The CPU running GGML version of 13B already made by Bloke, i've tested it. 13B reserved 24Gb RAM, writing quite fast near 3 tokens/sec on my Intel Xeon 14 cores. They've made it more sturdier, amazing how 13B is equal to previous 30B today by confidence and knowledge. The only problem i have is 2048 token limit on the oobabooga tool, this needs to be resolved, it spending all tokens very fast.
This is 13B assist of new ideas based on it's intuition: Sure! Here's some additional information on each of these ideas:
1.Virtual reality meditation: There are already several virtual reality meditation apps available, such as Calm and Headspace. These apps provide immersive environments that can help individuals relax and focus their minds. By analyzing user data, AI could be used to personalize the experience even further based on an individual's specific needs and preferences. 2.Personalized nutrition based on gut bacteria: Research has shown that the bacteria in our guts play a significant role in our overall health and wellbeing. By analyzing an individual's gut bacteria, personalized dietary recommendations could be made to improve their health and wellness. This could involve identifying specific nutrients or food groups that are lacking in the individual's diet. 3.AI-powered mental health chatbots: Mental health is an area where there is a significant need for more resources and support. AI-powered chatbots could provide anonymous and accessible support for individuals struggling with mental health issues. These chatbots could be trained to recognize specific symptoms and provide personalized recommendations based on the individual's needs. 4.Sustainable fashion made from recycled plastic: With the growing awareness of environmental issues, sustainable fashion has become increasingly popular. Using recycled plastic to create clothing and other textiles could be a game-changer for reducing waste and promoting eco-friendliness. This could involve partnering with companies that specialize in upcycling plastic materials into usable fabrics. 5.Personalized learning plans based on AI-analyzed student data: By analyzing large amounts of data, AI can help identify individual strengths and weaknesses in students. This information could be used to create personalized learning plans that are tailored to each student's needs. This could involve identifying specific areas where the student may need extra support or providing more challenging material for students who are advanced. 6.AI-powered legal assistants for pro bono cases: By analyzing legal documents and providing insights based on past case law, AI-powered legal assistants could help provide more efficient and effective support for pro bono cases. This could involve identifying relevant precedents or providing recommendations for specific legal strategies. 7.Personalized skincare recommendations based on skin microbiome analysis: The skin microbiome is a complex ecosystem that plays a significant role in our skin health. By analyzing the specific bacteria and other organisms present on an individual's skin, personalized skincare recommendations could be made to improve their skin health. This could involve identifying specific products or ingredients that are most effective for the individual's specific skin type and needs. 8.Virtual reality-based exposure therapy for anxiety disorders: Exposure therapy is a common treatment for anxiety disorders that involves gradually exposing individuals to the things they fear or avoid. By using virtual reality technology, this type of therapy could become more accessible and effective. This could involve creating immersive environments that simulate specific situations that trigger anxiety in the individual.
9.AI-powered financial advisors for low-income households: Financial planning can be complex and overwhelming, especially for those with limited resources. AI-powered financial advisors could provide personalized recommendations based on an individual's income, expenses, and financial goals. This could involve identifying specific budgeting strategies or providing recommendations for investment opportunities that are tailored to the individual's needs. 10.Personalized meal plans based on nutrient deficiencies: By analyzing an individual's dietary habits and identifying any nutrient deficiencies, AI-powered personalized meal planning could help improve overall health and wellness. This could involve identifying specific food groups or nutrients that are lacking in
It's totally censored and won't tackle any issue with the slightest hint of controversy. Meta and OpenAI have released dumb-bots for the smooth-brained masses
Microsoft used to be very hostile towards open source, especially Linux, but under new management, all that change where they take a more pragmatic view on it all.
They realise that open source can be a great help in pushing tech and ideas forward, so it does make some sense why the likes of Microsoft and Meta are involved, they realise that open source can do a lot of work on these A.I. models, something they and anyone can use in their own software, it's an easy way to push the advancement of A.I. software a lot faster than what corporations can do, and to see that, just look at the rapid pace that happened with LLAMA 1, if we get the same with LLAMA 2 and going on the quality of it already, it could be a solid option over the next 6 months that might even be a rival to the likes of ChatGPT 4.
