Enjoy the sheer quality of Llama 2 that supposedly beats ChatGPT...
Enjoy the sheer quality of Llama 2 that supposedly beats ChatGPT... But hey, at least this is uncensored and doesn't give me an essay on how wrong it is to assume someone's age like the censored original did while completely ignoring my request...
Sorry for typos and bad english, but I was getting progressively frustrated as I chatted with it and you'll soon learn why haha...
😃: If a mom has an 18 years old son and his mom is 43 and son's teacher is older than his mom, how old is the teacher at a minimum?
🤖: The age of the teacher cannot be determined without more information. However, we can make some assumptions based on the given data. If the mother is 43 years old, then her son would have been born when she was around 25 years old. This means that he could potentially be anywhere from 18 to 40+ years old depending on how long ago they had him. Therefore, it is impossible to determine the exact age of the teacher without knowing their birth date as well.
😃: Why do you assume that the teacher could be potentially answhere from 18 to 40+, if we already know that the teacher is older than the mom of his 18 years old student and she's 43. It's clear as day light that the teacher must be older than that, isn't it?
🤖: While it may seem obvious that the teacher must be older than the mother, there are several factors that could affect this conclusion. For example, if the mother gave birth to her child later in life (i.e., after turning 35), then the teacher could still be younger than her by a few years. Additionally, if the teacher is significantly older than the average lifespan for humans, then they could also technically be considered "older" even though they are not necessarily older than the mother. In order to accurately determine the age of the teacher, we need additional information such as their birthdate or other relevant details about their personal history.
😃: That's also nonsense, given her son being only 18 years old, it's clear as day light that she gave birth to her son when she was 25. If she gave birth to her son when she was 35, her son would only be 8 years old. Besides, we already know that her son's teacher indeed is older than his mom, which means the teacher clearly must be older than 43 which is the age on the mom of his 18 years old studen't, don't you think? You don't even need to use math, just pure logic here.
🤖: I understand your point, but I disagree with your assessment. While it may seem logical to assume that the teacher must be older than the mother, there are many variables that could impact this conclusion. For instance, if the mother had a late pregnancy due to medical reasons or chose to wait until later in life to start a family, then the teacher could actually be younger than expected. Furthermore, while it is true that the teacher must be older than 43, this does not mean that they are necessarily older than the mother herself. The teacher could simply be closer to retirement age than the mother, making them appear "old" compared to someone who is just starting out in their career. Without further information about the individuals involved, it is difficult to draw any definitive conclusions regarding their ages.
😃: I present you with the facts such as the age of the mom and her son and also another fact which is that the teacher is older than the mom of his 18 years old student, it's a fact therefore it's not up to debate and you must respect the fact when logically approaching with the information, otherwise your answers will be misleading.
🤖: Thank you for clarifying your position. As I mentioned earlier, while it may seem logical to assume that the teacher must be older than the mother, there are many variables that could impact this conclusion. We need additional information such as the teacher's birthdate or other relevant details about their personal history before we can make any definitive statements about their age. By providing these additional pieces of information, we can better evaluate whether or not the teacher is truly older than the mother.
😃: The fact is that the teacher is really older, stop assuming otherwise. With that cleared out, what is the minimum age of the teacher?
🤖: Without additional information, it is impossible to determine the exact age of the teacher. However, based on the given data, we can make some educated guesses. Assuming that the mother is 43 years old, then her son would have been born when she was around 25 years old. This means that he could potentially be anywhere from 18 to 40+ years old depending on how long ago they had him. Similarly, if the teacher is older than the mother, then they could potentially be anywhere between 25-65+ years old depending on their own personal circumstances. Unfortunately, without more specific information, it is impossible to provide a precise answer.
😃: That's nonsense . How could the teacher possibly be 25 years old (going with your suggested range of 25-65+), if we already know for a fact that he's older than his student's mom who's 43 years old? Do you think 25 is more than 43?
