kylebrodeur/ microfactory-node

Updated 1 month ago

The first fine-tune attempt. Trained on gemma-3-1b-it with rank 16 for 3 epochs on deterministic targets. It parroted the same settings template for every input — it memorized, it did not judge.

ollama run kylebrodeur/microfactory-node

curl http://localhost:11434/api/chat \
  -d '{
    "model": "kylebrodeur/microfactory-node",
    "messages": [{"role": "user", "content": "Hello!"}]
  }'

from ollama import chat

response = chat(
    model='kylebrodeur/microfactory-node',
    messages=[{'role': 'user', 'content': 'Hello!'}],
)
print(response.message.content)

import ollama from 'ollama'

const response = await ollama.chat({
  model: 'kylebrodeur/microfactory-node',
  messages: [{role: 'user', content: 'Hello!'}],
})
console.log(response.message.content)

Models

Name

1 model

Size / Usage

Context

Input

microfactory-node:latest

5.3GB · 128K context window · Text · 1 month ago

microfactory-node:latest

5.3GB

128K

Text

Readme

Microfactory Node: 3D Printer (LoRA v1 — historical)

⚠️ Do not use this model. It is a historical artifact kept for the paper trail.

The first fine-tune attempt. Trained on gemma-3-1b-it with rank 16 for 3 epochs on deterministic targets. It parroted the same settings template for every input — it memorized, it did not judge.

What went wrong

High rank (r=16) + many epochs (3) + deterministic targets = parrot
Base model (gemma-3-1b) was too small for the task

Lessons learned (applied in v2/v3)

Noisy targets force judgment — temperature=0.7 during dataset generation prevents template memorization
Low rank, single epoch — r=4 for 1 epoch: less capacity, less memorization, more attention to the actual job
Larger base model — gemma-4-E4B-it (~4B effective) instead of gemma-3-1b

Use these instead

Recommended: ollama run kylebrodeur/microfactory-node-v3-qat (QAT, highest quality)
Alternative: ollama run kylebrodeur/microfactory-node-v2 (non-QAT, solid)

Links

LoRA source: kylebrodeur/microfactory-node-lora-v1
GGUFs: kylebrodeur/microfactory-node-gguf
Code: github.com/kylebrodeur/microfactory-node
Live node: node.microfactory.space

License

Gemma license (inherited from base model).

# Microfactory Node: 3D Printer (LoRA v1 — historical)

⚠️ **Do not use this model.** It is a historical artifact kept for the paper trail.

The first fine-tune attempt. Trained on gemma-3-1b-it with rank 16 for 3 epochs on deterministic targets. It parroted the same settings template for every input — it memorized, it did not judge.

## What went wrong

- High rank (r=16) + many epochs (3) + deterministic targets = parrot
- Base model (gemma-3-1b) was too small for the task

## Lessons learned (applied in v2/v3)

1. **Noisy targets force judgment** — temperature=0.7 during dataset generation prevents template memorization
2. **Low rank, single epoch** — r=4 for 1 epoch: less capacity, less memorization, more attention to the actual job
3. **Larger base model** — gemma-4-E4B-it (~4B effective) instead of gemma-3-1b

## Use these instead

- **Recommended:** `ollama run kylebrodeur/microfactory-node-v3-qat` (QAT, highest quality)
- **Alternative:** `ollama run kylebrodeur/microfactory-node-v2` (non-QAT, solid)

## Links

- **LoRA source:** [kylebrodeur/microfactory-node-lora-v1](https://huggingface.co/kylebrodeur/microfactory-node-lora-v1)
- **GGUFs:** [kylebrodeur/microfactory-node-gguf](https://huggingface.co/kylebrodeur/microfactory-node-gguf)
- **Code:** [github.com/kylebrodeur/microfactory-node](https://github.com/kylebrodeur/microfactory-node)
- **Live node:** [node.microfactory.space](https://node.microfactory.space)

## License

[Gemma license](https://ai.google.dev/gemma/terms) (inherited from base model).

Paste, drop or click to upload images (.png, .jpeg, .jpg, .svg, .gif)