Pi-Pewter's coding model for raspberry pi5, and pi4 (coming later in 1b)

0.5b

ollama run Pi-Pewter/Hy-Bea:0.5b

curl http://localhost:11434/api/chat \
  -d '{
    "model": "Pi-Pewter/Hy-Bea:0.5b",
    "messages": [{"role": "user", "content": "Hello!"}]
  }'

from ollama import chat

response = chat(
    model='Pi-Pewter/Hy-Bea:0.5b',
    messages=[{'role': 'user', 'content': 'Hello!'}],
)
print(response.message.content)

import ollama from 'ollama'

const response = await ollama.chat({
  model: 'Pi-Pewter/Hy-Bea:0.5b',
  messages: [{role: 'user', content: 'Hello!'}],
})
console.log(response.message.content)

Models

View all →

Name

1 model

Size / Usage

Context

Input

Hy-Bea:0.5b

398MB · 32K context window · Text · 2 weeks ago

Hy-Bea:0.5b

398MB

32K

Text

Readme

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

Hy-Bea will be gaining tools by November 1 2026, or sooner

Hy-Bea

Hy-Bea is a very small coding assistant based on Qwen/Qwen2.5-Coder-0.5B-Instruct, fine-tuned with LoRA on a compact, hand-curated set of coding instructions, debug explanations, and shell utility examples. It is intentionally 0.5B so it stays fast and usable on low-resource hardware like a Raspberry Pi 5.

Nerd Zone🤓

Model

Base: Qwen/Qwen2.5-Coder-0.5B-Instruct
Fine-tune: LoRA (r=8, alpha=16), 1 epoch, 128-token context, batch size 1 with gradient accumulation 8
Dataset: 126 curated Python debugging, utility, and command-answer examples formatted as user/assistant instruction pairs
Export: Merged LoRA → Q4_K_M GGUF (~380 MB)
Runtime: Ollama / llama.cpp, tested on Raspberry Pi 5 (CPU)

Recommended usage

”`bash