Coming soon: David and Dawid's channel! 🚀 Join Dawid and me as we explore Artificial Intelligence, Machine Learning, Deep Learning & Coding. Go subscribe: www.youtube.com/@davidanddawid // PDF Reference // Download PDF here: davidbombal.wiki/localdeepseek // David's SOCIAL // Discord: discord.com/invite/usKSyzb X: twitter.com/davidbombal Instagram: instagram.com/davidbombal LinkedIn: www.linkedin.com/in/davidbombal Facebook: facebook.com/davidbombal.co TikTok: tiktok.com/@davidbombal UA-cam: www.youtube.com/@davidbombal // MY STUFF // www.amazon.com/shop/davidbombal // SPONSORS // Interested in sponsoring my videos? Reach out to my team here: sponsors@davidbombal.com // MENU // 0:00 - Installing private offline LLMs 0:58 - Reasons to use private offline LLMs 03:37 - How to Install private offline LLMs 04:37 - Downloading a censored LLM 05:20 - Running prompts in censored LLM 06:34 - Downloading an uncensored LLM 07:21 - Running prompts in uncensored LLM 08:30 - Conclusion deepseek llama ai privacy cybersecurity local ai private ai private deep seek r1 uncensored ai uncensored models lm studio lmstudio Please note that links listed may be affiliate links and provide me with a small percentage/kickback should you use them to purchase any of the items listed or recommended. Thank you for supporting me and this channel! Disclaimer: This video is for educational purposes only. #deepseek #llama #ai #privacy
I find it funny how everyone is talking about censorship when it comes to DeepSeek, but had no problem with censorship when it comes to all the others.
I feel like we take a blind eye to privacy when it comes to tech in general, not just LLM like browser, social media, OS, etc… not to mention email too which is inherently not private
Thank you so much for this video. I watch a lot of your content, but i am very much a noob with a lot of tech stuff. I really appreciated you guiding me through it and now i can run LLMs locally. Its amazing. All i need now is to figure out how to run text to speech locally, and ill be super happy.
Great video for many begginers, I hope more people will use local models. That being said though... deepseek-r1distilled (no matter whether it is lllama or qwen) is technically NOT a DeepSeek R1. it is distilled large language model based on Qwen 2.5 using outputs from DeepSeek R1 but it's not R1. full R1 version that get's a really good results needs something around 1,342 GB of VRAM so multi-GPU expensive hardware
That was a quick video reply. I literally commented on the short with Deepseek privacy claims that you should try running it locally, and there you go 3 hrs later posting a video about it. 😅
@@davidbombal I can confirm as a subscriber for quite a time that you never disappoint. One of the few people who teaches things and always on point. THANKS DAVID 🙏🏽
I was curious about CheapSeek and privacy, so I did a local install and watched for any packets coming out of the VM. There was only ordinary local arp traffic. That said, the local version is still censored , that’s why I use Grok. Further, I ran it on Apple Silicon, and this restricted my ability to add an external GPU, so it was very slow on the M1
I love dolphin and use it as my "daily driver". I also created a WICKED Modelfile to make it even more useful and remove "the rails" that are put in place for safety reasons.
Regarding the “bias” and “censorship” I think it was more of an issue with the distilled model you were using. If I put in the same questions into DeepSeek r1:32b I get quite comprehensive answers about the dates in question.
I'm assuming when you install and download the models from LM studio you can configure the model size? The default looks like the 4.7GB one, but it does allow us to go bigger is it?
