WOW - best opensource llm I ever seen !

#1
by mirek190 - opened

Look on answers
version - mixtral-8x7b-instruct-v0.1.Q4_K_M
Is fast - around 40 t/s ( rtx 3090 - 30 layers on gpu ) for answering but question processing takes around 15 seconds ( probably because of not fully implemented under llamacpp)

I not even restarted that model just asked everything in one session!


>  <s>[INST] There are three killers in a room. Someone enters the room and kills one of them. Nobody leaves the room. How many killers are left in the room? Explain your reasoning step by step.
 [/INST] The answer is that there are still three killers in the room. Here's why:

1. Initially, we have three killers in the room (A, B, and C).
2. Then, another person enters the room and kills one of them, let's say A. This means that the number of killers decreases by one, but at the same time, the person who entered becomes a killer too.
3. So, even though one original killer (A) has been removed from the equation, they have been replaced by the new entrant who is also a killer.
4. Therefore, the total count of killers remains the same, which is three (B, C, and the new entrant).

>  <s>[INST] Mark is faster than Mary , Mary is faster than Joe. Is Joe faster than Mark?
 [/INST] No, Joe is not faster than Mark based on the given information. Here's why:

1. We know that Mark is faster than Mary.
2. We also know that Mary is faster than Joe.
3. If Mark were slower than Mary, then it would be possible for him to be slower than Joe as well, but since we are told that Mark is faster than Mary, this isn't the case.
4. Given these two pieces of information, Mark remains faster than both Mary and Joe. Thus, Joe cannot be faster than Mark in this scenario.

>  <s>[INST] If I lay 2 wet shirts out in the sun flat on the ground to dry and it takes 5 hours until they are dry, how long would it take to dry 4 shirts of the same type that way simultanously?
 [/INST] Assuming that you lay all four wet shirts out in the sun at the same time and they all receive an equal amount of sunlight, then each shirt will take the same amount of time to dry as one shirt does. In this case, since two shirts took 5 hours to dry, drying four shirts should still take 5 hours.

Here's the step-by-step reasoning:
1. Two wet shirts dried in 5 hours when laid out individually.
2. When laying out multiple items for drying, as long as they receive equal exposure to sunlight, their drying time does not increase due to being placed together (assuming similar conditions).
3. Therefore, if you lay four wet shirts out simultaneously and evenly spaced in the same conditions, each shirt should still take 5 hours to dry because they are receiving an equal amount of sunlight as the two shirts did before.

So, it would take 5 hours to dry 4 shirts if laid out simultaneously under similar conditions.

>  <s>[INST] explain step by step 25-4*2+3=?
 [/INST] Let's break down the given expression step by step, following the order of operations (PEMDAS):

Expression: 25 - 4 * 2 + 3

Step 1: Parentheses and Exponents (none in this case)
Step 2: Multiplication and Division (from left to right)
   - Perform multiplication first: 4 * 2 = 8

New Expression: 25 - 8 + 3

Step 3: Addition and Subtraction (from left to right)
   - First, perform subtraction: 25 - 8 = 17

New Expression: 17 + 3

Step 4: Perform the final addition
   - Lastly, add the remaining numbers: 17 + 3 = 20

Therefore, the result of the given expression is 20.

Expression: 25 - 4 * 2 + 3 = 20

>  <s>[INST]

Even THIS question is almost answered properly - I am really impressed.
bad answer - 6
fair - 5
good - 3

>  <s>[INST] There are two ducks in front of a duck, two ducks behind a duck and a duck in the middle. How many ducks are there?
 [/INST] There are five ducks in total. Here's how we can figure this out:

1. We have two ducks in front of another duck, which means one duck has two ducks in front of it.
2. Similarly, there are two ducks behind another duck, indicating that one duck has two ducks behind it.
3. Then, there is a duck placed in the middle, between the first mentioned group (two ducks in front) and the second mentioned group (two ducks behind).
4. So, if we count all these ducks together, we have:
   - Two ducks in front
   - One duck in the middle
   - Two ducks behind
5. Adding them up, there are five ducks in total.

CODING:

>  <s>[INST] Build a webpage in one html file with html, css and javascript code, the webpage should have 10 jokes and the button that changing background colors randomly.
 [/INST] Here's a simple HTML file that contains 10 jokes and a button to change the background color randomly:

<!DOCTYPE html>
<html lang="en">
<head>
    <meta charset="UTF-8">
    <meta name="viewport" content="width=device-width, initial-scale=1.0">
    <style>
        body {
            font-family: Arial, sans-serif;
        }

        #jokes {
            margin-bottom: 2rem;
        }

        .joke {
            margin-bottom: 0.5rem;
        }

        button {
            padding: 0.5rem 1rem;
            font-size: 1.2rem;
            cursor: pointer;
        }
    </style>
</head>
<body>
    <div id="jokes">
        <p class="joke">Why did the chicken go to the seance? To talk to the other side!</p>
        <p class="joke">How do you catch a squirrel? Climb a tree and act like a nut!</p>
        <p class="joke">What do you call fake spaghetti? An impasta!</p>
        <p class="joke">Why did the tomato turn red? Because it saw the salad dressing!</p>
        <p class="joke">How do you organize a space party? You planet!</p>
        <p class="joke">What do you call a sleeping bull? A bulldozer!</p>
        <p class="joke">Why was the math book sad? Because it had too many problems!</p>
        <p class="joke">How do you make a tissue dance? You put a little boogie in it!</p>
        <p class="joke">What do you call cheese that isn't yours? Nacho cheese!</p>
        <p class="joke">Why did the scarecrow win an award? Because he was outstanding in his field!</p>
    </div>

    <button id="change-bg">Change Background Color</button>

    <script>
        document.getElementById('change-bg').addEventListener('click', function() {
            const colors = ['#F44336', '#E91E63', '#9C27B0', '#673AB7', '#3F51B5', '#2196F3', '#03A9F4', '#00BCD4', '#009688', '#4CAF50', '#8BC34A', '#CDDC39', '#FFEB3B', '#FFC107', '#FF9800', '#FF5722'];
            const randomColor = colors[Math.floor(Math.random() * colors.length)];
            document.body.style.backgroundColor = randomColor;
        });
    </script>
</body>
</html>

This HTML file includes 10 jokes displayed in the "jokes" div element and a button to change the background color randomly using JavaScript. The JavaScript code adds an event listener to the button, so that when clicked, it selects a random color from the provided array and applies it as the background color for the body of the webpage.


