I'm actually changing my mind here. We need a better definition for model, not of open source. The trained model and the code that defines the structure of the model and the training process cannot be both called "the model". Open sourcing the structural code doesn't mean the weights are reproducible. So maybe the issue it's in the definition of what open source is referring to.
Assuming that this presentation was intended for informing the larger public, One would have expected that the presenter would have, on the onset, REMINDER the audience what LLAMA stand for !
designed for on-device inference - small enough to run directly on smartphones and laptops so data never has to be sent to the cloud. This can both speed up processing and improve privacy -that is the basiC explanation that I found and can we also talk about IBM's quantum processors
Llama is not open source, the code might be, but the model is under a commercial license. Yes its not too restrictive for end users, but it is for businesses "You can use this model how you like, unless you make too much money, then you need to negotiate an alternative license with us". I do think self host able llm's are something that should be celebrated, and the permissive license is a step forward. But I hate that its being whitewashed as "open source" when its very very much not that, no matter how much meta claim's the model is open source for free clout.
The reason this matters, is because the code is nothing special or some super secret appropriate thing. What makes the open source community unable to create LLM's that work as well as chatgpt and llama. Has nothing to do with the code, and everything to do with the availability of training data and computational resources to train an LLM on that data. All the big tech companies are spending billions on custom chips from nvidia that specialist in training and running ai compute workloads, that the opensource community just doesn't have access too. API's for gathering data to train an ai on are becoming closed down and or super pricey. Publicly available data is being flooded by ai content, and image wise there's the whole artist rebellion where they're poisoning they're art, which will only effect smaller open source devs, big tech im sure will have the resources to avoid using poisoned art in their training..... In summery, The model is whats special, not the code, they open sourced the code, but the model is under a proprietary commercial license thats only permissive for personal use, and highly restrictive for commercial use.
There are just so many variations in hugging face, like for any model groups. Would be interesting to see model cluster information to help users choose the best for them.
I really wish open source LLM's actually existed, meta's claim this is open source isnt really true, see comment above, or just google "is llama really opensource"
There are only few LLMs where it is transparent in terms of openness. Most of models are open in terms of its parameter(weight and biases) but closed source code,closed training data, proprietary algorithms(neural network design or algorithm). Looking at Llama, it falls into first category, so it’s ok to use it for fine tuning towards one specific tasks but one will need additional commercial license agreement needed if anyway it is used in any of services rolled out > threshold number of users. So please beware of license agreements first before jump into making a product with these so called open source models
I appreciated the presentation, but I have found it a bit weird that Meta / Facebook which is the owner of Llama has not been mentioned even once despite the presentation being about the history of Llama.
Does this video feel condescending to anyone else? I suppose someone had to make a super basic, overly simple introduction to llama. Might as well be IBM. I guess.
Five minutes and thirty six second in, and I still have no idea what Llama is. All I've learned about it is different versions and stats about it. Another 'tech bro' explaining "use the AI, just trust us, bro".
In case someone is wondering Llama >>> Large Language Model Meta AI
Thank you. I was thinking about the Llama animal and wondering why that name... Dummy me :)
Llama 3.1 is Meta's latest flagship language model, boasting an impressive 405 billion parameters. 5:05
Great work bro.
LLama is not open source at all. It is freeware. Their license is not an open source license. The training data is not available at all.
Interesting comment, would be interesting to hear more about this nuance
It's not a nuance. Can you reproduce it, rebuild it from scratch exactly as the same model you get precompiled? You can't.
I'm actually changing my mind here. We need a better definition for model, not of open source. The trained model and the code that defines the structure of the model and the training process cannot be both called "the model". Open sourcing the structural code doesn't mean the weights are reproducible. So maybe the issue it's in the definition of what open source is referring to.
100%
I believe in common vernacular, Llama is considered "open weights" but not "open source"
Assuming that this presentation was intended for informing the larger public, One would have expected that the presenter would have, on the onset, REMINDER the audience what LLAMA stand for !
It stands for Llama, like the animal.
designed for on-device inference - small enough to run directly on smartphones and laptops so data never has to be sent to the cloud. This can both speed up processing and improve privacy -that is the basiC explanation that I found and can we also talk about IBM's quantum processors
One of the easiest explanation of llama ❤
hey, great video, btw i like the way this video is made. the parts writing .... what did u use to make it thx
Morning Coffee Tech Content! Thanks IBM !
