Create your own CUSTOMIZED Llama 3 model using Ollama

00:12:55
https://www.youtube.com/watch?v=k39a--Tu4h0

الملخص

TLDRLe contenu fournit un guide détaillé sur la façon de personnaliser le modèle Llama 3 récemment mis à jour, qui est un modèle d'intelligence artificielle. Il commence par expliquer comment installer Llama 3 en utilisant des commandes dans le terminal, avec des instructions pour télécharger le modèle en utilisant 'olama pull'. La vidéo guide ensuite l'utilisateur à travers la création d'un fichier modèle personnalisé, en décrivant chaque ligne de code nécessaire. Les paramètres personnalisables incluent la "temperature" et les prompts système, qui peuvent être ajustés selon les besoins spécifiques de l'utilisateur. Des conseils sont donnés sur la façon d'obtenir ces paramètres en consultant les cartes des modèles sur olama.com. Le processus est également illustré en utilisant VS Code, bien que l'utilisateur puisse choisir n'importe quel éditeur pour créer le fichier modèle. Enfin, des informations sont fournies sur la façon de tester le modèle pour vérifier son bon fonctionnement et où trouver des informations supplémentaires pour personnaliser davantage les paramètres.

الوجبات الجاهزة

  • 🐙 Comprendre Llama 3 et sa récente mise à jour sur Olama.
  • 💻 Télécharger et installer Llama 3 via le terminal.
  • 📝 Créer un fichier modèle personnalisé dans VS Code.
  • 🔧 Ajuster la 'temperature' pour la créativité du modèle.
  • 🛠️ Configurer un prompt système personnalisé.
  • 🔗 Récupérer les paramètres de modèle depuis olama.com.
  • 🚀 Tester le modèle personnalisé pour vérifier sa configuration.
  • 📊 Explorer d'autres paramètres dans la documentation GitHub.

الجدول الزمني

  • 00:00:00 - 00:05:00

    Dans cette vidéo, l'orateur explique comment personnaliser le modèle Lama 3, qui vient d'être mis à jour. Il commence par expliquer comment installer et configurer Lama 3, en dirigeant les spectateurs vers d'autres vidéos pour l'installation sur Mac et Windows, si nécessaire. Ensuite, l'orateur montre comment utiliser le terminal pour télécharger Lama 3 sur un ordinateur, vérifier si l'installation s'est bien passée en exécutant un test rapide, puis commencer à personnaliser le modèle avec des paramètres spécifiques. L'objectif est d'adapter le modèle aux besoins spécifiques de l'utilisateur en modifiant les invites système et certains paramètres clés.

  • 00:05:00 - 00:12:55

    Après avoir expliqué comment personnaliser les invites système et les paramètres de Lama 3, notamment en utilisant le code source VS pour créer un fichier de modèle personnalisé, l'orateur discute comment configurer les paramètres de manière appropriée pour éviter les sorties incorrectes. Il illustre comment obtenir et utiliser les paramètres d'arrêt à partir de la carte de modèle sur le site web. Ensuite, il explique comment tester le modèle personnalisé en posant des questions pour voir s'il fonctionne correctement par rapport au modèle de base. Enfin, l'orateur mentionne l'endroit où les utilisateurs peuvent trouver d'autres paramètres pour personnaliser davantage leur modèle sur la documentation GitHub d'Olama. Il conclut en dirigeant vers d'autres vidéos pour construire un chatbot utilisant Olama et encourage les spectateurs à s'abonner pour plus de contenu.

الخريطة الذهنية

Mind Map

الأسئلة الشائعة

  • Qu'est-ce que Llama 3 ?

    Llama 3 est un modèle d'intelligence artificielle récemment mis à jour, que l'on peut personnaliser pour répondre à des besoins spécifiques.

  • Comment télécharger et installer Llama 3 ?

    Vous pouvez télécharger et installer Llama 3 en utilisant le terminal avec des commandes précises comme 'olama pull' pour récupérer le modèle souhaité.

  • Quels sont les paramètres de base à modifier dans un fichier modèle ?

    Les paramètres de base incluent la température, les paramètres d'arrêt (stop parameters) et le prompt système.

  • Comment vérifier si le modèle Llama 3 est bien installé ?

    Vous pouvez utiliser la commande 'olama list' dans le terminal pour voir les modèles installés.

  • Que fait le paramètre 'temperature' ?