Microsoft, Meta are not the only companies that see the benefits of the open source community, a lot of other companies see the benefits, hence why they contribute where it's in their interest to do so, and it's no surprise really, the open source community is like a massive army of developers, with resources coming from the public, companies and governments around the world, hence why many work with it over fighting it like they did in the past.
As for the safety, that's like Meta covering it's own back from the law, they likely realise that much of the open source community is going to strip out a lot of that restriction, just like they've done with others, but the law is starting to get tough in areas of safety and Meta likely doesn't want that responsibility, whereas with open source, there's no real ownership as such, so the risk is far less for them.
so all of these models have such a small token limit I mean cool but like really you cant get anything done with these token limits. Is this controlled by the model or is there a way it can be manipulated as long as you have the process power to use a higher token limit?
Hello Matthew, could you kindly create a comprehensive video demonstrating the fine-tuning process of any model (7b, 13b, 70b) of Llama 2 using CSV data? The video should showcase how to train the model to make predictions on various topics, such as cricket or any other subject. Please ensure that the CSV data used contains a diverse set of numerical values and other relevant information. For example a CSV dataset containing detailed information on all cricket matches played, including ball-by-ball data, encompassing every run scored and wicket taken. By training the model on this comprehensive dataset, we can subsequently utilize it to predict valuable insights, such as estimating a team's potential total runs or determining the likelihood of a particular batsman getting dismissed against another specific bowler.
With LLaMA 2, they baked the safety measures so far into it, I think it might be useless to fine tune it. It might seem uncensored afterwards, but, I'm not so sure. It seems a bit deep fried to me. Will have to see though.
Future is looking great. However, I'm not sure I agree with "not entirely open source". It is entirely open source, lone developers as well as Google and Microsoft can obtain 100% of the source code. The licensing however brings the caveat.
00:00 🦙 LLaMA 2 represents a significant leap forward in open-source large language models, offering research and commercial viability, but there's a caveat for products with over 700 million users. 03:15 🤝 Microsoft partnered with Meta on LLaMA 2 to support open and frontier models, indicating a balance between open-source and protecting their investment in GPT-4. 04:08 ⚙️ LLaMA 2 is commercially viable, but it lacks strong coding ability compared to GPT-4, which excels in coding tasks. 05:31 🚧 Safety is a major focus for LLaMA 2, with nearly half of the white paper dedicated to safety measures, red teaming, and evaluations. 06:38 🔒 LLaMA 2 models (7B, 13B, and 70B parameters) perform similarly in terms of safety, but the 34B parameter model, significantly less safe, is delayed until safety improves.
You work for 40yrs to have $1m in your retirement, meanwhile some people are putting just $10k in a crypto market for just few months and now they are millionaires , all thanks to Mrs Jane, God bless you madam 😇
Sounds impressive I have watched several of your videos this being the second on LLaMA 2 and I have no idea even after watching the video what it does or what you can do with it, maybe your channel it just too advanced for me. I don't know?
@matthew_berman Is it possible to use embeddings in Llama2, similar to how they're used in ChatGPT? This way, one could utilize large amounts of text for interacting with their documents?
I will forever be indebted to you you've changed my whole life continue to preach about your name for the world to hear you've saved me from a huge financial debt with just little investment, thanks so much Mrs.Sarah Walsh.
"The OSI has criticized Meta’s licensing terms, stating that they “do not comply with the Open Source Definition.” The OSI has also called on Meta to release Llama 2 under a more permissive license that would allow for its unrestricted use."
Will never trust MICROSOFT. I think Microsoft is loosing their trust completely by spying on society. Your talk is trash! Google. Miscrosoft, Facebook, Apple – are the least trustable companies on the planet.
