything m missing?
I didn't add it to my Mac yet to compare, but installed locally shouldn't create a lag. I think you've got plenty of power, too. Maybe a quick restart of VS Code? As mentioned, I did have to restart Windows.
Thank you for the great content. I have a private ollama server running in my company with dedicated video card. Is it possible to connect Continue to this server in my LAN? Ollama is not installed on my local workstation, it is on a server here in my environment.
There is tab auto completion. That is the setting I changed in the config - but yeah, trying to fit it all into 10 minutes or less, I didn't demo everything.
This sounds pretty powerful. I'll give it a try. Does anyone have a model they would recommend for the following programming languages: Assembly x64, Kotlin, JavaScript C/C++, HTML/CSS and Java?
I've tried a few models locally, and I really like having an AI assistant locally. Too bad I can't run it on my laptop. Anyone know of any good free or low cost alternatives?
As an answer to myself, I just found out about Chat RTX from Nvidia...it can run local LLM's and train on your local data as well. Now I only need a extension to integrate it into VS Code.
Hello sir l'm learn the python from your channel but i forgot everything after some days and after a long time I'm learning again python please give some suggestions to learn python
Don't rush. Just learn one thing and then try to apply it. The more you use it, the easier it is to remember. Just learn a little something new every day.
@@DaveGrayTeachesCode ok thank you sir but in python course I couldn't understand the operators you wouldn't explain the some operators . How to understand that and were you learn python and other new technologies
@@TSMohanKumar there are many python resources available. If what I said or taught did not stick with you, sometimes it is good to reference other sources of information. Putting all of these together will help your understanding.
@@TSMohanKumar you don't have to understand everything. Be kind to yourself and everything will fall into place the more you code. So don't stress if you don't understand something
Excuse me please - *Oi!* - :D - thank u hello! Ummm... so at 3:07 u explain "...go back to your Terminal in VS Studio Code and run that with Ollama" - and I understand that, and that's fine... But what NOBODY EVER EXPLAINS or more accurately what it seems more like to be WHAT EVERYONE FAILS TO RESPONSIBLEY ADVISE is - {C:\in\which\gosh\darnit\someone\tellme\please\directory} ??? - it's either one of these so let's try multiple choice for ease and efficiency for all: Do I run the install command here (*): A) .../Users//> * B) .../Users//CODE/> * C) .../Users//CODE/Projects/> * D) .../Users//CODE/Projects/myProjectsRootDir> * or, simply even here: E) C:/ * ! ! ! Or are model packages installed as global dependencies so it matters none what directory your terminal is currently pointing to when sending the ollama run command? If so, is this true for all (or "most") ai chat models or just in this case for ollama? THANK YOU IN ADVANCE TO ANYONE HOW CAN CLARIFY THIS FOR ME
wow - erm there's a part in there that translates a bit harsh lol - wasn't tryna be... wasn't pointing fingers at anyone... sorry if it came across any bit a little fiery
Someone else commented about how Ollama stores the LLM models with weird filenames, but they didn't say where. They did mention another choice of theirs that makes the models available to other software, too. Might look for that comment from earlier today.
Hmm, I'm on a PC and that hasn't happened. Strange indeed. It does seem to be a bit power hungry according to some comments, but mine is far from a recent PC.
Can confirm as well. I did the same test as Dave and it took about 1-2 minutes for the entire code to be written using codeqwen. Don't know if it's CPU or GPU that slows it down. I have 32gb RAM a 9400k and a AMD 580RX. So I guess at least my PC is not as fast as Dave's. But fun little piece of software to use. I also use Cody Free version as an alternative to Copilot, as I'm still learning programming and only do a little coding at my current job, so I don't feel it's worth forking out $ yet.
You've already had time to compare both? I want to compare others. Can you choose your own free model with Codeium? If so, it comes down to the extension features and UI comparison.