    Le paramètre 'temperature' contrôle la créativité du modèle ; une valeur de 1 signifie que le modèle sera aussi créatif que possible.

  • Comment personnaliser le prompt système d'un modèle Llama 3 ?

    Vous pouvez modifier le prompt système dans le fichier modèle en définissant un texte qui sera utilisé par le modèle lors de son utilisation.

  • Est-il nécessaire d’utiliser VS Code pour créer un fichier modèle ?

    Non, vous pouvez utiliser n'importe quel éditeur de texte, mais VS Code offre l'avantage d'un terminal intégré.

  • Comment tester un modèle personnalisé ?

    Après avoir créé et enregistré un modèle personnalisé, vous pouvez le tester dans le terminal en exécutant 'olama run' suivi du nom de votre modèle.

عرض المزيد من ملخصات الفيديو

احصل على وصول فوري إلى ملخصات فيديو YouTube المجانية المدعومة بالذكاء الاصطناعي!
الترجمات
en
التمرير التلقائي:
  • 00:00:00
    okay let's jump right into customizing
  • 00:00:03
    llama 3 which was just updated about 16
  • 00:00:06
    minutes ago here on the olama website so
  • 00:00:09
    if you haven't already go ahead and
  • 00:00:11
    search for llama 3 on the olama website
  • 00:00:14
    if you don't have Ama already installed
  • 00:00:16
    check out the two videos Down Below in
  • 00:00:18
    the description I have a video on how to
  • 00:00:20
    install and set up AMA on both Macs and
  • 00:00:24
    on windows so go ahead and check that
  • 00:00:26
    out before we get going so this is Alama
  • 00:00:29
    3 that just dropped today we're going to
  • 00:00:32
    show how you can go about customizing
  • 00:00:34
    this model with system prompts and
  • 00:00:37
    different parameters that we have
  • 00:00:38
    available so if you haven't already go
  • 00:00:40
    ahead and navigate here check it out
  • 00:00:42
    read some documentation here we're going
  • 00:00:43
    to be using the Llama 3 8 billion
  • 00:00:46
    parameter model and for those who aren't
  • 00:00:47
    familiar you can get some different
  • 00:00:49
    information around the model here on the
  • 00:00:52
    model card so first thing I am going to
  • 00:00:55
    do is jump over into the terminal so
  • 00:00:58
    that we can start running some ol
  • 00:01:00
    commands to pull down llama 3 onto our
  • 00:01:03
    computers okay I've got my terminal
  • 00:01:05
    pulled up here now the first thing I'm
  • 00:01:07
    going to do is run the olama list
  • 00:01:09
    command which will list all the models
  • 00:01:11
    that I have installed here on AMA as you
  • 00:01:14
    can see I don't have any and install so
  • 00:01:16
    the next thing that I am going to do is
  • 00:01:17
    run AMA pull and I know the model that I
  • 00:01:22
    want is the brand new llama 3 Model here
  • 00:01:26
    and I want the 8 billion parameter model
  • 00:01:29
    so I'm going to go ahead and run this
  • 00:01:32
    take a few minutes to download it and
  • 00:01:34
    then come back after it's done
  • 00:01:35
    completing its download okay I'm back at
  • 00:01:38
    the terminal here and we can see that
  • 00:01:41
    llama 3 has officially downloaded onto
  • 00:01:43
    my laptop now the next command that I'm
  • 00:01:46
    going to run is AMA run and llama
  • 00:01:50
    38b just to make sure everything's
  • 00:01:52
    functioning properly I am going to ask
  • 00:01:55
    it a very simple question write me some
  • 00:01:59
    sample python
  • 00:02:02
    code all right so we can see that it's
  • 00:02:05
    printing out some output here and
  • 00:02:07
    writing me some sample python code so I
  • 00:02:11
    am going to stop that because we know
  • 00:02:12
    it's running I am going to type in
  • 00:02:15
    slashy to exit the model and clear this
  • 00:02:18
    out now the next step that we're going
  • 00:02:20
    to take is we're going to actually
  • 00:02:22
    customize this model the system props
  • 00:02:25
    and some of the parameters to just make
  • 00:02:27
    it more customized to our particular
  • 00:02:30
    needs maybe we want the system prompt to
  • 00:02:32
    say certain things so that's why you
  • 00:02:33
    would go about customizing your own
  • 00:02:36
    version of the model here so I'm going
  • 00:02:38
    to jump into VSS code and we'll walk
  • 00:02:41
    through how to create a custom version
  • 00:02:44
    of llama 3 okay I have vs code open up