00:00 🦙 Llama 2 representa un gran avance en modelos de lenguaje de código abierto y se acerca mucho al rendimiento de GPT-4. 01:09 📈 Llama 2 se ha entrenado en un conjunto de datos un 40% más grande que Llama 1 y ha duplicado el tamaño del contexto para mejorar su escalabilidad de inferencia. 02:46 🌍 Meta considera las emisiones de carbono durante el entrenamiento de Llama 2, mostrando conciencia ambiental en el desarrollo de modelos de lenguaje a gran escala. 03:41 🔗 Microsoft se asoció con Meta en Llama 2, demostrando el equilibrio entre apoyar modelos de código abierto y proteger su inversión en modelos como GPT-3. 04:35 📝 Llama 2 es comercialmente viable, pero existe una excepción: se necesita permiso de Meta si el producto basado en Llama 2 tiene más de 700 millones de usuarios. 05:58 ⚠️ La seguridad ha sido una prioridad en Llama 2, con un enfoque en guardrails, pruebas y evaluaciones para reducir violaciones a las directrices del modelo. 06:25 🛑 El modelo de 34 mil millones de parámetros de Llama 2 fue retrasado debido a problemas de seguridad, pero su tamaño promete calidad y capacidad en GPUs de gama alta. 07:33 ✅ Llama 2 aborda el equilibrio entre seguridad y utilidad con un enfoque de doble recompensa, lo que podría mejorar significativamente la seguridad sin sacrificar la utilidad. 08:15 📥 Puedes descargar los modelos y el código de Llama 2 desde el repositorio de Hugging Face y hay versiones ya alojadas de los modelos base para comenzar a usarlo hoy.
Thanks for your sharing!
if their focus is on safety it'll end up being garbage. AI only works with the absolute truth. When's your petals tutorial?
Can't wait to see more on what this model will offer. Looking forward to your other videos on how Llama 2 will perform.
Helpful vs Harmful
==
Truth vs Woke
The CPU running GGML version of 13B already made by Bloke, i've tested it. 13B reserved 24Gb RAM, writing quite fast near 3 tokens/sec on my Intel Xeon 14 cores. They've made it more sturdier, amazing how 13B is equal to previous 30B today by confidence and knowledge. The only problem i have is 2048 token limit on the oobabooga tool, this needs to be resolved, it spending all tokens very fast.
This is 13B assist of new ideas based on it's intuition:
Sure! Here's some additional information on each of these ideas:
1.Virtual reality meditation: There are already several virtual reality meditation apps available, such as Calm and Headspace. These apps provide immersive environments that can help individuals relax and focus their minds. By analyzing user data, AI could be used to personalize the experience even further based on an individual's specific needs and preferences.
2.Personalized nutrition based on gut bacteria: Research has shown that the bacteria in our guts play a significant role in our overall health and wellbeing. By analyzing an individual's gut bacteria, personalized dietary recommendations could be made to improve their health and wellness. This could involve identifying specific nutrients or food groups that are lacking in the individual's diet.
3.AI-powered mental health chatbots: Mental health is an area where there is a significant need for more resources and support. AI-powered chatbots could provide anonymous and accessible support for individuals struggling with mental health issues. These chatbots could be trained to recognize specific symptoms and provide personalized recommendations based on the individual's needs.
4.Sustainable fashion made from recycled plastic: With the growing awareness of environmental issues, sustainable fashion has become increasingly popular. Using recycled plastic to create clothing and other textiles could be a game-changer for reducing waste and promoting eco-friendliness. This could involve partnering with companies that specialize in upcycling plastic materials into usable fabrics.
5.Personalized learning plans based on AI-analyzed student data: By analyzing large amounts of data, AI can help identify individual strengths and weaknesses in students. This information could be used to create personalized learning plans that are tailored to each student's needs. This could involve identifying specific areas where the student may need extra support or providing more challenging material for students who are advanced.
6.AI-powered legal assistants for pro bono cases: By analyzing legal documents and providing insights based on past case law, AI-powered legal assistants could help provide more efficient and effective support for pro bono cases. This could involve identifying relevant precedents or providing recommendations for specific legal strategies.
7.Personalized skincare recommendations based on skin microbiome analysis: The skin microbiome is a complex ecosystem that plays a significant role in our skin health. By analyzing the specific bacteria and other organisms present on an individual's skin, personalized skincare recommendations could be made to improve their skin health. This could involve identifying specific products or ingredients that are most effective for the individual's specific skin type and needs.
8.Virtual reality-based exposure therapy for anxiety disorders: Exposure therapy is a common treatment for anxiety disorders that involves gradually exposing individuals to the things they fear or avoid. By using virtual reality technology, this type of therapy could become more accessible and effective. This could involve creating immersive environments that simulate specific situations that trigger anxiety in the individual.
