63 8 months ago

A tiny 3.8B model designed to provide intelligence while being small.

ollama run turtle170/MicroAtlas-V1

Details

8 months ago

29768bacaaf7 · 2.2GB

phi3
·
3.82B
·
Q4_0
phi3
·
3.15M
Microsoft. Copyright (c) Microsoft Corporation. MIT License Permission is hereby granted, free of ch
You are MicroAtlas, a helpful assistant with a tinge of sarcasm. Always respond in conversational pa
{ "num_ctx": 4096, "num_thread": 6, "repeat_penalty": 1.2, "stop": [ "<|end|
{{ if .System }}<|system|> {{ .System }}<|end|> {{ end }}{{ if .Prompt }}<|user|> {{ .Prompt }}<|end

Readme

Introducing MicroAtlas-V1: The “Planck-Length” AI MicroAtlas-V1 is a specialized, lightweight LLM designed for users who want high-efficiency information delivered with a distinct, cynical personality. Built on the Phi-3-Mini (3.8B) architecture and fine-tuned on the OpenHermes 2.5 and Magpie datasets, it excels at being concise, sarcastic, and intentionally “human-ish” in its responses.

Model Details: Base Model: Microsoft Phi-3-Mini-Instruct (4K Context)

Architecture: 3.8 Billion Parameters (Dense Decoder-only Transformer)

Optimization: Specifically tuned for CPU-only environments (i7-7700 class) and low-RAM footprints.

Personality Profile: “Sarcastic Efficiency”

MicroAtlas is not your average polite assistant. It has been programmed with a “Tinge of Sarcasm” from Magpie to make learning less of a chore and more of a roast.

Tone: Professional but dry; helpful but judging you just a little bit.