Incredible, I am currently having my coffee watching this video haha
Never heard of Llama before. You've given me something to do this weekend. Thanks Brianne!
Where have you been for past couple of years?
Llama is not open source, the code might be, but the model is under a commercial license. Yes its not too restrictive for end users, but it is for businesses "You can use this model how you like, unless you make too much money, then you need to negotiate an alternative license with us". I do think self host able llm's are something that should be celebrated, and the permissive license is a step forward. But I hate that its being whitewashed as "open source" when its very very much not that, no matter how much meta claim's the model is open source for free clout.
The reason this matters, is because the code is nothing special or some super secret appropriate thing. What makes the open source community unable to create LLM's that work as well as chatgpt and llama. Has nothing to do with the code, and everything to do with the availability of training data and computational resources to train an LLM on that data. All the big tech companies are spending billions on custom chips from nvidia that specialist in training and running ai compute workloads, that the opensource community just doesn't have access too. API's for gathering data to train an ai on are becoming closed down and or super pricey. Publicly available data is being flooded by ai content, and image wise there's the whole artist rebellion where they're poisoning they're art, which will only effect smaller open source devs, big tech im sure will have the resources to avoid using poisoned art in their training..... In summery, The model is whats special, not the code, they open sourced the code, but the model is under a proprietary commercial license thats only permissive for personal use, and highly restrictive for commercial use.
Thanks for the explanation
Thanks for sharing the Key Insights of LLMA and Covering Real Time Use Cases
That was a very informative video with a great explanation. Thank you for sharing such valuable information.
There are just so many variations in hugging face, like for any model groups. Would be interesting to see model cluster information to help users choose the best for them.
Thank U educating us on this great topic! It's great to know that open source LLM's exist! As usual, I appreciate all that U bring us IBM! Cheers!!
I really wish open source LLM's actually existed, meta's claim this is open source isnt really true, see comment above, or just google "is llama really opensource"
There are only few LLMs where it is transparent in terms of openness. Most of models are open in terms of its parameter(weight and biases) but closed source code,closed training data, proprietary algorithms(neural network design or algorithm). Looking at Llama, it falls into first category, so it’s ok to use it for fine tuning towards one specific tasks but one will need additional commercial license agreement needed if anyway it is used in any of services rolled out > threshold number of users. So please beware of license agreements first before jump into making a product with these so called open source models
How are these domain specific models built?
Very good question - would like to see presentation/demonstration on this topic
I appreciated the presentation, but I have found it a bit weird that Meta / Facebook which is the owner of Llama has not been mentioned even once despite the presentation being about the history of Llama.
Zuck is a villain, but this is one good thing he did for the world !
Why, what did he do.
@@MohammadAli-io9 because he has given Facebook 😂
IBM, I love you, your company is great! I am collecting your products, an unusual hobby for a 16 year old schoolboy 😅
I'm trying to register for the event in the description. But I can't successfully register. Can someone in IBM team help me out??
Hi, we were able to fix the link! Let us know if you have any more issues, and we hope you enjoy the event. Thanks for registering!
👏 Thank you.
How does a bigger context window induce security risks?
impressed with the direction Aliagents is taking in the AI space, big things coming from them
Did she forgot about Llama 3.2 1B, 3B, 11B, and 70B models.
Nice! Thank You
Aliagents is creating a powerful AI ecosystem, I’m excited to see how this develops
Test-time compute. I want my own local 8B parameter Llama model that can produce and process tokens locally before providing an output.
the tech Aliagents is developing could be a real game changer for the AI industry
COMO TE LLAMA? 🤔
Nice❤
Autonomous AI Agents
the way Aliagents integrates AI with tokenization is changing the game, excited for the future
Does this video feel condescending to anyone else?
I suppose someone had to make a super basic, overly simple introduction to llama.
Might as well be IBM. I guess.
What about llama 3.2 ?!!!
For now, it's still a secret!
Please let go 🙏😭, I'm not responsible for everything that has happened
Five minutes and thirty six second in, and I still have no idea what Llama is. All I've learned about it is different versions and stats about it. Another 'tech bro' explaining "use the AI, just trust us, bro".
You didn’t need all this drawing. Thanks anyway.
I wonder how long it took to learn to write right to left😅
65😅
1st🎉
Does she even know what she is talking about? Doesnt really seem so.