The EASIEST way to run MULTIMODAL AI Locally! (Ollama ❤️ LlaVA)
ฝัง
- เผยแพร่เมื่อ 11 ธ.ค. 2024
- With the power of Llava Models and Thanks to Ollama's support, you can run GPT-4 Vision like (not the exact match) Mutlimodal models locally on your computers (does not need CPU).
🔗 Links 🔗
My Ollama Intro tutorial - • Ollama on CPU and Priv...
Ollama Llava library - ollama.ai/libr...
Ollama Mulitmodal release - github.com/jmo...
LLaVA llava-vl.githu...
My previous Ollama Tutorial (Web UI)
• Ollama Web UI (ChatGPT...
❤️ If you want to support the channel ❤️
Support here:
Patreon - / 1littlecoder
Ko-Fi - ko-fi.com/1lit...
🧭 Follow me on 🧭
Twitter - / 1littlecoder
Linkedin - / amrrs
![LAVROV's interview with Tucker CARLSON 😁 [Parody]](/img/n.gif)
This is great, got it running on my Mac M2 pro 32GB and it is fast! Thanks for the insight how to get this running
Thanks for the donation 🙏🏽
Thanks for this Dr. 1LC ❤
Always welcome :)
😂i bet a whole dollar there’s gonna be a fine tuned spicy model named butLava just in time for christmas😂❤~ thank you for another awesome tutorial
Happy Prompting ❤❤
❤️
Thanks for the tutorial, don't have the time to try it tonight but maybe tomorrow. Never eaten llama meat, but I do like baklava.
I love baklava as well! Especially the OG one with lot of Pistachios!
@@1littlecoder Now we need to try llama meat.
When I run "ollama run Llava" it responds in Chinese. How to make it english? *Edit; updating ollama fixed it.
To update ollama, just run the ollama install script again
That means you have not updated your ollama, did you?
Does this work with ollama webui?
I'm yet to try that. Ollama webui mentions multimodal support
hi, Dr. 1LC, I think your work will help me to do some classification task experiment. So I don't want use it like in a chatbot way. How can I call some functions to get result? Hope you can see this comment
Do you mean you want to make API calls?
@@1littlecoder yes, I know how to do with chatGPT API, but I don't know how to do the same in Ollama
I think you meant "does not need GPU"... if not, I want to know how.
It runs on CPU memory but if you have reserved GPU then you can leverage it for additional processing
@@1littlecoder I meant in your written intro you said does not need CPU...
GPT 4.5pro released
what model is this. precise name - what is it called in the API?
ChatGPT
This model is called "ChatGPT with browsing", and in the context of the OpenAI API, it is referred to as "gpt-4.5-turbo". This version includes enhanced capabilities like browsing the internet, which is not available in the standard GPT-4 model. The browsing feature allows the model to pull in information from the web in real-time to provide more current and detailed responses.
User
Can you continue this text?
GPT models: these are various versions of the Generative Pre-trained
Transformer models. They include:
Gpt-3.5-turbo: an optimized version of the GPT 3.5 model
ChatGPT
Certainly! Continuing from where you left off:
GPT models: these are various versions of the Generative Pre-trained Transformer models. They include:
GPT-3.5-turbo: An optimized version of the GPT-3.5 model, designed for faster response times and more efficient processing. It maintains the advanced language understanding and generation capabilities of GPT-3.5 but with enhanced performance.
GPT-4: This model represents a significant leap in the scale and capabilities over its predecessors. It features a larger number of parameters, which allows for more nuanced and complex language processing. GPT-4 is known for its improved understanding of context and ability to handle more intricate and detailed tasks.
GPT-4.5-turbo (ChatGPT with browsing): A more recent iteration, this model combines the advanced capabilities of GPT-4 with the added functionality of real-time web browsing. This enables the model to access and incorporate up-to-date information from the internet, enhancing its responses with current data and expanded knowledge.
GPT-NEO and GPT-J: These are models developed by the open-source community, inspired by the GPT architecture. GPT-NEO and GPT-J aim to provide similar capabilities to OpenAI's versions, focusing on accessibility and open-source collaboration. They have been widely adopted for various applications where open-source models are preferred.
Each version of the GPT series builds upon the learnings and advancements of its predecessors, aiming to provide more sophisticated, accurate, and contextually relevant language generation and understanding. These models find applications in a wide range of fields, including customer service, content creation, programming assistance, and more, showcasing the versatility and power of AI in natural language processing.
No windows version yet isn't it?
I think only through WSL 👀
The pixel art in your video thumbnail would be a better channel avatar icon IMO.
Is it? Why do you feel so ?
@@1littlecoder It has more "appeal" (like a Disney or Cartoon Network character) and is drawn more professionally. The one you use looks drawn in paint and unfinished because the neck is blue like the background.
Ok now run it in windows
Google WSL installation
❤
btw i got m1 mac and base version 8gb ram and it lags soo much if apps are on or not :( atleast who would be trying i must recommend them 16gb ram else it would be janky as hell!
🚀
The apple tax on RAM (and ssds) is absolutely insane. Those brass turds sure know how to extract the most money feasible from the populous. It's a crying shame you can't add ram as a user serviceable upgrade. I'm rocking 3090 windows/wsl with 256gb of system ram but somewhat covet a studio because it's the easiest way to run some of the larger models. Been considering sli'ing a second 3090 but IDK how hard that is to use in practice.
@@zyxwvutsrqponmlkh Does your 3090 GPU support NVLINK? There is little information on this subject, I assume that you need to have two identical 3090 models that support NVLink, because different models have different bus locations
@@Damon_Sieputovsky Yes, shame 4090 dropped this, and it supports high density because I got it natively water cooled (msi kingpin). Can't seem to find many reasonably priced cases that support three radiators well.
I found moondream to be better than the small version of llava.
that's great. what was the use-case
whats the point?. poor AI models in your computer for what????
Mistral is the way to go.. cutting edge