Hi, One thing I would love to know that wasn't covered in this video is what the downsides and/or privacy implications of the dolphin model is. You discussed the problems with deepseek calling home like it's ET, but i would love to know if dolphin has any privacy and data collection downsides like that. Great video as always. I would love to see even more, or potentially even a series, on how to get back more privacy in our everyday lives. Perhaps a series that tackles certain aspects with each video. One could be how to set up your home network to be private like with home run VPNs, changing router settings, etc. then the next could be on making windows more private and how to use some tools to gut it from all the 'spyware' that microsoft puts in there, as well as alternative OS options. Next could be how to do similar things for your phone's OS. Next could be migrating to software that is more private from ones commonly in use, e.g. email, password managers, messaging, etc. Essentially just a series showing how to take control of our data better in this day and age, because as much as a lot of us would want to do that, sadly there aren't too many resources on it that are definitive and well researched, so a series that walks people through how to do that for all aspects of everyday life would be a godsend. Anyway, thats just an idea, but I think it would fit your channel well as you already have dabbled in some things like that in the past, I just think that a structured series would be great for people who aren't too aware of where to start and don't want to do days worth or research, and sometimes still not have an easy answer. Keep up the amazing work. 😊
Hi david , I have a doubt.... whether the ai that running locally can be trained and used as a professional personal assistant. Can you make a video about it like.... " How can you train your own AI " actually its possible right!!😅
Hey David I am following you from past 2-3 year and I love your content. Can you please create a video on how to setup a network topology with SDN in it in GNS3. It will be very helpful. I watched your old videos on youtube but didn't got a clear idea. Actually I am currently working on a project and I need to setup it. I hope you will help.
I am shocked that uncensored LLMs are not illegal. The kinds of things they know... I love the idea of it but good god they will probably explain how to implode the entire universe into a singularity if you ask them nicely.
it would be nice to see the same method you used for deepseek app aplied to local llms interface programs... so much "freeness" gets one suspicious about information leaks... if one disconects internet from the laptop it can always encrypt the info and ones it reaches internet the info can be sent. just like it happens with android phones
Excellent. Love your channel. I dropped my smartphone, and the screen is completely shattered. How can I recover my data, such as my photos and documents, knowing that I can no longer enter my PIN to authorize access from my laptop when the phone is connected? Thanks in advance for any help you can provide.
Amazing dav. Please could share with us the laptop specs you use for this be4 we try it? Or you could give us the minimum requirements to run it smoothly. Thanks
Hi David, Which software/site are you using to record your videos along with screen sharing & blurring a particular part in video? also it would be good if you can tell about the Laptop that you are using for all these recordings.
Can you imagine we would be running AI assistants in our computers with just a few clicks? Now we just need some hoverboards for the full 2025 experience. Thanks for the tutorial David.
It's important to remember why engineers use cencoring filters on AI models in the first place. I understand if someone (including myself) doesn't want source code or security things cencored. However ; There are other queries to llm's that potentially could lead to people getting hurt or causing damage to the world we live in. If the really bad queries (hurting or killing) are cencored, it's for the sake of all of us who live in this world and want it to persist. Political bias and other such things are certainly present, but that's not the important thing. Censorship in significant areas is justified by good reasons. That's why it's a good thing that the dolphin models cannot deliver on OpenAI or DeepSeek level. Let's keep it that way, is my advice.
What about while you are using local model then there is no traffic.. But later, when you don't expect it to, activate background process and send data to server.. Is that stupid or maybe possible?
No, ive ran models for days with network monitoring software. Also you can always run it through a docker container or a virtual machine with 0 internet access.
You can remove all wires or wifi and use it locally. Models can't access internet in Lm Studio or oobabooga, it's in fact not easy to make if you need it to search for you.
Hello, dear David I ask u so many time to considere doing videos aboout the outcome of networking with quantum computing because the trasnport protocols will all be affected pls do me the favours thx any way
Once LM studio is installed is it possible to configure it in a client sever situation where other computers on the same LAN network can access the LLM of thier choice and use it via a web browser?
Thanks for this video, I followed it and installed LM Studio and Deepseek & Dolphin (same as your selections). I tried them both with a simple question: "How old is Gwyneth Paltrow" and got two different answers so I tried a Google search and got a third answer (which, by consensus of multiple links, appears to be correct). So how are we to trust anything from these AI searches?
You are getting different results based on the knowledge cutoff for each LLM. In the case of Dolphin shown in this video, the cutoff is 2023. For Deepseek shown in this video it is 2024. They are not accessing live data so you will get different results.