Good sir@@DaveGrayTeachesCode 1. Codeium doesn't require me to install and run a local ollama instance which hogs the compute and memory of my MBA making it unbearable to code. 2. Codeium free version doesn't give the ability to choose models but the Pro model has ability to choose GPT 4 models. Haven't tried that, don't have that kinda money but free is fast and good enough for my use case - Web dev. 3. Codeium setup is just 1 click. 🙂
@@unknotmiguel Umm if someone actually reads through the privacy section of these Copilots-as-service apps, they don't send your code back to their server lol. But they send metrics and that too depending on Paid version can be turned off!!! Also, PRIVACY is a MYTH. And privacy in code is a joke. My poor 2 cents.
Thank you Dave for these great tutorials, as a non-native English speaker you speak English so evident that I can comprehend I'm stiil learning HTML and CSS Dave with your tutorials, your tutorials are so invaluable 🌻❤
Glad I can help!
One important caveat (which wasn't mentioned) is that you need a pretty good PC to use this comfortably.
When it comes to running ai model locally, it's kind of a given that you will need a strong computing power.
GPU
Thank you for the tutorial. I found out that deepcoder v2 runs faster (and better according to the leaderboard) than qwen on my computer.
Thanks, that worked really well. Used it to write some nodejs / express / ejs code, a combo i hand not done and it got the code running nicely
Glad to hear that!
Things you are showing here no longer align with Continue when I installed it today in Oct 24. :(
Exactly what I was looking for. Thank you very much!
Will take a look, but I'm finding hard to believe there's something out there to replace supermaven. Copilot is trash compared to it now.
Yes so many new things. I like this because you can keep changing the models as they improve and the extension is constantly being updated as well.
I agree supermaven is awesome. Switched from copilot
I tried it and instantly loved it-it's even better than my Tabnine Pro subscription! Thanks for letting me know it exists
Whats the models name?
Supermaven? It's fast, but the code it produces is... Not that great... It completely neglects my coding style. Cursor is much better
I tried this and it's amazing. The only drawback is the response is really slow. Do you have any idea about that? I'm running on a MacBook Pro 2019 16inch Intel with 32GB RAM, 2.6GHZ 6-core CPU
I have the same issue. Did you manage to fix that?
intel macs are slow not the ARM versions
I, also, have this issue. Sometimes it never responds and I have to cancel the question. Not having a good experience. How do I uninstall the 4GB LLM? Going back to ChatGPT4.
@q_pato_o I have a 64gb arm mac and 8b will still run slow. It because it is using the CPU.
M1 8gb ARM macbook air can run 8gb no problem. for me.
Thanks so very much, Local LLMs are the next big thing. so thankful for this video. Just earned a new sub here 🌟🌟🌟🌟🌟
You're welcome!
Thank you. Exactly what i was looking for.
Thank you for discovering new AI code completion. I have a question: what are the minimum system requirements to run this local code completion?
Is the model aware of the entire codebase of the project?
You can do this with Aider also and while it has a map of the repo, it only knows the files you add
how it compares to codium?
From Nigeria Africa! Love your content.
Thank you!
Thank you, Dave. Have you tried cursor? How do you like it?
Do you have some gpu for powering that beast
This is great! Thank you! How does this compare with Codeium Free? Which one is better?
what is hardware requirments for run locally ?
Does it have inline suggestion?
Yes it does
With the advent of *Apple Intelligence*, I am thinking there may be a way to integrate your code editor into the model like this. If and when its possible PLEASE make a video on that. 😄
Great work Dave!!! 🎉🎉🎉🎉❤❤❤
Thanks so much!!
hey dave, do you use linux and if you do which distro do you use, i'm trying to switch from windows.
I used to. My favorite for a long time was Debian. Then everyone went to Ubuntu. Last I knew, Ubuntu was easiest to switch to.
@@DaveGrayTeachesCode thanks, i'll try that.
Everybody has an opinion on which Linux distro is best, and all of us are 100% right. :)
I use PopOS as a daily driver and suggest you give it a try. Zorin and Mint are also often suggested for somebody just coming over from Windows.
Ubuntu is a good choice as well.
Depending on your hardware, you should be able to boot from a USB stick and try any of these before you commit to an installation.
@@drkgumby Yes, exactly. I was confused between PopOS and Mint, also I've used Ubuntu in the past.
@@Peacemaker.404 If you can boot from USB for a test drive then it's easier to try them all and pick the one that best meets your needs. Unless you are doing something that requires some specific software that requires Windows, you will probably find them to all work for you. Then you can pick the one that looks/feels the best.