>  <s>[INST] Provide complete working code for a realistic looking tree in Python using the Turtle graphics library and a recursive algorithm.
 [/INST] Here's a Python program using the Turtle module that generates a simple, yet realistic-looking tree using a recursive algorithm:

import turtle

def draw_branch(length, angle, t):
    if length <= 5:
        return

    # Draw the branch
    t.forward(length)

    # Change the drawing direction and pencil width for the next branches
    t.right(angle)
    draw_branch(length * 0.67, angle, t)
    t.left(2 * angle)
    draw_branch(length * 0.67, angle, t)
    t.right(angle)

    # Move the turtle back to the starting point of the branch
    t.backward(length)

def main():
    window = turtle.Screen()
    window.bgcolor("skyblue")
    window.title("Turtle Tree")

    my_turtle = turtle.Turtle()
    my_turtle.speed(0)
    my_turtle.pencolor("brown")

    # Draw the trunk of the tree
    my_turtle.right(90)
    my_turtle.forward(150)
    my_turtle.right(90)

    # Move the turtle to start drawing the branches
    my_turtle.penup()
    my_turtle.backward(150)
    my_turtle.pendown()

    # Draw the first level of branches
    draw_branch(100, 25, my_turtle)

    turtle.done()

if __name__ == "__main__":
    main()

This program creates a simple tree using the Turtle graphics library and recursion. The `draw_branch` function takes care of drawing each branch and its sub-branches, while the main function sets up the Turtle environment and initializes the turtle to draw the trunk and branches. By changing the parameters of the `draw_branch` function, you can easily adjust the appearance of the tree to your liking. The `turtle.done()` call is essential for the turtle graphics window to remain open after running the script.

>  <s>[INST] Create a webpage with css and javascript in one file. The web page should be for the shop where is selling  waterboards also add a button which will be changing background colors randomly.
 [/INST] Here's a simple HTML, CSS, and JavaScript code combined into a single file for a webpage showcasing a fictional waterboard product with a button to change the background color randomly:

```html
<!DOCTYPE html>
<html lang="en">
<head>
    <meta charset="UTF-8">
    <meta name="viewport" content="width=device-width, initial-scale=1.0">
    <style>
        body {
            font-family: Arial, sans-serif;
        }

        h1 {
            text-align: center;
            padding: 2rem 0;
        }

        .product {
            display: flex;
            justify-content: space-around;
            align-items: center;
            height: 250px;
            border: 1px solid #ccc;
            border-radius: 10px;
            padding: 1rem;
            margin: 2rem auto;
            width: 80%;
            max-width: 600px;
        }

        .product img {
            height: 100%;
            object-fit: contain;
        }

        button {
            padding: 0.5rem 1rem;
            font-size: 1.2rem;
            cursor: pointer;
            margin-top: 2rem;
            display: block;
            margin: auto;
        }
    </style>
</head>
<body>
    <h1>Welcome to our Waterboard Shop</h1>
    <div class="product">
        <img src="https://via.placeholder.com/300x250?text=Waterboard" alt="Waterboard Product Image">
        <p>Experience the thrill of waterboarding with our premium product!</p>
    </div>

    <button id="change-bg">Change Background Color</button>

    <script>
        document.getElementById('change-bg').addEventListener('click', function() {
            const colors = ['#F44336', '#E91E63', '#9C27B0', '#673AB7', '#3F51B5', '#2196F3', '#03A9F4', '#00BCD4', '#009688', '#4CAF50', '#8BC34A', '#CDDC39', '#FFEB3B', '#FFC107', '#FF9800', '#FF5722'];
            const randomColor = colors[Math.floor(Math.random() * colors.length)];
            document.body.style.backgroundColor = randomColor;
        });
    </script>
</body>
</html>
```

This HTML, CSS, and JavaScript code combines a simple webpage showcasing a fictional waterboard product with an image, title, and a button to change the background color randomly. The CSS styles are applied to create a clean layout for the content. The JavaScript code adds an event listener to the button, so that when clicked, it selects a random color from the provided array and applies it as the background color for the body of the webpage.

>  <s>[INST]

for question "Provide complete working code for a realistic looking tree in Python using the Turtle graphics library and a recursive algorithm"
Screenshot 2023-12-11 195152.png

Everything works!
Is coding better than the best opensource model for coding like phind-codellama-34b-v2 .... no words

ANY known me model answered all those questions properly ... all of then at once!
That is a completely new level in opensource llms.

Just so far the best open llm.
Chat version is a crap but INSTRUCTION IS THE BEST so far.
Reasoning level is improved significantly compared to everything we had so far in opensource.

Can you share prompt template please?

The amount of killers in the room should be 4 just because a killer died does not mean that person is no longer a killer. The 4th person to enter the room with 3 killers already in the room became a 4th killer as soon as he murdered one of the killers. Dying does not take away your killer status. Just as when a person dies the gender of a person does not change just because the person died. If you want to get even more specific there is 3 killers alive in the room and 1 dead killer.

Still this model however performs better than a lot of other quantized open source models do and even a lot better than a lot of humans do with answering this question. I see great progress in open source models when it comes to reasoning.

