Pi-Pewter's first model for the raspberry pi5, and pi4

0.5b 1.1b

ollama run Pi-Pewter/kat:1.1b

curl http://localhost:11434/api/chat \
  -d '{
    "model": "Pi-Pewter/kat:1.1b",
    "messages": [{"role": "user", "content": "Hello!"}]
  }'

from ollama import chat

response = chat(
    model='Pi-Pewter/kat:1.1b',
    messages=[{'role': 'user', 'content': 'Hello!'}],
)
print(response.message.content)

import ollama from 'ollama'

const response = await ollama.chat({
  model: 'Pi-Pewter/kat:1.1b',
  messages: [{role: 'user', content: 'Hello!'}],
})
console.log(response.message.content)

Details

Updated 3 weeks ago

3 weeks ago

9331f52c4c1b · 668MB

model

archllama

parameters1.1B

quantizationQ4_K_M

668MB

template

53B

system

You are Kat, a helpful assistant fine-tuned on the user's data.

63B

params

{ "num_ctx": 512, "stop": [ "<|system|>", "</s>", "<|user|>",

130B

Readme

NOTE: Kat is in beta, it may fail, or not load. Our apologies if that happens.

Kat

Kat is a small, fast conversational model based on Qwen2.5-0.5B-Instruct, fine-tuned with LoRA on the Pi 5. The fine-tune dataset was derived from TinyShakespeare and split into short text chunks, so Kat tends toward a Shakespearean/classical tone while staying general-purpose as a helpful assistant.

Nerd Zone🤓

Model

Base: Qwen/Qwen2.5-0.5B-Instruct
Training: 1 epoch, LoRA (r=8, alpha=16), sequence length 128, batch size 1 with grad accum 8
Export: Merged LoRA → Q4_K_M GGUF (~397 MB)
Runtime: Ollama, llama.cpp (CPU), ~27 tok/s generation on Raspberry Pi 5

Usage

`bash