  • 00:02:47
    here and I have a file called custom-
  • 00:02:50
    llama 3 holded up here on my laptop you
  • 00:02:53
    can call this file whatever you want but
  • 00:02:55
    you'll have to be sure when we type in
  • 00:02:57
    the command to reference our model file
  • 00:02:59
    file that you're typing the correct name
  • 00:03:01
    of your file here also you could just
  • 00:03:03
    type this in text editor if you want it
  • 00:03:05
    to I'm using vs code cuz it has an
  • 00:03:08
    integrated terminal now we're going to
  • 00:03:10
    briefly walk through each line here of
  • 00:03:15
    what it takes to create your own model
  • 00:03:17
    file now the first thing that we're
  • 00:03:18
    going to do is call from llama 3 colon
  • 00:03:24
    8B now this is our base model that we're
  • 00:03:27
    going to be using here we're going to be
  • 00:03:28
    using the bass llama 3 8 billion
  • 00:03:31
    parameter model as our starting point
  • 00:03:34
    for our custom model so that's why you
  • 00:03:36
    have the from llama 3 colon 8B now one
  • 00:03:39
    thing to call out here the from is not
  • 00:03:41
    case sensitive you could type that in
  • 00:03:43
    all lower cases if you want but the
  • 00:03:45
    convention when creating your custom
  • 00:03:46
    model files is to type in all caps there
  • 00:03:50
    now that's the reason that we have the
  • 00:03:51
    from llama 3 now the next thing that
  • 00:03:54
    we're going to do here is we're going to
  • 00:03:56
    set up some custom parameters now you
  • 00:04:00
    have to set these correctly for each
  • 00:04:02
    model or when you create your custom
  • 00:04:04
    model then your output will be incorrect
  • 00:04:08
    or you may see some weird output if you
  • 00:04:10
    don't type these parameters correctly
  • 00:04:12
    here now I'm just going to go here and
  • 00:04:16
    the first parameter that I am going to
  • 00:04:18
    set is temperature by default the
  • 00:04:20
    underlying temperature is 08 or 07 for
  • 00:04:25
    most of the models so we're going to say
  • 00:04:28
    we're going to set this temperature to
  • 00:04:29
    one so this is going to make the model
  • 00:04:31
    as creative as it can possibly be that's
  • 00:04:34
    what the value of one means when we set
  • 00:04:37
    a temperature parameter and if you're
  • 00:04:39
    used to playing with open Ai and things
  • 00:04:41
    like that and other large language
  • 00:04:43
    models this parameter here shouldn't be
  • 00:04:46
    foreign to you at all now we're going to
  • 00:04:49
    set some other parameters here that I'm
  • 00:04:51
    going to just paste in these are our
  • 00:04:53
    stop parameters now these are set based
  • 00:04:56
    on the model now you can get these
  • 00:04:58
    values here by going to look at the
  • 00:05:00
    model card on olama so let's jump back
  • 00:05:03
    over into the browser so you can
  • 00:05:05
    understand how I came up with these
  • 00:05:06
    parameters here and it's not just
  • 00:05:08
    something I'm making up okay we're back
  • 00:05:10
    at ama.com and we're at the Llama 3
  • 00:05:13
    Model here now I was going to show you
  • 00:05:16
    how I came about getting the stop
  • 00:05:19
    parameters and what you're going to do
  • 00:05:20
    is go down to this section here and go
  • 00:05:22
    to params and then click on that now we
  • 00:05:25
    can see that the stop parameters are
  • 00:05:27
    these sets of values here so you're
  • 00:05:29
    going to take these values one by one
  • 00:05:31
    and add them individually like I did in
  • 00:05:34
    my model file so again each model is
  • 00:05:36
    going to be different so if we're
  • 00:05:38
    customizing minstral you would have to
  • 00:05:40
    make sure that you were actually using
  • 00:05:44
    their stop parameters okay so now you
  • 00:05:46
    know how I came about those stop
  • 00:05:48
    parameters we're going to go back over
  • 00:05:49
    into our model file okay I'm back in my
  • 00:05:51
    model file here in vs code here and now
  • 00:05:54
    hopefully you have a better
  • 00:05:55
    understanding of how I came about
  • 00:05:57
    getting these stop parameters now the
  • 00:05:59
    process is the same for any model that
  • 00:06:01
    is hosted on ama.com so you can follow