I really appreciate your videos! Could you recommend a large language model for coding that I can run locally without worrying about my computer slowing down?
Thank you Dave, good quality content right there ❤
You're welcome!
You definitely deserve your entry to the Nirvana!
(The thing is: you bring so much enlightenment to our world that if you do so, you will be missed a lot! Hopefully you volunteer to reincarnate on this planet to bring your light!)
Thanks a lot Dave!
What hardware are you running this with? I am running a 11th gen i7 with 32gb ram and SSD and unfortunately codeqwen is running at an absolute snails pace to point that it is unusable 😢
Same on macbook pro M1Pro 32g RAM - unusable. Few promps in terminal - and CPU goes to 100% load.
Will try to move ollama with codeqwen to the server with GPU
Thanks alot Dave, As per the request for upcoming videos I would be happy if you could provide honoJS, react / nextJs and postgress tutorial
Many possibilities there! Thanks for the request!
A Nextjs frontend with Django backend tutorial would be nice 🙏
thanks Dave, your always as ever on the money so to speak. Local LLMs may well be the next thing and privacy is king so this is well worth the time and investment. Thanks agai, your a ⭐cheers
You're welcome and I agree!
@@DaveGrayTeachesCode yep, I've got deepseek-coder-v2 running already, thanks to your prompting me to take a look on now windows
up to now I've been running on WSL so in order to get this to work I had to first stop ollama in WSL with syttemctl commands to stop and disable but that aside, running in windows now which is terrific, its even working with dart / flutter which is amazing
I am using Cursor AI currently,...i have the bare feeling that those guys use Qwen in the background...
Does this also do code completion / suggestions?
Yes
I installed deepseek but that extention is not picking up it self can you tell me what i can write in config file? Please help
hey Dev, your tutorials are awsome and provide deep incites. i tried this with qwen2 modal but its not working as expected. i have setup everthing, but its very slow on my system, so better to go with chatgpt. but i wonder is there any lightweight application that we can use for auto code completion similar to copilot.
hi mate, does this work for regular visual studio too? I cant stand vscode haha
This is crazy. This will eventually greatly reduce the need for programmers.
Would this be able to access all the code in the codebase? Or only the opened files?
Yes you can use @codebase or @folder. Reference: docs.continue.dev/walkthroughs/codebase-embeddings
Hey Dave, can you share the hardware specs of the machine where you tried this? I am curious if my MBA M1 with 16GB can handle it or not.
I'm using a Windows PC that's about 3 years old. Your m1 is good, but RAM might be a concern. I added a lot of RAM to render my videos faster.
Hi anything for vs studio?
Mine is "indexing" for my first code generation. I'm guessing it's a one time thing though.
Cheers. Haven’t tried CodeQwen yet. I prefer LM Studio to Ollama if nothing else because it keeps the downloaded file in a for. That’s usable by other programs. Where as Ollama just hides the download intoweirdly named files.
Good info! Thanks!
Dave, you missed the entire install on VSCode 4:18
I installed at 3:56 so not sure what you mean? I do mention that I previously had it installed so I did not get the splash screen again. I talk about what you should see.
@@DaveGrayTeachesCode Thanks for the reply, its a screen that tells users to install llama3 and run it as well as starcoder2. Easy spot to get lost. thanks!
Ur a big help mate
If you want to try any Local LLM make sure you have a Powerful workstation
Installed in VSCode, but can’t get the chat in intellij
Great Dave!
Thank you!
Hello, I can see that you have python and js lessons on your channel, but what language has more opportunities to start job at least for free or for food:/? I learned python and Django before, but dropped it because people said that it's too slow, and i'm practicing MERN now, of course it feels a bit harder than Django. And i'm a bit confused now, was it right choice or wrong to drop python web
There are jobs for both. Difficult to say which would be better. Both are among the most popular programming languages.
Great! More of this please!
Thanks for the feedback! I do want to compare more of these solutions. I really like this one because it is local and you choose your own model - which allows you to upgrade as models improve.
@@DaveGrayTeachesCode 🙏
thanks for your tutorial
Can we set the token size since there is a limit with copilot. Thank you another wonderful tutorial.
