Experimenting with AI models

[pwgit-create]

Here are suggestions for effective AI models that you can experiment with using Appwish!

The model can be modified by accessing the "src/main/resources/" from the release folder, and modifying the model name in file: ollama_model.props

Appwish's default models

Appwish's current default model for the amd64 version (Linux) iscodestral:22b.
https://ollama.com/library/codestral:22b

Appwish's current default model for the arm version (Raspberry pie) is llama3 by Meta.
https://ollama.com/library/llama3:latest

Model tips

These models are also fun to experiment with:

1. Gemma by Google
   https://ollama.com/library/gemma:latest

This model is both entertaining and easy on your hardware. The quality of this model is really good, and it's kind of amazing considering its size of less than 5 GB. It´ is very cencored though , so keep that in mind if your thinking of generating network apps.

2. dolphin-mixtral by Eric Hartford
   https://ollama.com/library/dolphin-mixtral:8x7b

This is an uncencored model that is good for advanced console apps. The truth is, it's really good. It requires a lot of your hardware, so it's not worth trying unless you run WSL with Nvidia Cuda.

3. dolphin-llama3 Eric Hartford
   https://ollama.com/library/dolphin-llama3:latest

Uncencored model based on the llama3 model from META. This model is a good choice if you want to generate more freely than with the default model. This model's small size makes it suitable for limited hardware resources. It's important to note that this model is limited and will result in more compile errors than the original llama3 model.

4. granite-code:34b by IBM
   https://ollama.com/library/granite-code:34b

A version of the IBM granite-code model with 34 billion parameters. If you possess the required hardware, give the 34b version of this model a shot to achieve outstanding and dependable applications.

[pwgit-create]

Appwish running with Nvidia GPU.