9.AI-powered financial advisors for low-income households: Financial planning can be complex and overwhelming, especially for those with limited resources. AI-powered financial advisors could provide personalized recommendations based on an individual's income, expenses, and financial goals. This could involve identifying specific budgeting strategies or providing recommendations for investment opportunities that are tailored to the individual's needs.
10.Personalized meal plans based on nutrient deficiencies: By analyzing an individual's dietary habits and identifying any nutrient deficiencies, AI-powered personalized meal planning could help improve overall health and wellness. This could involve identifying specific food groups or nutrients that are lacking in
if it's not under one of the traditional opensource licences like MIT, GPL it may be Source viewable but it's not really opensource.
It's totally censored and won't tackle any issue with the slightest hint of controversy. Meta and OpenAI have released dumb-bots for the smooth-brained masses
Still holding out for a GPT4 level open source model before i do a lot more local processing. Guess ill need to wait for 2 years.
The reason Microsoft and meta are teaming up on this is clear: censorship and controlling the message
was waiting for this
Thanks for the Video. And you can download for Mac as well ?
Microsoft used to be very hostile towards open source, especially Linux, but under new management, all that change where they take a more pragmatic view on it all.
They realise that open source can be a great help in pushing tech and ideas forward, so it does make some sense why the likes of Microsoft and Meta are involved, they realise that open source can do a lot of work on these A.I. models, something they and anyone can use in their own software, it's an easy way to push the advancement of A.I. software a lot faster than what corporations can do, and to see that, just look at the rapid pace that happened with LLAMA 1, if we get the same with LLAMA 2 and going on the quality of it already, it could be a solid option over the next 6 months that might even be a rival to the likes of ChatGPT 4.
Microsoft, Meta are not the only companies that see the benefits of the open source community, a lot of other companies see the benefits, hence why they contribute where it's in their interest to do so, and it's no surprise really, the open source community is like a massive army of developers, with resources coming from the public, companies and governments around the world, hence why many work with it over fighting it like they did in the past.
As for the safety, that's like Meta covering it's own back from the law, they likely realise that much of the open source community is going to strip out a lot of that restriction, just like they've done with others, but the law is starting to get tough in areas of safety and Meta likely doesn't want that responsibility, whereas with open source, there's no real ownership as such, so the risk is far less for them.
so all of these models have such a small token limit I mean cool but like really you cant get anything done with these token limits. Is this controlled by the model or is there a way it can be manipulated as long as you have the process power to use a higher token limit?
Obviously MSFT wants to protect their investment with openai.
Hello Matthew, could you kindly create a comprehensive video demonstrating the fine-tuning process of any model (7b, 13b, 70b) of Llama 2 using CSV data? The video should showcase how to train the model to make predictions on various topics, such as cricket or any other subject. Please ensure that the CSV data used contains a diverse set of numerical values and other relevant information.
For example a CSV dataset containing detailed information on all cricket matches played, including ball-by-ball data, encompassing every run scored and wicket taken. By training the model on this comprehensive dataset, we can subsequently utilize it to predict valuable insights, such as estimating a team's potential total runs or determining the likelihood of a particular batsman getting dismissed against another specific bowler.
😮
Embrace, extend, extinguish.
With LLaMA 2, they baked the safety measures so far into it, I think it might be useless to fine tune it. It might seem uncensored afterwards, but, I'm not so sure. It seems a bit deep fried to me. Will have to see though.
Just tried it. It's a crap
Future is looking great. However, I'm not sure I agree with "not entirely open source". It is entirely open source, lone developers as well as Google and Microsoft can obtain 100% of the source code. The licensing however brings the caveat.
00:00 🦙 LLaMA 2 represents a significant leap forward in open-source large language models, offering research and commercial viability, but there's a caveat for products with over 700 million users.
03:15 🤝 Microsoft partnered with Meta on LLaMA 2 to support open and frontier models, indicating a balance between open-source and protecting their investment in GPT-4.
04:08 ⚙️ LLaMA 2 is commercially viable, but it lacks strong coding ability compared to GPT-4, which excels in coding tasks.
05:31 🚧 Safety is a major focus for LLaMA 2, with nearly half of the white paper dedicated to safety measures, red teaming, and evaluations.