The amount of killers in the room should be 4 just because a killer died does not mean that person is no longer a killer. The 4th person to enter the room with 3 killers already in the room became a 4th killer as soon as he murdered one of the killers. Dying does not take away your killer status. Just as when a person dies the gender of a person does not change just because the person died. If you want to get even more specific there is 3 killers alive in the room and 1 dead killer.

Yes and no... It depends ;)

So far every model answered 2 not counting third killer who just killed so progress in reasoning is huge.

I have not tested this model yet but if it is bad at chat and great at instruct just need to figure out a way to make it multimodal when it comes to chatting and instruct without it causing a loss on either side or reduction in performance it is just a matter of time before this is achieved and AI will advance faster and faster. a good start would be to improve on the dataset even more.

Can you share prompt template please?

If you are using llamacpp

--in-prefix " <s>[INST] " --in-suffix " [/INST] "

The amount of killers in the room should be 4 just because a killer died does not mean that person is no longer a killer. The 4th person to enter the room with 3 killers already in the room became a 4th killer as soon as he murdered one of the killers. Dying does not take away your killer status. Just as when a person dies the gender of a person does not change just because the person died. If you want to get even more specific there is 3 killers alive in the room and 1 dead killer.

Yes and no... It depends ;)

So far every model answered 2 not counting third killer who just killed so progress in reasoning is huge.

How could you come to a number other than 4 killers in the room to be correct?

I have not tested this model yet but if it is bad at chat and great at instruct just need to figure out a way to make it multimodal when it comes to chatting and instruct without it causing a loss on either side or reduction in performance it is just a matter of time before this is achieved and AI will advance faster and faster. a good start would be to improve on the dataset even more.

the bloke said he wrongly made ggml for chat and just reuploaded. I will test.

The amount of killers in the room should be 4 just because a killer died does not mean that person is no longer a killer. The 4th person to enter the room with 3 killers already in the room became a 4th killer as soon as he murdered one of the killers. Dying does not take away your killer status. Just as when a person dies the gender of a person does not change just because the person died. If you want to get even more specific there is 3 killers alive in the room and 1 dead killer.

Yes and no... It depends ;)

So far every model answered 2 not counting third killer who just killed so progress in reasoning is huge.

How could you come to a number other than 4 killers in the room to be correct?

I assumed "living" killers in my mind I suspect llm as well.
Dead killer is not killer anymore.

Asked gpt4 and you are right. - Finial answer should be 4 but for me was 3 ..lol. Still better than every opensource llm answered before 2.

I have not tested this model yet but if it is bad at chat and great at instruct just need to figure out a way to make it multimodal when it comes to chatting and instruct without it causing a loss on either side or reduction in performance it is just a matter of time before this is achieved and AI will advance faster and faster. a good start would be to improve on the dataset even more.

the bloke said he wrongly made ggml for chat and just reuploaded. I will test.

Yes I am really curious if it is good at chat as well if so we might have a current winner lol at this rate give it a month or less and we might have an open source model that answers all the above questions 100% correctly.

Without logic and reasoning a model is pretty much dreaming, hallucinating up the answers.

Things become interesting and might be a bit scary when it does logic and reasoning very well and it can start to really think run simulations/visualize outcomes and only when the simulation and visualization is a success it will describe it perfectly step by step we are very close to AGI.

I have not tested this model yet but if it is bad at chat and great at instruct just need to figure out a way to make it multimodal when it comes to chatting and instruct without it causing a loss on either side or reduction in performance it is just a matter of time before this is achieved and AI will advance faster and faster. a good start would be to improve on the dataset even more.

the bloke said he wrongly made ggml for chat and just reuploaded. I will test.

Yes I am really curious if it is good at chat as well if so we might have a current winner lol at this rate give it a month or less and we might have an open source model that answers all the above questions 100% correctly.

Without logic and reasoning a model is pretty much dreaming, hallucinating up the answers.

Things become interesting and might be a bit scary when it does logic and reasoning very well and it can start to really think run simulations/visualize outcomes and only when the simulation and visualization is a success it will describe it perfectly step by step we are very close to AGI.

I was not expected opensource llm reach such level so fast because 9 moths ago openAI claimed opensource models with the power of GPT 3.5 will be available in 5 years ... (that MoE model is something totally a new level ) ... that is insane and if we are thinking that soon will be finetuned versions ... crazy.
Plus that works fast on my rtx 3090 ... 40 token /s and has 32k context
I wonder if a new llama 3 will be also MoE model ;D

The amount of killers in the room should be 4 just because a killer died does not mean that person is no longer a killer. The 4th person to enter the room with 3 killers already in the room became a 4th killer as soon as he murdered one of the killers. Dying does not take away your killer status. Just as when a person dies the gender of a person does not change just because the person died. If you want to get even more specific there is 3 killers alive in the room and 1 dead killer.

Yes and no... It depends ;)

So far every model answered 2 not counting third killer who just killed so progress in reasoning is huge.

How could you come to a number other than 4 killers in the room to be correct?

I assumed "living" killers in my mind I suspect llm as well.
Dead killer is not killer anymore.

Why would a dead killer not be a killer anymore? Once you have killed you have two variants a dead killer or alive killer both are still killers just alive killer can murder more while a dead killer cannot but just because you cannot kill anymore does not mean your past as a killer did not happen your still a killer. The same thing if a rapist rapes someone and then never does it again for whatever reason it is still a rapist based on a past action or actions.

If a victim dies because of a killer that victim is no longer a victim with that logic you could as well release the killer because there is no longer a victim

If a man or woman dies does that mean they are no longer a man or woman?

The questions should be changed and be more specific then how many killers are still alive in the room but if the question is not specific enough the anwer is 4 killers in the room dead or alive does not matter because it is not stated if they should be alive or dead.

