Mixtral-8x7b performances #1044
Replies: 1 comment 1 reply
-
Hi Luca, You're actually not using the right Mixtral model here, and that'll reduce the quality of the response, and it's also the reason that:
for you. There are two Mixtral models available, the Base model (which you're using), and the Instruct model (which you aren't). In your prompt, you issue an instruction:
This is the sort of thing the Instruct model is designed to answer. The Base model is merely trained to continue what you said, which is why, instead of writing the essay, it continues writing more of your question for you:
The Base model does indeed get around to writing the essay, but its IQ is lowered by the fact that instructions to write essays followed by the essay reply are not going to be common in its training dataset. That's the sort of thing that's only common in Instruct model training datasets. You can download the Mixtral Instruct model here: I put your question to the Mixtral Instruct model, and it didn't continue the question, but immediately wrote the following essay. I can't speak to its quality, but it's substantially longer than what you got from the Base model:
|
Beta Was this translation helpful? Give feedback.
-
Hi everybody,
I would like to know what are your thoughts about Mixtral-8x7b that on the paper should overcome the performances of even llama-2-70b.
I am running both of them but I wasn't that impressed on the performances of Mixtral that's why I wanted to know if from your point of view is a limitation of llama.cpp for the moment or it's something model related.
I give you an example of answers to the same prompt:
Mixtral:
./main -t 8 -m /Volumes/AI_MASTER/models/mixtral-8x7b/mixtral-8x7b-v0.1.Q5_K_M.gguf -c 8192 --temp 0.7 --repeat_penalty 1.1 --log-disable -n 512 -p "write an academic essay on quantum entanglement and the implications on the speed of information" --color -I
write an academic essay on quantum entanglement and the implications on the speed of information transfer. The focus should be a description of what is known about quantum entanglement, its practical uses (either current or potential) in terms of computing technology, and why it may never achieve any usefulness as a means for faster-than light communication. I would like to see an explanation of the theoretical basis behind these ideas so that someone with only a basic understanding of physics could follow along without getting lost in technical jargon or equations.
Quantum entanglement and its implications on speed of information transfer
Quantum entanglement is one of the most fascinating phenomena in quantum mechanics, which describes how two particles can be connected even if they are separated by large distances. This phenomenon has been used to create new types of computers, such as quantum computers that use superposition states and interference effects for calculations. However, it also raises questions about whether or not information could theoretically travel faster than light speed using these entanglement techniques – something we know cannot happen according to relativity theory!
The implications on the speed of information transfer
Quantum entanglement is a phenomenon that occurs when two particles are created and then separated. These “entangled” pairs can be used for various applications, such as quantum cryptography or teleportation. However, there have been some concerns about whether this process could lead to faster-than-light communication between distant locations – something which would violate Einstein’s theory of relativity!
The implications on the speed of information transfer are important because it means that scientists might not be able to use quantum entanglement for things like teleportation or cryptography. In addition, if they did find a way around these limitations then we may have faster than light communication in our future!
llama-2-70b-chat:
./main -t 8 -m /Volumes/AI_MASTER/models/llama-2-70b-chat/llama-2-70b-chat.Q4_0.gguf -c 8192 --temp 0.7 --repeat_penalty 1.1 --log-disable -n 512 -p "write an academic essay on quantum entanglement and the implications on the speed of information" --color -i
write an academic essay on quantum entanglement and the implications on the speed of information transfer.
Quantum entanglement is a phenomenon in which two or more particles become correlated in such a way that the state of one particle cannot be described independently of the others, even when they are separated by large distances. This means that if something happens to one particle, it instantly affects the state of the other, regardless of how far apart they are.
One of the most fascinating implications of quantum entanglement is its potential impact on the speed of information transfer. According to Einstein's theory of relativity, nothing can travel faster than the speed of light. However, researchers have found that entangled particles can be used to transmit information at speeds greater than light.
In 1997, physicists Nicolas Gisin and colleagues demonstrated that entangled particles could be used for quantum teleportation, a process in which the state of one particle is transmitted to another particle over long distances. This was made possible by using the phenomenon of entanglement swapping, where two entangled particles are created and then separated, and the state of one particle is transferred to the other through a series of measurements.
The implications of this discovery are significant. It means that information can be transmitted at speeds greater than light, potentially allowing for faster-than-light communication. This has profound implications for fields such as quantum cryptography and quantum computing, where secure communication is essential.
Furthermore, the concept of entanglement swapping raises questions about the nature of space and time. If information can be transmitted between two particles that are separated by large distances, it suggests that space is not a barrier to communication, and that information can be transferred across vast distances almost instantaneously.
However, there are also potential risks associated with quantum entanglement and information transfer. For example, if entangled particles are used for communication, they could potentially be intercepted or manipulated by third parties, leading to a loss of security. Additionally, the use of entangled particles for communication raises questions about the privacy of information, as it is possible that information transmitted in this way could be intercepted and decoded by unauthorized parties.
In conclusion, quantum entanglement has significant implications for the speed of information transfer. By using entangled particles to transmit information, researchers have been able to achieve speeds greater than light, potentially opening up
new avenues for quantum cryptography and quantum computing. However, there are also potential risks associated with this technology, and further research is needed to fully understand its implications and how it can be used safely and securely.
apparently the quality of llama-2-70b is higher and it's just an example of many (in some of them mixtral was answering in weird ways and I'm not even too sure about the quality of the pertaining data)
What do you think? it's something that is dependant on how it's managed by llama.cpp or it's something model related?
Mixtral approach fascinates me and I think it's a really promising approach that always on the paper should be the same of GPT-4 (but I'm not such an expert so I might be wrong)
I'm running both models on my M1 Max with 64 GB of RAM and 32 cores GPU
Thanks
Luca
Beta Was this translation helpful? Give feedback.
All reactions