Hi, is there something like that but for which I could inject my data - files on which it will work? I think I can't demand more than to put text but if possible, would be awesome if such a "beast" could operate on my private data such as photos, Excel files, gpx files etc.
weird, seems my dolphin is censored and my deepseek isnt... I am running Ollama and i get the opposite responses from those two models. Asked Dolphin how to break into a car, absolutely refused, asked deepseek about tank man, more info than I needed, it even refers to him has heroic and couragous and talks negatively about the chinese government...
David bro, you have made several videos on installing nethunter rootless.i wonder if you make video on running tools( which tools not need root or can be run without root but have limits ) on rootless android.
abliterated models are better than "uncensored" ones. abliterated ones are fine-tuned to ignore, or go harder on the refusals. while uncensored ones tuned on data someone sees as "uncensored", and they still end up having corporate guardrails. abliterated ones just say anything. also they are easier to instruct, because they wont refuse the instructions.
Is it possible that when they created deep sink they put sleeper agents in that could be triggered? I read a paper last year about this. I just can’t remember where I read it.
David, I've seen a few videos where authors either use dockers, plus openwebui in Windows or Ollama to ran these models. Are there any significant differences? (performance?) Also, are all deepseek models censored? How to tell before you download a model if it is censored?
I tried this with a different program (Ollama). I noticed that DeepSeek verbose for its answers were way to loud. One simple question will trigger 10 min plus of text and didn’t know how to stop it, so I decided to delete everything instead. I’m gonna try this new way, but in case im encountering the same issue, what would it be the alternatives to fix it? Thanks
I'm a bit confused about the GPU offloading part. I'm using a Lenovo Legion 5 Pro laptop with an 8C/16T CPU, 32GB RAM and a 4070 (8GB). What is the "correct value" for that setting? I can go up to 32 model layers (whatever that means). I can see VRAM being used, but my GPU Utilization stays at 0%. Is that normal? Also I get quite a few unfinished queries from a point on.
Am really looking for an AI model that can help me with my studies am not that bright in the AI tech field but i wanna adapt and use this to up my carrer and studies would love a suggestion of which LLM that u advice me of using
Ollama is last one today which can't use model in parts, so getting by it one file model can take on certain speed at least 10 hours or more for original size.
@fontenbleau Well even if you ask a local deepseek model some historical questions it will refuse to answer so I believe there is a way to fine tune it so its uncensored and can say and answer whatever I ask it. Also if its isnt 671B model then it will not code as well as o1
Coming soon: David and Dawid's channel! 🚀 Join Dawid and me as we explore Artificial Intelligence, Machine Learning, Deep Learning & Coding.
Go subscribe: www.youtube.com/@davidanddawid
// PDF Reference //
Download PDF here: davidbombal.wiki/localdeepseek
// David's SOCIAL //
Discord: discord.com/invite/usKSyzb
X: twitter.com/davidbombal
Instagram: instagram.com/davidbombal
LinkedIn: www.linkedin.com/in/davidbombal
Facebook: facebook.com/davidbombal.co
TikTok: tiktok.com/@davidbombal
UA-cam: www.youtube.com/@davidbombal
// MY STUFF //
www.amazon.com/shop/davidbombal
// SPONSORS //
Interested in sponsoring my videos? Reach out to my team here: sponsors@davidbombal.com
// MENU //
0:00 - Installing private offline LLMs
0:58 - Reasons to use private offline LLMs
03:37 - How to Install private offline LLMs
04:37 - Downloading a censored LLM
05:20 - Running prompts in censored LLM
06:34 - Downloading an uncensored LLM
07:21 - Running prompts in uncensored LLM
08:30 - Conclusion
deepseek
llama
ai
privacy
cybersecurity
local ai
private ai
private
deep seek r1
uncensored ai
uncensored models
lm studio
lmstudio
Please note that links listed may be affiliate links and provide me with a small percentage/kickback should you use them to purchase any of the items listed or recommended. Thank you for supporting me and this channel!
Disclaimer: This video is for educational purposes only.