It will only be limited by the limits of the model you choose.
@@DaveGrayTeachesCode Thank you.
Thanks.Which is the best Ollama LLM for nextJS 14 free / licensed ?
I don't think you can target a framework like that unless someone specifically creates an LLM for it. Just go with the rankings for coding like I do with EvalPlus in this video.
Hi! Thanks for this video!
Is the Ollama and codeqwen 100% free, has IntelliSense and autocomplete aside from code chat?
Yes to all
thank you dave, you are amazing
You're welcome!
Thanks Dave
Thank you this is an amazing video! 🎉 I have a question concerning it's ability or not to understand the full context of a codebase/project/folder or is it limited to selected code/limited number of open files
from continue extension docs:
Completions don't know about my code
We are working on this! Right now Continue uses the Language Server Protocol to add definitions to the prompt, as well as using similarity search over recently edited files. We will be improving the accuracy of this system greatly over the next few weeks.
Which model you choose and how many tokens it supports will impact that, but yes, continue supports that. Reference: docs.continue.dev/walkthroughs/codebase-embeddings
Does this all happen locally? No posting your code to some server in Vietnam somewhere?
Right! And yes, 100% local.
@@DaveGrayTeachesCode This assumes we have a really good computer?
I am using MacBook m1 and did everything as mentioned in the video. It seems like vscode is lagging after applying all the configuration. Also, auto suggestion is not working for me. Is there anything m missing?
I didn't add it to my Mac yet to compare, but installed locally shouldn't create a lag. I think you've got plenty of power, too. Maybe a quick restart of VS Code? As mentioned, I did have to restart Windows.
you need a helluva PC to run this, but let's give it a try... if anything i can just buy a bit more RAM ;p
Thank you for the great content.
I have a private ollama server running in my company with dedicated video card.
Is it possible to connect Continue to this server in my LAN?
Ollama is not installed on my local workstation, it is on a server here in my environment.
I don't know, but if you find out, please share here. Interesting!
Will the Assistan able to auto complete react and next.js code ??
Yes, it has auto complete. The setting I change in the video is tab auto complete.
Thanks Dave.❤❤❤
Did I miss something? I did not see code completion as you type your own code.
There is tab auto completion. That is the setting I changed in the config - but yeah, trying to fit it all into 10 minutes or less, I didn't demo everything.
Thanks Dave!
You're welcome!
awesome. thanks.
Can we do video on detail of ecma script used forjs
Thank you teacher.
This sounds pretty powerful. I'll give it a try. Does anyone have a model they would recommend for the following programming languages: Assembly x64, Kotlin, JavaScript C/C++, HTML/CSS and Java?
Dave will ever make a course about the new nextjs futures
I'm waiting until Next.js 15 is promoted beyond release candidate. Then I will consider it.
I've tried a few models locally, and I really like having an AI assistant locally. Too bad I can't run it on my laptop. Anyone know of any good free or low cost alternatives?
Yeah running locally will take some power. If your laptop runs short on that, you might want to look at services that don't run locally.
As an answer to myself, I just found out about Chat RTX from Nvidia...it can run local LLM's and train on your local data as well. Now I only need a extension to integrate it into VS Code.
have auto complete?
Hello sir l'm learn the python from your channel but i forgot everything after some days and after a long time I'm learning again python please give some suggestions to learn python
Don't rush. Just learn one thing and then try to apply it. The more you use it, the easier it is to remember. Just learn a little something new every day.
@@DaveGrayTeachesCode ok thank you sir but in python course I couldn't understand the operators you wouldn't explain the some operators . How to understand that and were you learn python and other new technologies
@@TSMohanKumar there are many python resources available. If what I said or taught did not stick with you, sometimes it is good to reference other sources of information. Putting all of these together will help your understanding.
@@TSMohanKumar you don't have to understand everything. Be kind to yourself and everything will fall into place the more you code. So don't stress if you don't understand something
I was wondering if you could make a postgresql tutorial
Nice request! I've been thinking about that 🙌
you're legend
For me it seems way slower than it should be. Running on AMD 4800U with 16gb of ram and the starcoder2:3b model.
