Imported from https://unsloth/Mistral-Small-4-119B-2603-GGUF

Note vision is still a work in progress, see below.

Chat/thinking

Note that unlike most models, mistral-small-4 starts with thinking disabled, and has to be enabled for thinking traces.

$ ollama run frob/mistral-small-4
>>> hello
Hello! How can I assist you today? 😊

>>> /set think high
Set 'think' mode to 'high'.
>>> /clear
Cleared session context
>>> hello
Thinking...
The user greeted me in English. I should respond in English in a friendly 
and helpful manner. Since I don't have any context about their location or 
intent, I'll keep it general and ask how they're doing.
...done thinking.

Hello! How can I assist you today?

Tools

$ ollama launch opencode \
    --model frob/mistral-small-4 \
    -- \
    run 'create a file called README with the contents "hello world"'

> build · frob/mistral-small-4:latest

Task is too short to split. Falling back to single general.
I'll do this in one step.

← Write README
Wrote file successfully.

1:  
```1: hello world
```

$ cat README 
"hello world"

Vision

Vision seems to have artefacts that confuse the vision tower. The test image is a small white dog on a stone step with a collar and a bell. This example and others shows that the model perceives textures (eg, rug, rice grains, spots on the dog) which are not present in the image.

$ ollama run frob/mistral-small-4
>>> describe ./image1.jpg
Added image './image1.jpg'
The image shows a cute, fluffy white kitten sitting on a textured surface, 
possibly a carpet or rug. The kitten has a small, pinkish-red bow tied 
around its neck, giving it a charming and playful appearance. Its eyes are 
wide open, and it seems to be looking directly at the camera with a 
curious expression. The lighting in the image highlights its soft fur and 
makes its features stand out. Overall, the kitten looks adorable and 
innocent.