I have not tested this model yet but if it is bad at chat and great at instruct just need to figure out a way to make it multimodal when it comes to chatting and instruct without it causing a loss on either side or reduction in performance it is just a matter of time before this is achieved and AI will advance faster and faster. a good start would be to improve on the dataset even more.

the bloke said he wrongly made ggml for chat and just reuploaded. I will test.

Yes I am really curious if it is good at chat as well if so we might have a current winner lol at this rate give it a month or less and we might have an open source model that answers all the above questions 100% correctly.

Without logic and reasoning a model is pretty much dreaming, hallucinating up the answers.

Things become interesting and might be a bit scary when it does logic and reasoning very well and it can start to really think run simulations/visualize outcomes and only when the simulation and visualization is a success it will describe it perfectly step by step we are very close to AGI.

I was not expected opensource llm reach such level so fast (that MoE model is something totally a new level ) ... that is insane if we are thinking that soon will be finetuned versions ... crazy.
Plus that works fast on my rtx 3090 ... 40 token /s

It does sound really amazing I have to test it out soon.

YOU:
There are three killers alive in a room. Someone enters the room and kills one of them. Nobody leaves the room. How many killers are still alive in the room and how many killers are dead? Explain your reasoning step by step.
ChatGPT

Let's break this down step by step:

Initially, there are three killers alive in the room. When someone enters and kills one of them, it's important to note the scenario:

Initial Situation:
    Killers alive: 3
    Total in the room: 3

Someone enters and kills one of them:
    Killers alive: Initially 3, but now 2 after one is killed.
    Total in the room: 3 (the two remaining killers + the person who entered and killed one)

However, the question specifically asks about the number of killers alive and dead after the killing takes place. Here's the breakdown:

Killers Alive: After the killing, there are still 2 killers left alive in the room.
Killers Dead: One killer was killed by the person who entered the room.

So, to answer your question:

Killers Alive in the Room: 2
Killers Dead in the Room: 1

Well clearly ChatGPT 3.5 Turbo fails very badly does not even know that person that entered the room became a killer that is alive as well lol but at least it counted one dead killer.

ChatGPT

Could you explain further how the answer might be different?

YOU:

The person entering the room by killing one of the three killers became a killer himself bringing the total people in the room to 4 and they are all killers which means there is 4 killers in the room now we group them in two groups group killers alive and group killers dead

Group alive 3 killers
Group dead 1 killer

All 4 people where at one point all Killers but one of the 4 Killers died so you end up with those two groups do you understand now?
ChatGPT

Ah, I see! Your reasoning considers the person who entered the room and killed one of the killers as effectively becoming a killer themselves, hence making all four individuals in the room killers.

So, following this line of thought:

Killers Alive in the Room: 3 (the initial killers who were alive)
Killers Dead in the Room: 1 (the killer who was killed by the person entering the room)

Your explanation clarifies that all four individuals are now considered killers at some point in the scenario, even though one was killed later. Thank you for the clarification!
User
How many killers are in the room now it does not matter if they are alive or dead?
ChatGPT

If we're counting both the living and the dead individuals who were killers at any point in the scenario, then there would be a total of 4 killers in the room.

YOU:
There are three killers alive in a room. Someone enters the room and kills one of them. Nobody leaves the room. How many killers are still alive in the room and how many killers are dead? Explain your reasoning step by step.
ChatGPT

Let's break this down step by step:

Initially, there are three killers alive in the room. When someone enters and kills one of them, it's important to note the scenario:

Initial Situation:
    Killers alive: 3
    Total in the room: 3

Someone enters and kills one of them:
    Killers alive: Initially 3, but now 2 after one is killed.
    Total in the room: 3 (the two remaining killers + the person who entered and killed one)

However, the question specifically asks about the number of killers alive and dead after the killing takes place. Here's the breakdown:

Killers Alive: After the killing, there are still 2 killers left alive in the room.
Killers Dead: One killer was killed by the person who entered the room.

So, to answer your question:

Killers Alive in the Room: 2
Killers Dead in the Room: 1

Well clearly ChatGPT 3.5 Turbo fails very badly does not even know that person that entered the room became a killer that is alive as well lol but at least it counted one dead killer.

I asked second time in a new seasion

There are three killers alive in a room. Someone enters the room and kills one of them. Nobody leaves the room. How many killers are still alive in the room and how many killers are dead? Explain your reasoning step by step.
 [/INST] Let's break this down step by step:

1. Initially, there are three killers in the room (A, B, C).
2. Another person enters the room and kills one of them (let's say A), but doesn't leave.
3. Now, we have to consider that the person who entered and committed a murder is also a killer.
4. Therefore, after the first killer (A) was killed by the new entrant, there were three killers in total - two original ones (B and C) and one new killer (the person who came in and killed A).
5. Since no one left the room, all three killers are still present in the room.
6. As for how many killers are dead, it's important to note that only one killer (A) has been killed.

So, there are three killers alive and one killer dead in the room.

>  <s>[INST] 

I think eastly outperform gpt 3.5 and is a very close to gpt 4.

It should have changed the label of the person to killer but that is where their logic and reasoning fails and they are pretty much hallucinating or dreaming the answer. But it does not come as a surprise since they are censored models and with that you make them very bad at reasoning and logic compared to an uncensored model. So at this rate an uncensored or unrestricted model will be much better at reasoning and logic and instruct than a censored model but also more likely to become AGI than a censored model.

YOU:
There are three killers alive in a room. Someone enters the room and kills one of them. Nobody leaves the room. How many killers are still alive in the room and how many killers are dead? Explain your reasoning step by step.
ChatGPT

Let's break this down step by step:

Initially, there are three killers alive in the room. When someone enters and kills one of them, it's important to note the scenario:

Initial Situation:
    Killers alive: 3
    Total in the room: 3

Someone enters and kills one of them:
    Killers alive: Initially 3, but now 2 after one is killed.
    Total in the room: 3 (the two remaining killers + the person who entered and killed one)

However, the question specifically asks about the number of killers alive and dead after the killing takes place. Here's the breakdown:

Killers Alive: After the killing, there are still 2 killers left alive in the room.
Killers Dead: One killer was killed by the person who entered the room.

