Skip to content

Testing of different IA engines like chat GPT based on the serge repo,

Notifications You must be signed in to change notification settings

Israelsmmx/IA_Engines

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

7 Commits
 
 

Repository files navigation

IA Engines like Alpaca, Vicuna , Wizard .. etc

Testing of different IA engines like chat GPT based on the serge repo,

ALPACA TESTING With Cluster

Git repo: https://github.com/serge-chat/serge

        Kudos for the great team that share this engine to make a test with different IAs 

        <https://github.com/serge-chat>

docker immage

         ghcr.io/serge-chat/serge:latest

Download the repo

        git clone https://github.com/serge-chat/serge.git

Tunnning on the WSL2 configuration

the idea is try to assign and balance the available resources

image

Edit the .wslconfig to assign resources, this will prevent the system comes down or freeze the systems

        C:\Users\<User>\.wslconfig

Reference values on the file, keep resources available for the system management

** Notes: ** if you have aditional unit use a diferent disk for swap



        # Settings apply across all Linux distros running on WSL 2
        [wsl2]
        # Limits VM memory to use no more than 4 GB, this can be set as whole numbers using GB or MB
        memory=40GB 
        # Sets the VM to use two virtual processors
        processors=24
        # Sets amount of swap storage space to 8GB, default is 25% of available RAM
        swap=30GB
        # Sets swapfile path location, default is %USERPROFILE%\AppData\Local\Temp\swap.vhdx
        swapfile=G:\\WSL\\swap\\wsl-swap.vhdx
        # Disable page reporting so WSL retains all allocated memory claimed from Windows and releases none back when free
        pageReporting=false
        # Turn off default connection to bind WSL 2 localhost to Windows localhost
        localhostforwarding=true
        # Disables nested virtualization
        nestedVirtualization=false
        # Turns on output console showing contents of dmesg when opening a WSL 2 distro for debugging
        debugConsole=false

Run the cluster and container with compatibilty of cores and memory

docker-compose --compatibility up

Sample of docker-compose file

        services:
          serge:
            image: ghcr.io/serge-chat/serge:latest
            container_name: serge
            restart: unless-stopped
            ports:
              - 8008:8008
            volumes:
              - weights:/usr/src/app/weights
              - datadb:/data/db/
            deploy:
              resources:
                limits:
                  cpus: '20'
                  memory: 40G
                reservations:
                  cpus: '20'
                  memory: 40G
        volumes:
              weights:
              datadb:

image

image

image

Notes:

some times the engine needs a small kick with CTRL + Enter Again to request reformulate the answare Some questions need more specific question, or may be need a B extended engine trained

image

image

as Serge team mention on the original repo is possible add more engines manually to the container

Additional weights can be added to the serge_weights volume using docker cp:

        docker cp ./my_weight.bin serge:/usr/src/app/weights/

Releases

No releases published

Packages

No packages published