  • 00:06:04
    this process for any model there all
  • 00:06:06
    right so the next thing that we're going
  • 00:06:07
    to do is we need to actually have the
  • 00:06:11
    template that this model uses to produce
  • 00:06:15
    its output now I'm going to grab that
  • 00:06:18
    and paste that in so you can see it and
  • 00:06:21
    again we'll jump back over to the model
  • 00:06:23
    file so you'll know exactly how did I
  • 00:06:25
    get these values now I'm going to paste
  • 00:06:28
    that in here and this is the template
  • 00:06:30
    for the Llama 3 Model now let's jump
  • 00:06:34
    back over into the browser so you can
  • 00:06:36
    see what that looks like in the model
  • 00:06:38
    card here now we're back here at the
  • 00:06:39
    Llama 3 Model card here and where we get
  • 00:06:43
    the template information is right here
  • 00:06:45
    where we click template and the model
  • 00:06:47
    card location now we can see the
  • 00:06:50
    template here that we need to use for
  • 00:06:52
    the model to produce output properly for
  • 00:06:55
    us so all I've done is copied this value
  • 00:06:58
    here and then paste that into my model
  • 00:07:01
    file so that's all it is to it again
  • 00:07:02
    this process is the same for any model
  • 00:07:04
    that you want to customize that's on
  • 00:07:06
    ama.com all right so now let's jump back
  • 00:07:08
    over into vs code and set up our last
  • 00:07:12
    parameter okay so we're back here in vs
  • 00:07:14
    code and we're going to set up our last
  • 00:07:16
    parameter which is going to be our
  • 00:07:19
    system parameter which is basically our
  • 00:07:21
    system prompt now I am going to use this
  • 00:07:23
    as my system prop here I'm just saying
  • 00:07:26
    you are a helpful AI assistant named
  • 00:07:28
    llama 3 Droid so so that's all I'm going
  • 00:07:30
    to set you can set it to whatever you
  • 00:07:31
    want you can even make this a little bit
  • 00:07:33
    longer if you want it to but for this
  • 00:07:35
    example here I'm going to keep it pretty
  • 00:07:37
    simple here so that's all it is here for
  • 00:07:40
    us to create our model file so now we
  • 00:07:42
    need to create our new model based on
  • 00:07:46
    this model file here so I'm going to go
  • 00:07:47
    ahead and click save here and the next
  • 00:07:50
    thing I am going to do here is open up
  • 00:07:53
    terminal so I'm going to go here and I
  • 00:07:56
    have my terminal window open now I'm
  • 00:07:59
    going to type in the AMA command to
  • 00:08:01
    create our new model so we're going to
  • 00:08:03
    type in olama and then we're going to
  • 00:08:06
    type create and then I'm just going to
  • 00:08:09
    call this my llama 3 Model you can name
  • 00:08:15
    it whatever you want to now the next
  • 00:08:18
    thing we need to do is set the file flag
  • 00:08:21
    so this is going to be where is my
  • 00:08:23
    custom model located my model is located
  • 00:08:26
    in this current directory here so I can
  • 00:08:28
    just reference
  • 00:08:30
    custom- llama 3 because again that's
  • 00:08:33
    what I named my model here and I can
  • 00:08:37
    simply hit enter so I'm going to hit
  • 00:08:39
    enter here and it ran pretty quickly
  • 00:08:42
    here it's just transferring the model
  • 00:08:44
    reading the model metadata and then
  • 00:08:46
    creating the model layers there so
  • 00:08:47
    that's all it is to create my custom
  • 00:08:49
    model with the create command now let's
  • 00:08:52
    see if our model shows up for us I'm
  • 00:08:55
    going to type in ama list and we can now
  • 00:08:58
    see that I have the Llama 8 billion
  • 00:09:01
    parameter model and then llama latest
  • 00:09:03
    here with the latest tags and the last
  • 00:09:06
    Model that you can see here is my llama
  • 00:09:10
    3- model that we just created a few
  • 00:09:13
    seconds now let's go ahead and test out
  • 00:09:15
    our model to see how it funed I'm going
  • 00:09:17
    to just go ahead and open up a brand new
  • 00:09:19
    terminal window so that you can have a
  • 00:09:21
    better view okay we're back at the
  • 00:09:23
    terminal here and we're going to test
  • 00:09:25
    out our models now the first one we're
  • 00:09:28