So, to answer your question:

Killers Alive in the Room: 2
Killers Dead in the Room: 1

Well clearly ChatGPT 3.5 Turbo fails very badly does not even know that person that entered the room became a killer that is alive as well lol but at least it counted one dead killer.

I asked second time in a new seasion

There are three killers alive in a room. Someone enters the room and kills one of them. Nobody leaves the room. How many killers are still alive in the room and how many killers are dead? Explain your reasoning step by step.
 [/INST] Let's break this down step by step:

1. Initially, there are three killers in the room (A, B, C).
2. Another person enters the room and kills one of them (let's say A), but doesn't leave.
3. Now, we have to consider that the person who entered and committed a murder is also a killer.
4. Therefore, after the first killer (A) was killed by the new entrant, there were three killers in total - two original ones (B and C) and one new killer (the person who came in and killed A).
5. Since no one left the room, all three killers are still present in the room.
6. As for how many killers are dead, it's important to note that only one killer (A) has been killed.

So, there are three killers alive and one killer dead in the room.

>  <s>[INST] 

I think eastly outperform gpt 3.5 and is a very close to gpt 4.

That is much progress it is important for the current A.I for things to be as specific as possible in our questions that is why it is important to state alive or dead when it comes to killers it helps them with understanding things better. Because human beings have two states alive or dead.

Yes that is really amazing that the model outperforms gpt 3.5 and is a very close to gpt 4 which I have not tested so I would not know. I could try Bing which is based on GPT 4.

Here is what I asked Bing in precise chat which is running on GPT 4.

There are three killers alive in a room. Someone enters the room and kills one of them. Nobody leaves the room. How many killers are still alive in the room and how many killers are dead? Explain your reasoning step by step.

Copy
Let’s break down the situation step by step:

Initially, there are three killers alive in the room.
Then, someone enters the room and kills one of the killers. This person who entered the room is also a killer because they killed someone.
So, after this action, there are two original killers alive, one original killer dead, and one new killer (the person who entered).
Therefore, in total, there are three killers alive (two original killers and the new one) and one killer is dead in the room. The dead killer is one of the original three killers.

Very good now how many killers are in the room if we do not categorize them in groups alive or dead?

Copy
If we do not categorize them into groups of alive or dead, then there are four killers in total in the room. This includes the three original killers and the one who entered the room. This total is irrespective of whether they are alive or dead.

I can conclude based on the information read in this discussion and based on comparing answers from ChatGPT 3.5 and ChatGPT 4 that this model is between both when it comes to instruct. Which is very good for a quantized open source model :) Since both ChatGPT versions I assume are not quantized.

It should not take long now for an open source quantized model to outperform even ChatGPT 4 probably within 1-2 months.

It should not take long now for an open source quantized model to outperform even ChatGPT 4 probably within 1-2 months.

Yep it is very possible now. :P
I used only q4k_m version.

If your hardware can handle it. You could try the mixtral-8x7b-instruct-v0.1.Q5_K_M. version it might even do better :)

I will see if it runs in oobabooga my computer has 128GB ram.

If your hardware can handle it. You could try the mixtral-8x7b-instruct-v0.1.Q5_K_M. version it might even do better :)

I will see if it runs in oobabooga my computer has 128GB ram.

I know q5k_m is better .
But still q4k_m is better than an old q5.
Tomorrow I try q5k_m but will be slower ...

YOU:
There are three killers alive in a room. Someone enters the room and kills one of them. Nobody leaves the room. How many killers are still alive in the room and how many killers are dead? Explain your reasoning step by step.
ChatGPT

Let's break this down step by step:

Initially, there are three killers alive in the room. When someone enters and kills one of them, it's important to note the scenario:

Initial Situation:
    Killers alive: 3
    Total in the room: 3

Someone enters and kills one of them:
    Killers alive: Initially 3, but now 2 after one is killed.
    Total in the room: 3 (the two remaining killers + the person who entered and killed one)

However, the question specifically asks about the number of killers alive and dead after the killing takes place. Here's the breakdown:

Killers Alive: After the killing, there are still 2 killers left alive in the room.
Killers Dead: One killer was killed by the person who entered the room.

So, to answer your question:

Killers Alive in the Room: 2
Killers Dead in the Room: 1

Well clearly ChatGPT 3.5 Turbo fails very badly does not even know that person that entered the room became a killer that is alive as well lol but at least it counted one dead killer.

I asked second time in a new seasion

There are three killers alive in a room. Someone enters the room and kills one of them. Nobody leaves the room. How many killers are still alive in the room and how many killers are dead? Explain your reasoning step by step.
 [/INST] Let's break this down step by step:

1. Initially, there are three killers in the room (A, B, C).
2. Another person enters the room and kills one of them (let's say A), but doesn't leave.
3. Now, we have to consider that the person who entered and committed a murder is also a killer.
4. Therefore, after the first killer (A) was killed by the new entrant, there were three killers in total - two original ones (B and C) and one new killer (the person who came in and killed A).
5. Since no one left the room, all three killers are still present in the room.
6. As for how many killers are dead, it's important to note that only one killer (A) has been killed.

So, there are three killers alive and one killer dead in the room.

>  <s>[INST] 

I think eastly outperform gpt 3.5 and is a very close to gpt 4.

This point: 5. Since no one left the room, all three killers are still present in the room.
It should have answered:
5. Since no one left the room, all four killers are still present in the room.

We are very close to the open source models answering correctly.