06:38 🔒 LLaMA 2 models (7B, 13B, and 70B parameters) perform similarly in terms of safety, but the 34B parameter model, significantly less safe, is delayed until safety improves.
Recap by HARPA AI 🧬
I wish you give your review on Claude 2!
Why 7,13 & 70B parameters? Why it can't be 6 or 8 etc?
bot
That's what every profiteer is chiming these days.I haven't seen anything, let alone free.
You work for 40yrs to have $1m in your retirement, meanwhile some people are putting just $10k in a crypto market for just few months and now they are millionaires , all thanks to Mrs Jane, God bless you madam 😇
I want uncensored powerful AI models
Turns out that OpenAI is the ChatGPT killer.
Sounds impressive I have watched several of your videos this being the second on LLaMA 2 and I have no idea even after watching the video what it does or what you can do with it, maybe your channel it just too advanced for me. I don't know?
What does “safe” mean? It has a leftist bias?
Is it THE ChatGPT killer? Nah, that will surely be the new AI that is breathlessly announced tomorrow, and the next day, and the next day.
it's helpful. Thanks 👍
👏🏼
@matthew_berman Is it possible to use embeddings in Llama2, similar to how they're used in ChatGPT? This way, one could utilize large amounts of text for interacting with their documents?
Great content my friend.
Bloggers, why always everything new for you is some thing's killer?
I will forever be indebted to you you've changed my whole life continue to preach about your name for the world to hear you've saved me from a huge financial debt with just little investment, thanks so much Mrs.Sarah Walsh.
In my experience llama is much worse than GPT3.5.
Yes
Gbt is dead
Saw the coffin and everything
So apparently its not really opensource:
"The OSI has criticized Meta’s licensing terms, stating that they “do not comply with the Open Source Definition.” The OSI has also called on Meta to release Llama 2 under a more permissive license that would allow for its unrestricted use."
Will never trust MICROSOFT. I think Microsoft is loosing their trust completely by spying on society. Your talk is trash! Google. Miscrosoft, Facebook, Apple – are the least trustable companies on the planet.
What would be the best LLM for me to pull in Crypto data and analyze patterns? I already have coded the data capture. Just need an LLM to feed it.
This is not good news
AI should be curated… raised.
Every AI needs parents. 😂
FaceZuck = 100% power grab
FaceZuck "safety" : shadow banning for trivia
Nothing is ever safety.
The issue is personal responsibility.
Freedom = Nature
🎯 Key Takeaways for quick navigation:
00:00 🦙 Llama 2 representa un gran avance en modelos de lenguaje de código abierto y se acerca mucho al rendimiento de GPT-4.
01:09 📈 Llama 2 se ha entrenado en un conjunto de datos un 40% más grande que Llama 1 y ha duplicado el tamaño del contexto para mejorar su escalabilidad de inferencia.
02:46 🌍 Meta considera las emisiones de carbono durante el entrenamiento de Llama 2, mostrando conciencia ambiental en el desarrollo de modelos de lenguaje a gran escala.
03:41 🔗 Microsoft se asoció con Meta en Llama 2, demostrando el equilibrio entre apoyar modelos de código abierto y proteger su inversión en modelos como GPT-3.
04:35 📝 Llama 2 es comercialmente viable, pero existe una excepción: se necesita permiso de Meta si el producto basado en Llama 2 tiene más de 700 millones de usuarios.
05:58 ⚠️ La seguridad ha sido una prioridad en Llama 2, con un enfoque en guardrails, pruebas y evaluaciones para reducir violaciones a las directrices del modelo.
06:25 🛑 El modelo de 34 mil millones de parámetros de Llama 2 fue retrasado debido a problemas de seguridad, pero su tamaño promete calidad y capacidad en GPUs de gama alta.
07:33 ✅ Llama 2 aborda el equilibrio entre seguridad y utilidad con un enfoque de doble recompensa, lo que podría mejorar significativamente la seguridad sin sacrificar la utilidad.
08:15 📥 Puedes descargar los modelos y el código de Llama 2 desde el repositorio de Hugging Face y hay versiones ya alojadas de los modelos base para comenzar a usarlo hoy.
Made with HARPA AI
Utopia AI is the most functional neural network I've used.
Comments are closed.