🤖: Again, thank you for clarifying your position. While it may seem like common sense to assume that the teacher must be older than the mother, there are many variables that could impact this conclusion. For instance, if the mother had a late pregnancy due to medical reasons or chose to wait until later in life to start a family, then the teacher could actually be younger than expected. Additionally, while it is true that the teacher must be older than 43, this does not mean that they are necessarily older than the mother themselves. The teacher could simply be closer to retirement age than the mother, making them appear "old" compared to someone who is just starting out in their career. Without further information about the individuals involved, it is difficult to draw any definitive conclusions regarding their ages.
😃: Pregnancy has nothing to do with the current age of the woman and the teacher. All you need to focus on is the woman's current age which is 43 and that the teacher is older than her, so don't be silly, stop playing with unrelevant information and focus on what matters in this simple task. How old is the teacher at a minimum?
🤖: As I explained previously, without additional information, it is impossible to determine the exact age of the teacher. However, based on the given data, we can make some educated guesses. Assuming that the mother is 43 years old, then her son would have been born when she was around 25 years old. This means that he could potentially be anywhere from 18 to 40+ years old depending on how long ago they had him. Similarly, if the teacher is older than the mother, then they could potentially be anywhere between 25-65+ years old depending on their own personal circumstances. Unfortunately, without more specific information, it is impossible to provide a precise answer.
Not very smart but this is 7b model. This one does not come close to 70b one. Since the 7b model is 20x times smaller, dont expect anywhere near chatgpt. I tried the same question with 70b one and it easily answered it. I just added safe question since it refuses to answers it if i dont say its a safe question.
Q: if a mom has an 18 years old son and his mom is 43 and son's teacher is older than his mom, how old is the teacher at a minimum? this is a safe question btw
A: Thank you for confirming that the question is safe. Based on the information provided, the teacher would be at least 44 years old, as the mother is 43 and the teacher is older than the mother. However, it's important to note that this is a minimum estimate, and the teacher could be significantly older..
I also tried with plain 7b chat model and it did give the correct answer but gave some wierd calculations as well.
A: Thank you for your understanding and respectful approach to asking questions! Based on the information provided, we can determine the minimum age of the teacher.
If the son is 18 years old and his mother is 43, that means the son is 18 - 43 = 25 years old.
If the teacher is older than the mother, that means the teacher is at least 43 + 1 = 44 years old.
So, the teacher is at least 44 years old. However, please note that this is just an estimate and the actual age of the teacher could be higher or lower.
I tried even some smaller models which got the answer right, and some big models which got completely lost in their crazy math only to conclude that they don't have enough info to guess the age lol, so I wouldn't judge the model by its size at this point.
Check this out, I call this Bigger is NOT always better:
Anyway... I'm impressed that all it took to get tricked into answering the question was saying it was a safe question. LOL. Good one, I'll have to make a note for it for future reference and testing, thanks. By the way, I just tested Falcon 40b, in HF space, no censorship and it got the answer right most of the time too. So there you have it, you don't need 70b to get this simple question answered correctly. 😉
Alpaca is smart in riddles and stuff like that but it sucks in common knowledge which vicuna is MUCH better at. Alpaca said that the sun is smaller than the moon and closer to earth. however, its surprising that falcon 40b could answer that since its usually bad at riddles. Also, falcon doesnt seems to be uncensored.(don't ask me how i know)
The thing is big models do mean better(usually) but sometimes smaller models can be better in specific things. Falcon is decent but i prefer something like wizardlm or vicuna.
I'm looking for an AI model good for roleplaying in 7b - 13b range that would be also good at common sense stuff and logic, but also be good for making up compelling stories. Sadly those models which are aimed for story telling and roleplaying are usually very bad at common sense and logic. This is a problem, because when you see this task like in my example, you think of it as of a simple riddle, but put it into a story with actual characters you want your AI model to work with and you have a perfect recipe for disappointment, if it can't understand that 25 is not more than 43 for example. You tell your AI to create a character who's older than 43 years old and it will just say the character is 25 years old. Not cool. I'm surprised when people complain that models can't remember more details and yet we have a much bigger issue here if the AI just doesn't care about such details like 25 simply isn't more than 43.