#deepseek #llama #ai #privacy
Horey 🎉
There is no video in d&d channel
@@CodeWithJoe watch the video. Dolphin 3.0 by Cognitive Computations is a different and uncensored video
@@ayushkandari2339working on it
@@CodeWithJoe he never claimed DeepSeek would not be censored. He provided two models and the 2nd one is the one that is not censored.
I find it funny how everyone is talking about censorship when it comes to DeepSeek, but had no problem with censorship when it comes to all the others.
There are different types of data collection and deepseeks flaws cause way worse information leakage then anything that exists like this in the usa
Every government sensor some information, but no other government sensors more information than any communist government.
Chinese applications employ extensive data collection methods, incorporating diverse data types, often exceeding those *other ones*
Sorry. I cannot answer that comment. I'm an UA-cam user that should only provide helpful and harmless responses.
Exactly. DeepSeek threatens all American apps and markets as we’ve seen in the market two weeks ago. SMH.
Hi David ,This year the content you're bringing to the channel is top notch ,I can't wait for more 💯
Thank you very much! I appreciate that!
I feel like we take a blind eye to privacy when it comes to tech in general, not just LLM
like browser, social media, OS, etc… not to mention email too which is inherently not private
Often very true and sad 😞
Including gaming with programs like Battleye anti-cheat
This is the most useful video on the internet currently
Thank you so much for this video. I watch a lot of your content, but i am very much a noob with a lot of tech stuff. I really appreciated you guiding me through it and now i can run LLMs locally. Its amazing. All i need now is to figure out how to run text to speech locally, and ill be super happy.
Thank you David for the valuable video- short & to the point.❤ Keep Educating US ❤❤
Hi david how about a video on how we run local LLM inside docker containers , and which LLM are best for this
Another vote as i run everything in docker
Great video for many begginers, I hope more people will use local models.
That being said though... deepseek-r1distilled (no matter whether it is lllama or qwen) is technically NOT a DeepSeek R1. it is distilled large language model based on Qwen 2.5 using outputs from DeepSeek R1 but it's not R1. full R1 version that get's a really good results needs something around 1,342 GB of VRAM so multi-GPU expensive hardware
That was a quick video reply. I literally commented on the short with Deepseek privacy claims that you should try running it locally, and there you go 3 hrs later posting a video about it. 😅
My audience demands. I deliver 😀
Thank you to both of you!
@@davidbombal I can confirm as a subscriber for quite a time that you never disappoint. One of the few people who teaches things and always on point. THANKS DAVID 🙏🏽
I was curious about CheapSeek and privacy, so I did a local install and watched for any packets coming out of the VM.
There was only ordinary local arp traffic.
That said, the local version is still censored , that’s why I use Grok.
Further, I ran it on Apple Silicon, and this restricted my ability to add an external GPU, so it was very slow on the M1
Thank you :)
I love dolphin and use it as my "daily driver". I also created a WICKED Modelfile to make it even more useful and remove "the rails" that are put in place for safety reasons.
Can you please share that model file or explain how we can make that ourselves?
I wanted to download it locally, but I really want to give my data to China at this point so guess I'll be using the online one.
Thank you for concentrating on deepseek and not the Western ai.
Now I can download deepseek.
Wow. I'd no idea. What an eye-opener. Thanks.
Glad it was helpful!
Love the videos boet. I feel like we are in a world/time where it is nearly impossible to have digital privacy.
Baie Dankie! 😀
It's funny that people forgot about Snowden and NSA etc.
Hello
Excellent video. Thanks.
awesome. i thought doing this would be much more difficult so i never looked into it. thanks!
Regarding the “bias” and “censorship”
I think it was more of an issue with the distilled model you were using.
If I put in the same questions into DeepSeek r1:32b I get quite comprehensive answers about the dates in question.
Make a video explaining how neural networks work on a fundamental level, easy to follow :D
Great suggestion!
@@davidbombal Can I download the DeepSeek on my phone and I will be secured
@@gentiljhack
but still you don't get much information as.that model is trained with limited information..
