New HYBRID AI Model Just SHOCKED The Open-Source World - JAMBA 1.5

AI21 Labs has released two new open-source AI models, Jamba 1.5 Mini and Jamba 1.5 Large, featuring a unique hybrid architecture called SSM-Transformer that combines the strengths of Transformers and Structured State Space Models. These Jamba models excel in handling long context windows, offering faster processing and lower resource usage, making them ideal for complex, data-heavy AI tasks. Jamba 1.5 models outperform other popular models like Llama 3.1 and Mistral in benchmarks, providing developers with powerful, efficient AI tools for diverse applications.

#ai #opensource

[Lemure_Noah]

05:02 - Not correct! The HF Jamba page states that *_"a minimum of 2x 80GB GPUs is required..."_* for 200K context and *_"a 80GB GPU"_* for 140k context. sorry, but It indeed requires a MASSIVE infrastruture!

While Llama 3.1 8b and other modern LLM models can run on GPU as low as a 8GB RTX-class GPUs, when using quantizations like GGUF, Jamba seems to be more of a cloud datacenter solution

[Echosintheattic]

I don't think anyone is **SHOCKED** by anything these days..

[antoniobortoni]

Wow, small fast powerful... an update of Android and maybe we get a really good IA in any smartphone, a good doctor, lawer, etc... helper ever. Good software free can change the world for the better of all like nothing.

[sirishkumar-mz]

While hybrid AI models are revolutionary, you may find that other technologies offer more capabilities or insights that better fit the needs of your project.

[LJSheffRBLX]

AI Revolution, cool video keep it up dude

[zeeker__]

Imagine if you were able to use RAG with JAMBA 1.5

[serikazero]

@05:05
Sorry but what?
Dude, I can run Lamma 3.1, 70b on a cpu from 4 years ago, on a laptop with just 64 gb of ram. Yes its slow but it runs.
And Llamma 3.1, 8b runs fast.
And they both do a 128k context window. So what are you smoking?
Other models can run "on a single gpu" or on just cpu if you have enough ram. So while maybe its faster or better, or deals better with the context window that's fine.
Just having a large context window? other models do that as well now.
And "a single gpu" means nothing because a "single gpu" can have different amount of VRAM. There's 1 thing to run AI models on a Nvidia Tesla 100, desktop GPU, and another to run it on an RTX 2070 on a laptop, which is what I used to do until my laptop GPU burned.

Sorry but at very least that segment is as wrong as you can be

[LJSheffRBLX]

AI Revolution, nice content you deserve more views

[KC_]

I have tried it, but the output is always very short, like a few sentences short. which is not ideal for my purpose.

[chanpasadopolska]

There is need for alternative for Poe that will incorporate those smaller but important models.

[WCKEDGOOD]

There will be an entire echo system of AI models out there one day. We are truly watching the future unfold before us.

[zabagar]

Ooohh, SHOCKED in all caps you say?! Wow this video must be awesome. I’ll never know.

[MacDonaldWilliamKabonja]

Love 💕 you no the meaning of jamba in Swahili language fart=jamba ahahaha 😅😅😅