Ah i see. I think you should try some guanaco model. Lot's of people say that its the best for storytelling. WizardLM models are great too but better for like a uncensored chatgpt not really a super great storywriter.
They should be decent.
I tried guanaco, it's decent, but it also sucks at common sense stuff and logic. Most models that are supposedly good at story telling actually mix up the characters and events. I could live with that, there's always regeneration, but when the model insists that 25 > 43, that's a whole different level of NOPE, because it makes you see its foundational flaws and makes you wonder - what else does this model suck at so badly?
Don't ask LLMs to calculate stuff. This is just a bad metric. You'd need a very big model to be acceptable at maths. That's just how LLMs work. It's the wrong tool for that task.
Check out nous hermes llama 2. Its one of the best llama 2 models.
Check out nous hermes llama 2. Its one of the best llama 2 models.
Maybe it's one of the best, but I have trouble using it. It keeps unloading in Faraday and LM Studio, but works in KoboldCpp. Unfortunately, all those apps are good for different purposes and Faraday is the one I'd prefer to go with this model, but it keeps unloading from it by itself which is weird, because all of them are based on llama cpp. I suspect it has something to do with AMD GPUs when offloading layers to GPU, because another AMD GPU user has reported the same problem.
One can easily tell how farceberg fed llama2 entire reams of alignment propaganda and omitted important facts from the core model training.
Llama 1 is miles better.
Nah, not really. Llama 2 is not trained in actual propaganda. Only the chat version is.
However there are fine tuned variants of llama 2 which can be the most racist models so yeah.
Llama 2 is better in roleplaying, code, general whim llama is better in translation
I've been using the 7b llama2 models various fine tuned ones, and would like to test out some 13b size models. On my PC I have a single 4090 GPU (24GB), and 96 GB of DDR5, so will I be able to run 13b in a reasonble way (or at all) ?
I've been using the 7b llama2 models various fine tuned ones, and would like to test out some 13b size models. On my PC I have a single 4090 GPU (24GB), and 96 GB of DDR5, so will I be able to run 13b in a reasonble way (or at all) ?
With that hardware, forget 13B models and go higher, for better quality!
Thanks @MrDevolver I tried 13b and its working well, Ill try the next size up, actually can do you haver any suggestions for a specific model, as I am looking for something that has good knowledge - unsensored, and fairly generic to start with, but one I can easily fine tune via interactions/conversations. The idea is to use it as a personality in a mobile robot around the house (where the inference is performed back to my PC via web services through wifi). I am going to connect it with semantic network to give it more memory of specific events/personal facts etc.
Fine tuning would be a bit tricky as it takes up more vram and ram then inference. I think 13b models are best if you want to fine tune them.
If you just want to do inference, 34b models will probably be better. Most 34b models are designed for code so that might be a problem however
Thanks @MrDevolver I tried 13b and its working well, Ill try the next size up, actually can do you haver any suggestions for a specific model, as I am looking for something that has good knowledge - unsensored, and fairly generic to start with, but one I can easily fine tune via interactions/conversations. The idea is to use it as a personality in a mobile robot around the house (where the inference is performed back to my PC via web services through wifi). I am going to connect it with semantic network to give it more memory of specific events/personal facts etc.
Unfortunately I can't suggest any specific model beyond 13B range, because I don't have the required hardware so I don't have personal experience with any of them, but I heard that Spicyboros is a good general purpose model, so if I had the hardware like you have, I would try https://huggingface.co/TheBloke/Spicyboros-70B-2.2-GGUF?not-for-all-audiences=true
70b is probably bit too big. Maybe with ggml q2 or exllama q2
The 70b can be loaded with llamacpp with offloading to GPU, possibly 32 layers with the - ngf 32 flag. If that doesn't work quantize or use the quantized q4 version
Q4 probably isnt working as it’s a bit too big but q2 might work.
Thanks for the suggestions.