I'm assuming when you install and download the models from LM studio you can configure the model size? The default looks like the 4.7GB one, but it does allow us to go bigger is it?
Hi,
One thing I would love to know that wasn't covered in this video is what the downsides and/or privacy implications of the dolphin model is. You discussed the problems with deepseek calling home like it's ET, but i would love to know if dolphin has any privacy and data collection downsides like that.
Great video as always. I would love to see even more, or potentially even a series, on how to get back more privacy in our everyday lives. Perhaps a series that tackles certain aspects with each video. One could be how to set up your home network to be private like with home run VPNs, changing router settings, etc. then the next could be on making windows more private and how to use some tools to gut it from all the 'spyware' that microsoft puts in there, as well as alternative OS options. Next could be how to do similar things for your phone's OS. Next could be migrating to software that is more private from ones commonly in use, e.g. email, password managers, messaging, etc.
Essentially just a series showing how to take control of our data better in this day and age, because as much as a lot of us would want to do that, sadly there aren't too many resources on it that are definitive and well researched, so a series that walks people through how to do that for all aspects of everyday life would be a godsend. Anyway, thats just an idea, but I think it would fit your channel well as you already have dabbled in some things like that in the past, I just think that a structured series would be great for people who aren't too aware of where to start and don't want to do days worth or research, and sometimes still not have an easy answer.
Keep up the amazing work. 😊
to my knowledge dolphin is only run locally so there's no privacy implications.
Thanks David excellent piece of work.
I am just amazed how much knowledge and intelligence is in these models. Just a few gb of data and it seems to know a whole lot more than me.
Very informative. Thanks David.
Cool! Thank you for the information!
Was waiting for your video David sir
Hope you enjoyed it!
Thanks for sharing this info.
Hi david , I have a doubt.... whether the ai that running locally can be trained and used as a professional personal assistant. Can you make a video about it like.... " How can you train your own AI " actually its possible right!!😅
A great video, as always!
Great video david . Want more from you regarding use of this locally install llm for my other application or program
tiny models are really taking off, glad I decided to jump in, fine tuning seems to be the real magick tho, cant wait to start grinding away gpu cycles
You are the man David !
Thank you!
Epic video! Had to immediately install this on my linux test bench. 4790k gives 6tok/sec.
What is the biggest difference when it comes to size of llm is it just more knowledge and a bigger data set it can work from offline?
Give this man a medal
Hey David I am following you from past 2-3 year and I love your content. Can you please create a video on how to setup a network topology with SDN in it in GNS3. It will be very helpful. I watched your old videos on youtube but didn't got a clear idea. Actually I am currently working on a project and I need to setup it. I hope you will help.
Have you analized the traffic with Wireshark? Do they make suspicious calls?
I am shocked that uncensored LLMs are not illegal. The kinds of things they know... I love the idea of it but good god they will probably explain how to implode the entire universe into a singularity if you ask them nicely.
Thank you
it would be nice to see the same method you used for deepseek app aplied to local llms interface programs... so much "freeness" gets one suspicious about information leaks... if one disconects internet from the laptop it can always encrypt the info and ones it reaches internet the info can be sent. just like it happens with android phones
I hope that in the near future, OpenAI will consider going open source, like DeepSeek.
Thanks David.
I'll leave a comment here so i can backtrack to this video when I will need it later.
THanks !!
Excellent. Love your channel. I dropped my smartphone, and the screen is completely shattered. How can I recover my data, such as my photos and documents, knowing that I can no longer enter my PIN to authorize access from my laptop when the phone is connected? Thanks in advance for any help you can provide.
great video, thanks a lot.
so do these models now train by my prompts?
Amazing dav. Please could share with us the laptop specs you use for this be4 we try it? Or you could give us the minimum requirements to run it smoothly.
Thanks
Hi David,
Which software/site are you using to record your videos along with screen sharing & blurring a particular part in video?
also it would be good if you can tell about the Laptop that you are using for all these recordings.
Can you imagine we would be running AI assistants in our computers with just a few clicks? Now we just need some hoverboards for the full 2025 experience.
