How to Run Google's New FREE AI Model on Your PC - But is it Any Good?
Вставка
- Опубліковано 22 лют 2024
- Google has released Gemma, a family of lightweight open LLM models, built from the same research and technology used to create its Gemini models. You can run them in the cloud, but you can also run them on your PC using LMStudio or llama.cpp. In this video I test them out for speed, accuracy, and capabilities.
---
LMStudio video: • Run Your Own ChatGPT-l...
llama.cpp video: • Run a ChatGPT-like AI ...
Twitter: / garyexplains
#garyexplains - Наука та технологія
Been trying LLMs on my Raspberry Pi5 8GB, not fast but they work. Have not tried these LLMs. Will add them to my collection. Ollama makes it easy to try them.
At some point, we will be able to train this models on our own data I hope
Could it really get significantly better if it has so few parameters compared to Chatgpt?
The "point" of training it on Your own data is ((I suspect)) that You want it to be able to answer questions that is "specifically about Your own data" !!! Which a "general LLM" will not be able to do.
Best regards
Could it really get significantly better owver time if it has so few parameters compared to Chatgpt?
In you previous AI video - I was surprised by The Princess Bride link but the model went on to explain its reasoning which was sound so there you go.
As for these 'on device' models, they are going to be limited because they are not referring to the cloud for information and context. They could still be useful though in specialist applications.
Even the cloud LLMs are offline in that they don't refer to the internet for more information. The exception is Bing which Microsoft built to act like a half way between an LLM and a search engine. Try asking ChatGPT about something that happened last week and it can't tell you.
Maybe you covered this is another video, but are there security risks with running models shared on the internet?
None that I am aware of or can think of. The LLM model doesn't have access to your system. The underlying runtime code is based on llama.cpp which is open source.
Did try it, Gemma is not good when compare to llama 2, don't understrand simple phase, change subject/question and context. Side by side, Llama2 is clear better, 7B vs 7B.
is it censured? why is it so slow?
❤
Test algebra, calculus, and code generation on these different models.
Little point. The big LLMs struggle with maths. These little ones have no chance.
I've seen the model in action on youtube and it looks... awful
And it insists Empire State Building is in Tokyo and not NYC. 😂
😂
You know why its free?
You should find out.
Eh?
To be kind , they are still pretty crap aren't they 😄.
Hopefully they will get better.
You are too hard on Gemma with the spelling. There are a lot of people I know who do the same spelling mistake. I swear.
😂
They giving it away because its basically unusable