@@skewlines4152 might need more RAM to speed it up.
@@DaveGrayTeachesCode I"ll try it on my gaming pc tomorrow and report the difference.
I would like to see python to adobe pdf
thank you so much
You're welcome!
you are big brother thank you
Want to see docker, docker compose from scratch
Compliment approved
Excuse me please - *Oi!* - :D - thank u hello! Ummm... so at 3:07 u explain "...go back to your Terminal in VS Studio Code and run that with Ollama" - and I understand that, and that's fine... But what NOBODY EVER EXPLAINS or more accurately what it seems more like to be WHAT EVERYONE FAILS TO RESPONSIBLEY ADVISE is - {C:\in\which\gosh\darnit\someone\tellme\please\directory} ??? - it's either one of these so let's try multiple choice for ease and efficiency for all:
Do I run the install command here (*):
A) .../Users//> *
B) .../Users//CODE/> *
C) .../Users//CODE/Projects/> *
D) .../Users//CODE/Projects/myProjectsRootDir> *
or, simply even here:
E) C:/ *
! ! !
Or are model packages installed as global dependencies so it matters none what directory your terminal is currently pointing to when sending the ollama run command? If so, is this true for all (or "most") ai chat models or just in this case for ollama?
THANK YOU IN ADVANCE TO ANYONE HOW CAN CLARIFY THIS FOR ME
wow - erm there's a part in there that translates a bit harsh lol - wasn't tryna be... wasn't pointing fingers at anyone... sorry if it came across any bit a little fiery
Someone else commented about how Ollama stores the LLM models with weird filenames, but they didn't say where. They did mention another choice of theirs that makes the models available to other software, too. Might look for that comment from earlier today.
@@DaveGrayTeachesCode did u ever locate that post/choice?
It is so cool but it crashes my pc every few minutes. Even though my PC meets the requirements.
Hmm, I'm on a PC and that hasn't happened. Strange indeed. It does seem to be a bit power hungry according to some comments, but mine is far from a recent PC.
Awesome 😍....
you're the best (must be exhausting)
Legend
Better resalts
Well its good. But takes away many system resources.
Could be a drawback of keeping everything local depending on machine power. I'm not usually running many other tasks while coding/chatting with it.
but you need strong strong CPU
Mine isn't too strong. I'd say mid.
You need a high end computer to run that locally though. My mac m1 lags lol so I'll stick to copilot
Hmm, I have a mid PC running it without issues. I do have extra RAM though. Might make a difference.
Can confirm as well. I did the same test as Dave and it took about 1-2 minutes for the entire code to be written using codeqwen. Don't know if it's CPU or GPU that slows it down. I have 32gb RAM a 9400k and a AMD 580RX. So I guess at least my PC is not as fast as Dave's. But fun little piece of software to use. I also use Cody Free version as an alternative to Copilot, as I'm still learning programming and only do a little coding at my current job, so I don't feel it's worth forking out $ yet.
This is lagging so much in my M2 MacBook Air
From what I'm hearing in the comments, it seems like more RAM helps. I have a PC that's a few years old but lots of RAM.
I think Codeium is better even with free version
You've already had time to compare both? I want to compare others. Can you choose your own free model with Codeium? If so, it comes down to the extension features and UI comparison.
Good sir@@DaveGrayTeachesCode
1. Codeium doesn't require me to install and run a local ollama instance which hogs the compute and memory of my MBA making it unbearable to code.
2. Codeium free version doesn't give the ability to choose models but the Pro model has ability to choose GPT 4 models. Haven't tried that, don't have that kinda money but free is fast and good enough for my use case - Web dev.
3. Codeium setup is just 1 click. 🙂
The goal usually is to run locally due to privacy of code concerns, I believe...
@@unknotmiguel Umm if someone actually reads through the privacy section of these Copilots-as-service apps, they don't send your code back to their server lol. But they send metrics and that too depending on Paid version can be turned off!!!
Also, PRIVACY is a MYTH. And privacy in code is a joke. My poor 2 cents.
Спасибо огромное
You're welcome!
It's old news. 😢
Definitely not for everyone. But yeah, we do hear about things at different times.
first time hearing about this