Thanks for the tutorial David.
Would be nice if you give a short description between different versions of the same LLM.
Hi David, you did a video on deepseek sending data to China when using your phone, on a public instance. Does private instance also do the same? TIA
Great vid
Thank you!
Good theres an open source version now
Hi David. Thanks for High quality content and videos.
Why there are two names in this llm names ? like qwen...llama3.0..?
i've found that deepseek r1 distilled doesn't do deepthinking, which is why i use deepseek personally, when i need general info, i use gemini though
Can we do this with GrapheneOS ?
That's a great suggestion!! 😀
Wouldn't the local LLM's knowledge be very limited with just gigabytes of data? I thought the knowledge base of LLMs was measured in petabytes
Grate video as always ... Is LM Studio safe?
Yes, LM Studio is safe. It was even mentioned in Andrej Karpathy's latest video on LLMs.
Is there any way to generate a prompt along with the response for stable diffusion to generate a scene picture?
It's important to remember why engineers use cencoring filters on AI models in the first place. I understand if someone (including myself) doesn't want source code or security things cencored.
However ; There are other queries to llm's that potentially could lead to people getting hurt or causing damage to the world we live in. If the really bad queries (hurting or killing) are cencored, it's for the sake of all of us who live in this world and want it to persist.
Political bias and other such things are certainly present, but that's not the important thing. Censorship in significant areas is justified by good reasons. That's why it's a good thing that the dolphin models cannot deliver on OpenAI or DeepSeek level. Let's keep it that way, is my advice.
Hello,what does the input count mean when you ask the llm a question?is it limiting the number of questions asked?.
Thanks
It's the number of tokens.
David, we need to learn memory forensics !!
Hey David, it would be nice of you if you helped us remove the restrictions from the DeepSeek R1 itself i.e uncensoring it.
Use a vpn.
@@noindemnity-en1ydnothing to do with a vpn.
What about while you are using local model then there is no traffic.. But later, when you don't expect it to, activate background process and send data to server.. Is that stupid or maybe possible?
given where Deepseek originated (PRC) is there concerns about running their model even locally and offline?
No, ive ran models for days with network monitoring software. Also you can always run it through a docker container or a virtual machine with 0 internet access.
should i use Docker or Lm studio to install it ?
David, is this local version of the Deepseek installation the one that you documented Deepseek sending back the information to China?
You can remove all wires or wifi and use it locally. Models can't access internet in Lm Studio or oobabooga, it's in fact not easy to make if you need it to search for you.
Hello, dear David I ask u so many time to considere doing videos aboout the outcome of networking with quantum computing because the trasnport protocols will all be affected pls do me the favours thx any way
Did you watch this video? ua-cam.com/video/MPzkkIknNOo/v-deo.htmlfeature=shared
@@davidbombal Thx a million david I love it
Once LM studio is installed is it possible to configure it in a client sever situation where other computers on the same LAN network can access the LLM of thier choice and use it via a web browser?
Hey bro can you share some information about vpn hosting in one of your videos
Great suggestion!
Will you do a guide how to run those LLMs without someones frontend?
Either I'm blind or I don't see the pdf reference link?
Fixed! Thank!
Great Video Sir, First to Comment
Thank you! Much Appreciated!
Thanks for this video, I followed it and installed LM Studio and Deepseek & Dolphin (same as your selections). I tried them both with a simple question: "How old is Gwyneth Paltrow" and got two different answers so I tried a Google search and got a third answer (which, by consensus of multiple links, appears to be correct). So how are we to trust anything from these AI searches?
You are getting different results based on the knowledge cutoff for each LLM. In the case of Dolphin shown in this video, the cutoff is 2023. For Deepseek shown in this video it is 2024. They are not accessing live data so you will get different results.
@@erichhaemmerle2654 Ms Paltrow was born fifty-something years ago, you'd think that was before any cutoff?