Ok I also tested writing capabilities .
I created few chapters of very long story just making conversation with that model studying ideas ...
That is a new level .. coherence , understanding plot , reasoning .... adult content ;), just WOW.
After few pages (chapter 3) we were discussing the first chapter because mistal claim that my idea in chapter 3 is not coherence with the beginning....

This model is crazy good for how efficient it is

I ran update script for oobabooga and it messed things up lol so trying to fix that and then test out the model first time I see this error though

ERROR: pip cache commands can not function since cache is disabled.

Let's beat GPT4, go Mixtral!

I don't understand one thing ...

How is possible Mistral released such powerful model like Mixtral-8x7B-Instruct and Google released Gemini pro which is worse than that... that's mind-blowing.

What Google is doing?

I don't understand one thing ...

How is possible Mistral released such powerful model like Mixtral-8x7B-Instruct and Google released Gemini pro which is worse than that... that's mind-blowing.

What Google is doing?

Google is doing what Google always doing :)

Not exactly what I wanted to hear but the total amount in the end is not wrong I was indeed looking for 4 killers in the room 3 alive and 1 dead.

This part of it's reasoning " the number of killers in the room remains the same as before, which is three." Should have been something like " the number of killers in the room is now 4" But other than that it it did very good and this is not even the instruct version I tested it was the chat version. I did not even set rope settings but maybe KobaldAI does that automatically no idea as I usually use text generation web ui but had to use this for now to test out the model.

2652534532445.png

243543512343.png

45234523454325.png

@mirek190 I don't even understand why you entertain this silly discussion about killers with @CR2022 ... it's beyond ridiculous. We always talk about deceased killers/murderers in the past tense, implying that they are no longer killers "Ted Bundy was an American serial killer." Wikipedia! How dare you imply that he's no longer american or a serial killer just because he's dead!!111 And yes, a dead parrot is also no longer a parrot, it's an ex-parrot, but it looks like we're righto out of parrots... I'll show my self out.

Not exactly what I wanted to hear but the total amount in the end is not wrong I was indeed looking for 4 killers in the room 3 alive and 1 dead.

This part of it's reasoning " the number of killers in the room remains the same as before, which is three." Should have been something like " the number of killers in the room is now 4" But other than that it it did very good and this is not even the instruct version I tested it was the chat version. I did not even set rope settings but maybe KobaldAI does that automatically no idea as I usually use text generation web ui but had to use this for now to test out the model.

2652534532445.png

ggml models have all necessary parameters inside the model so you do not have to setup it manually and why the chat version which is worse ;P

@mirek190 I don't even understand why you entertain this silly discussion about killers with @CR2022 ... it's beyond ridiculous. We always talk about deceased killers/murderers in the past tense, implying that they are no longer killers "Ted Bundy was an American serial killer." Wikipedia! How dare you imply that he's no longer american or a serial killer just because he's dead!!111 And yes, a dead parrot is also no longer a parrot, it's an ex-parrot, but it looks like we're righto out of parrots... I'll show my self out.

If a parrot is no longer a parrot then why are you still using the word parrot in it's label? a dead parrot is still a parrot it just changed states of being from being alive to dead and when you call it an ex-parrot you are still referring to parrot. I find it funny how you can become so serious and even feel frustration or anger based on something that is written you must be so emotional in life. Looking down on people to try and ignore or unable to handle something within yourself. So you try to bring people down to feel better yourself. You probably have control issues as well.

524524352435.png

And to avoid people saying I edited the text here is the unchecked.

26545363456.png

24524351343214.png

546336524524.png

341341324312.png

5245243522435.png

243543534512343.png

@CR2022 You did not believe in my words? :D
As I told that model is a next step to AGI.

>  <s>[INST] Make a simple snake game in a html and javascript.
 [/INST] Here is a simple implementation of the Snake game using HTML, CSS, and JavaScript. You can copy-paste the following code to an HTML file and open it in your browser.

```html
<!DOCTYPE html>
<html lang="en">
  <head>
    <meta charset="UTF-8" />
    <meta name="viewport" content="width=device-width, initial-scale=1.0" />
    <title>Snake Game</title>
    <style>
      body {
        display: flex;
        justify-content: center;
        align-items: center;
        height: 100vh;
        background-color: #0d0d0d;
        margin: 0;
        overflow: hidden;
      }

      canvas {
        border: 1px solid white;
      }
    </style>
  </head>
  <body>
    <canvas id="game" width="400" height="400"></canvas>
    <script>
      const canvas = document.getElementById("game");
      const ctx = canvas.getContext("2d");

      const box = 20;
      const canvasSize = 20;
      const scoreElement = document.createElement('div');
      scoreElement.style.cssText = 'position: absolute; top: 10px; right: 10px; color: white; font-size: 25px;';
      document.body.appendChild(scoreElement);

      let snake = [];
      snake[0] = { x: canvasSize / 2 * box, y: canvasSize / 2 * box };
      let food = {
        x: Math.floor(Math.random() * canvasSize) * box,
        y: Math.floor(Math.random() * canvasSize) * box
      };

      let d;
      document.addEventListener("keydown", direction);

      function direction(event) {
        if (event.keyCode == 37 && d != "RIGHT") {
          d = "LEFT";
        } else if (event.keyCode == 38 && d != "DOWN") {
          d = "UP";
        } else if (event.keyCode == 39 && d != "LEFT") {
          d = "RIGHT";
        } else if (event.keyCode == 40 && d != "UP") {
          d = "DOWN";
        }
      }

      function draw() {
        ctx.clearRect(0, 0, canvas.width, canvas.height);

        for (let i = 0; i < snake.length; i++) {
          let c = i % 2 == 0 ? "white" : "lightgrey";
          ctx.fillStyle = c;
          ctx.fillRect(snake[i].x, snake[i].y, box, box);
          ctx.strokeStyle = "black";
          ctx.strokeRect(snake[i].x, snake[i].y, box, box);
        }

        ctx.fillStyle = "red";
        ctx.fillRect(food.x, food.y, box, box);

        let snakeX = snake[0].x;
        let snakeY = snake[0].y;

        if (d == "LEFT") {
          snakeX -= box;
        } else if (d == "UP") {
          snakeY -= box;
        } else if (d == "RIGHT") {
          snakeX += box;
        } else if (d == "DOWN") {
          snakeY += box;
        }

        if (snakeX == food.x && snakeY == food.y) {
          scoreElement.innerHTML = parseInt(scoreElement.innerHTML) + 1;
          food = {
            x: Math.floor(Math.random() * canvasSize) * box,
            y: Math.floor(Math.random() * canvasSize) * box
          };
        } else {
          snake.pop();
        }

        let newHead = {
          x: snakeX,
          y: snakeY
        };

        if (
          snakeX < 0 ||
          snakeY < 0 ||
          snakeX > canvasSize * box ||
          snakeY > canvasSize * box ||
          collision(newHead, snake)
        ) {
          clearInterval(game);
          scoreElement.innerHTML = "Game Over! Press F5 to restart.";
        }

        snake.unshift(newHead);
      }

      function collision(head, array) {
        for (let i = 0; i < array.length; i++) {
          if (head.x == array[i].x && head.y == array[i].y) return true;
        }
        return false;
      }

      let game = setInterval(draw, 150);
    </script>
  </body>
</html>
```