    going to test is the Baseline model for
  • 00:09:31
    llama 3 now I'm going to ask it what is
  • 00:09:35
    your
  • 00:09:36
    name and it says it doesn't have a
  • 00:09:38
    personal name there so we don't expect
  • 00:09:40
    it to give us a name let's exit out of
  • 00:09:43
    it so that's the base model now let's
  • 00:09:45
    use our new model here which is the my
  • 00:09:50
    llama 3 model and we're going to type in
  • 00:09:53
    olama run and then my model name and
  • 00:09:55
    just a heads up you don't have to add
  • 00:09:57
    the tag at the end if you don't want to
  • 00:10:00
    you only need to use the tag when you're
  • 00:10:01
    calling out certain parameter counts so
  • 00:10:04
    if I had llama 3 8 billion parameter and
  • 00:10:07
    llama 3 the 70 bilar parameter model I
  • 00:10:11
    would have to do that to call those
  • 00:10:13
    particular Models All right so let's ask
  • 00:10:16
    this model our custom model remember we
  • 00:10:18
    set the system prompt to be llama 3
  • 00:10:20
    Droid
  • 00:10:22
    what is your
  • 00:10:26
    name and you can see hello there my name
  • 00:10:28
    is llama 3 Droid but you can call me
  • 00:10:31
    llama for short so that's just a example
  • 00:10:35
    of how you can customize certain
  • 00:10:37
    parameters such as your system prompt
  • 00:10:39
    your temperature and things of that
  • 00:10:41
    nature with a custom model file now
  • 00:10:44
    let's ask it one more question can
  • 00:10:47
    you write me a simple
  • 00:10:51
    Java
  • 00:10:53
    program and it's writing me a simple
  • 00:10:56
    Java program so it still functions as it
  • 00:10:59
    did before is just we added new
  • 00:11:02
    parameters on top of the base model
  • 00:11:05
    there for you so that's what we did to
  • 00:11:06
    make this model our own now I'm going to
  • 00:11:09
    exit out of here and type in
  • 00:11:13
    slby and the next thing I'm going to do
  • 00:11:16
    is show you where you can check out
  • 00:11:18
    other parameters that we could have set
  • 00:11:20
    in our model file okay I'm here at the
  • 00:11:23
    Alama documentation on GitHub here I'm
  • 00:11:26
    at the olama model file document a now
  • 00:11:30
    we can see the different options that we
  • 00:11:32
    have here when we want to create a model
  • 00:11:35
    file and some of these things we've
  • 00:11:37
    already gone over but like I said I
  • 00:11:39
    wanted to show you where you could get
  • 00:11:41
    other information related to other
  • 00:11:43
    parameters that you could set so I'm
  • 00:11:44
    going to go to the parameter section
  • 00:11:46
    here and click on that and there's a
  • 00:11:47
    host of other parameters that I could
  • 00:11:50
    have set so for example I could have set
  • 00:11:53
    the context window to be larger if that
  • 00:11:56
    model supported a larger token contacts
  • 00:12:00
    there or I could also set the things
  • 00:12:02
    like top K or top P if I wanted to also
  • 00:12:06
    you can see where you had the stop
  • 00:12:08
    parameter and the temperature parameter
  • 00:12:10
    so that's where a lot of these different
  • 00:12:12
    things came from I will put this link
  • 00:12:15
    into the description section also the
  • 00:12:18
    link to the model file will be in a
  • 00:12:21
    GitHub repo if you want to pull that and
  • 00:12:23
    alter it to create your own model files
  • 00:12:25
    like I said you'll have to create a
  • 00:12:27
    model file for any model you want to
  • 00:12:28
    cover customiz onama anyways so you're
  • 00:12:31
    probably wondering okay I've got a
  • 00:12:33
    custom model how can I go about using
  • 00:12:35
    that model in an application well check
  • 00:12:37
    out the two videos that show up on the
  • 00:12:39
    screen on how you could go about
  • 00:12:41
    building your very own olama chatbot and
  • 00:12:44
    if you like this video and the content
  • 00:12:46
    hit like subscribe I try to put content
  • 00:12:49
    out like this on a weekly basis
  • 00:12:51
    appreciate you hanging around to the end
  • 00:12:52
    of the video hope you like it and have a
  • 00:12:54
    great day
الوسوم
  • Llama 3
  • Customisation
  • Modèle IA
  • Terminal
  • VS Code
  • Paramètres modèle
  • Prompt système
  • olama