@@davidchamberlain6466 That is a very good point. 😂
Hi, is there something like that but for which I could inject my data - files on which it will work? I think I can't demand more than to put text but if possible, would be awesome if such a "beast" could operate on my private data such as photos, Excel files, gpx files etc.
take love from bangaladesh
Thank you! I appreciate it!
spent a lot of time with Dolphin. Definitely has censorship issues.
Hi David, what about content describing installation and setup of jailbreak GPT models ?
Can you please make a video about building an agents?
Nice video
Thank you!
weird, seems my dolphin is censored and my deepseek isnt... I am running Ollama and i get the opposite responses from those two models. Asked Dolphin how to break into a car, absolutely refused, asked deepseek about tank man, more info than I needed, it even refers to him has heroic and couragous and talks negatively about the chinese government...
David bro, you have made several videos on installing nethunter rootless.i wonder if you make video on running tools( which tools not need root or can be run without root but have limits ) on rootless android.
abliterated models are better than "uncensored" ones.
abliterated ones are fine-tuned to ignore, or go harder on the refusals.
while uncensored ones tuned on data someone sees as "uncensored", and they still end up having corporate guardrails.
abliterated ones just say anything.
also they are easier to instruct, because they wont refuse the instructions.
What are some examples of abliterated models?
Please make a video comparing Ollama vs LM Studio. Thanks!
Ollama can't process model in parts, that is all, it lost. Also there's oobabooga or even Kobold app
Great video but Chatgpt also has egregious censorship examples.
Like I said in the video, all LLMs have biases
Is it possible that when they created deep sink they put sleeper agents in that could be triggered? I read a paper last year about this. I just can’t remember where I read it.
David, I've seen a few videos where authors either use dockers, plus openwebui in Windows or Ollama to ran these models. Are there any significant differences? (performance?) Also, are all deepseek models censored? How to tell before you download a model if it is censored?
The Dolphin models by Cognitive Computations are uncensored
I tried this with a different program (Ollama). I noticed that DeepSeek verbose for its answers were way to loud. One simple question will trigger 10 min plus of text and didn’t know how to stop it, so I decided to delete everything instead. I’m gonna try this new way, but in case im encountering the same issue, what would it be the alternatives to fix it? Thanks
I'm a bit confused about the GPU offloading part. I'm using a Lenovo Legion 5 Pro laptop with an 8C/16T CPU, 32GB RAM and a 4070 (8GB). What is the "correct value" for that setting? I can go up to 32 model layers (whatever that means). I can see VRAM being used, but my GPU Utilization stays at 0%. Is that normal? Also I get quite a few unfinished queries from a point on.
is there any way to run this through my phone remote connected to my pc???
Am really looking for an AI model that can help me with my studies am not that bright in the AI tech field but i wanna adapt and use this to up my carrer and studies would love a suggestion of which LLM that u advice me of using
What should be the specs of laptop to run this thin, ya haven't shared em
This is probably a stupid question but can you do this on a phone or does it have to be a desktop?
when i clicked to download a model, only Gemma came up and couldnt find deepseek, anyone any ideas?
Same here
Can we use it with Termux locally
HEY DAVID, DOES THIS WORK ON AN i3 LAPTOP WITH NO GRAPHIC CARD AND 4GB RAM?
What do you know about Venice?
Is better Lm studio or Ollama + something for the gui (open webui or Msty App)?
Ollama is last one today which can't use model in parts, so getting by it one file model can take on certain speed at least 10 hours or more for original size.
Is there a way to run deepseek-r1 locally and make it uncensored?
I wanna know what happens if you ask Dolphin3.0 that exact question 🤣
Get a good hardware. For Q6 you need 567GB RAM and at least 20 cores CPU, in GGUF the Ram=storage size+10%
@fontenbleau Yeah but how do you make deepseek fully uncensored?
@@redgamestream uncensored where? What do you need it to? I personally interested in coding but it's awful in it.
@fontenbleau Well even if you ask a local deepseek model some historical questions it will refuse to answer so I believe there is a way to fine tune it so its uncensored and can say and answer whatever I ask it. Also if its isnt 671B model then it will not code as well as o1
Hay alguna forma de conectarle nuestras propias bds ?