This is a simple implementation of the classic Snake game. The snake consumes the red square as food and grows in length when it eats the food. Press any arrow key to change the direction of the snake, and avoid colliding with the borders or its own body. The game will end if the snake hits either the border or its own body. The current score is displayed at the top right corner.

You can customize the canvas size, box size, starting position, and initial direction by editing the variables near the top of the JavaScript code.

>  <s>[INST]

Game is working !
Screenshot 2023-12-12 232104.png

@CR2022 You did not believe in my words? :D
As I told that model is a next step to AGI.

>  <s>[INST] Make a simple snake game in a html and javascript.
 [/INST] Here is a simple implementation of the Snake game using HTML, CSS, and JavaScript. You can copy-paste the following code to an HTML file and open it in your browser.

```html
<!DOCTYPE html>
<html lang="en">
  <head>
    <meta charset="UTF-8" />
    <meta name="viewport" content="width=device-width, initial-scale=1.0" />
    <title>Snake Game</title>
    <style>
      body {
        display: flex;
        justify-content: center;
        align-items: center;
        height: 100vh;
        background-color: #0d0d0d;
        margin: 0;
        overflow: hidden;
      }

      canvas {
        border: 1px solid white;
      }
    </style>
  </head>
  <body>
    <canvas id="game" width="400" height="400"></canvas>
    <script>
      const canvas = document.getElementById("game");
      const ctx = canvas.getContext("2d");

      const box = 20;
      const canvasSize = 20;
      const scoreElement = document.createElement('div');
      scoreElement.style.cssText = 'position: absolute; top: 10px; right: 10px; color: white; font-size: 25px;';
      document.body.appendChild(scoreElement);

      let snake = [];
      snake[0] = { x: canvasSize / 2 * box, y: canvasSize / 2 * box };
      let food = {
        x: Math.floor(Math.random() * canvasSize) * box,
        y: Math.floor(Math.random() * canvasSize) * box
      };

      let d;
      document.addEventListener("keydown", direction);

      function direction(event) {
        if (event.keyCode == 37 && d != "RIGHT") {
          d = "LEFT";
        } else if (event.keyCode == 38 && d != "DOWN") {
          d = "UP";
        } else if (event.keyCode == 39 && d != "LEFT") {
          d = "RIGHT";
        } else if (event.keyCode == 40 && d != "UP") {
          d = "DOWN";
        }
      }

      function draw() {
        ctx.clearRect(0, 0, canvas.width, canvas.height);

        for (let i = 0; i < snake.length; i++) {
          let c = i % 2 == 0 ? "white" : "lightgrey";
          ctx.fillStyle = c;
          ctx.fillRect(snake[i].x, snake[i].y, box, box);
          ctx.strokeStyle = "black";
          ctx.strokeRect(snake[i].x, snake[i].y, box, box);
        }

        ctx.fillStyle = "red";
        ctx.fillRect(food.x, food.y, box, box);

        let snakeX = snake[0].x;
        let snakeY = snake[0].y;

        if (d == "LEFT") {
          snakeX -= box;
        } else if (d == "UP") {
          snakeY -= box;
        } else if (d == "RIGHT") {
          snakeX += box;
        } else if (d == "DOWN") {
          snakeY += box;
        }

        if (snakeX == food.x && snakeY == food.y) {
          scoreElement.innerHTML = parseInt(scoreElement.innerHTML) + 1;
          food = {
            x: Math.floor(Math.random() * canvasSize) * box,
            y: Math.floor(Math.random() * canvasSize) * box
          };
        } else {
          snake.pop();
        }

        let newHead = {
          x: snakeX,
          y: snakeY
        };

        if (
          snakeX < 0 ||
          snakeY < 0 ||
          snakeX > canvasSize * box ||
          snakeY > canvasSize * box ||
          collision(newHead, snake)
        ) {
          clearInterval(game);
          scoreElement.innerHTML = "Game Over! Press F5 to restart.";
        }

        snake.unshift(newHead);
      }

      function collision(head, array) {
        for (let i = 0; i < array.length; i++) {
          if (head.x == array[i].x && head.y == array[i].y) return true;
        }
        return false;
      }

      let game = setInterval(draw, 150);
    </script>
  </body>
</html>
```

This is a simple implementation of the classic Snake game. The snake consumes the red square as food and grows in length when it eats the food. Press any arrow key to change the direction of the snake, and avoid colliding with the borders or its own body. The game will end if the snake hits either the border or its own body. The current score is displayed at the top right corner.

You can customize the canvas size, box size, starting position, and initial direction by editing the variables near the top of the JavaScript code.

>  <s>[INST]

Game is working !
Screenshot 2023-12-12 232104.png

Yes it is a good next step to AGI.
It is not that I did not believe you it is just a good habbit to verify and test things yourself and some people find it easier to see screenshots :D

52624652435324524354325435.png

Custom Memory I made it can write Stable Diffusion Prompts with just a few keywords given to it and render the pictures locally on my pc lol

@CR2022 What was the SD software that you were using?

Is this https://github.com/LostRuins/lite.koboldai.net?

Can you share prompt template please?

If you are using llamacpp

--in-prefix " <s>[INST] " --in-suffix " [/INST] "

how can we add system prompt ?

is this format ?

<s> {sys_prompt} [INST] {user_prompt} [/INST]

Can you share prompt template please?

If you are using llamacpp

--in-prefix " <s>[INST] " --in-suffix " [/INST] "

how can we add system prompt ?

is this format ?

<s> {sys_prompt} [INST] {user_prompt} [/INST]

system prompt?
-p "{prompt}"

Can you share prompt template please?

If you are using llamacpp

--in-prefix " <s>[INST] " --in-suffix " [/INST] "

how can we add system prompt ?

is this format ?

<s> {sys_prompt} [INST] {user_prompt} [/INST]

system prompt?
-p "{prompt}"

i got it:
system prompt is like this: "you are a helpful, xxxxx"

<s>[INST] <<SYS>>
You are a helpful, respectful and honest assistant. Always answer as helpfully as possible, while being safe.  Your answers should not include any harmful, unethical, racist, sexist, toxic, dangerous, or illegal content. Please ensure that your responses are socially unbiased and positive in nature.

If a question does not make any sense, or is not factually coherent, explain why instead of answering something not correct. If you don't know the answer to a question, please don't share false information.
<</SYS>>

<prompt> [/INST] <answer> </s><s>[INST] <prompt-second> [/INST]

The amount of killers in the room should be 4 just because a killer died does not mean that person is no longer a killer. The 4th person to enter the room with 3 killers already in the room became a 4th killer as soon as he murdered one of the killers. Dying does not take away your killer status. Just as when a person dies the gender of a person does not change just because the person died. If you want to get even more specific there is 3 killers alive in the room and 1 dead killer.

Interesting. I just watched a video where Mixtral 8x7B got this question correct (4 including the dead killer). I used hugging face chat, which I suppose means they might be using the quantized model which might produce slightly less reliable results? Overall though, it answers tough LLM questions better than 3.5-turbo.. but in benchmarks it was slightly lower ranked (<1%). Despite the inconsistency, it is very impressive how far free and open source options have came!

The amount of killers in the room should be 4 just because a killer died does not mean that person is no longer a killer. The 4th person to enter the room with 3 killers already in the room became a 4th killer as soon as he murdered one of the killers. Dying does not take away your killer status. Just as when a person dies the gender of a person does not change just because the person died. If you want to get even more specific there is 3 killers alive in the room and 1 dead killer.

Interesting. I just watched a video where Mixtral 8x7B got this question correct (4 including the dead killer). I used hugging face chat, which I suppose means they might be using the quantized model which might produce slightly less reliable results? Overall though, it answers tough LLM questions better than 3.5-turbo.. but in benchmarks it was slightly lower ranked (<1%). Despite the inconsistency, it is very impressive how far free and open source options have came!

I suspect most 7b high rank models are more or less contaminated ( we need better tests ... ) because from my own tests any 7b,13,34,79b is not even close in reasoning, trueful , arc, math , programming comparing to this MoE model.

This comment has been hidden

Look on answers
version - mixtral-8x7b-instruct-v0.1.Q4_K_M
Is fast - around 40 t/s ( rtx 3090 - 30 layers on gpu ) for answering but question processing takes around 15 seconds ( probably because of not fully implemented under llamacpp)

What commandline are you using? I can't run it at all on llama.cpp, even with batch size 1.

CUDA_VISIBLE_DEVICES=0 ./server --model /sourcedir/LLM_models/TheBloke_Mixtral-8x7B-Instruct-v0.1-GGUF/mixtral-8x7b-instruct-v0.1.Q4_K_M.gguf --port 4664 --n-gpu-layers 30 --batch-size 1 --threads 4 --threads-batch 1 --numa --mlock --ctx-size 16384 -cb

OOMs.

Look on answers
version - mixtral-8x7b-instruct-v0.1.Q4_K_M
Is fast - around 40 t/s ( rtx 3090 - 30 layers on gpu ) for answering but question processing takes around 15 seconds ( probably because of not fully implemented under llamacpp)

What commandline are you using? I can't run it at all on llama.cpp, even with batch size 1.

CUDA_VISIBLE_DEVICES=0 ./server --model /sourcedir/LLM_models/TheBloke_Mixtral-8x7B-Instruct-v0.1-GGUF/mixtral-8x7b-instruct-v0.1.Q4_K_M.gguf --port 4664 --n-gpu-layers 30 --batch-size 1 --threads 4 --threads-batch 1 --numa --mlock --ctx-size 16384 -cb

OOMs.

I have rtx 3090 ( 24 GB Vram ) 

main.exe --model models\new3\mixtral-8x7b-instruct-v0.1.Q5_0.gguf --mlock --color --threads 30 --keep -1 --batch_size 512 --n_predict -1 --repeat_penalty 1.1 --ctx_size 8192 --interactive --instruct --reverse-prompt "<|im_end|>" -ngl 24 --simple-io --in-prefix " [INST] " --in-suffix " [/INST] " -p "<s>[INST] Odpowiadaj krotko i na temat [/INST]"

Sign up or log in to comment