This repository contains a personal experimental model released mainly for research, learning, and proof-of-concept purposes. It is not intended to be a production-ready or fully validated model release. As a very small MoE language model, its output quality, robustness, and practical usefulness may be limited.

Overview

SchoolLM-6M-A3M is a tiny Korean Mixture-of-Experts language model shared primarily as an experimental public artifact. This repository is intended for implementation-level exploration, architecture experimentation, and educational inspection rather than reliable downstream deployment.

Highlights

  • architectures: SchoolMoEForCausalLM
  • tiny sparse MoE with fish-school expert routing
  • total params: about 5.84M
  • active params per token: about 3.00M
  • attention: GQA (8 query heads / 2 KV heads)
  • routed experts: 8
  • shared experts: 2
  • top-k routed experts per token: 2
  • YaRN scaling from 128 to 512

Load

from transformers import AutoConfig, AutoModelForCausalLM, AutoTokenizer

config = AutoConfig.from_pretrained("drlee1/SchoolLM-6M-A3M", trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained("drlee1/SchoolLM-6M-A3M", trust_remote_code=True)
tokenizer = AutoTokenizer.from_pretrained("drlee1/SchoolLM-6M-A3M", trust_remote_code=True, use_fast=False)

Intended Use

  • Personal experimentation
  • Educational inspection of a tiny MoE language model
  • Proof-of-concept validation
  • Research on compact language model architecture

Limitations

  • Not benchmarked thoroughly
  • Not production-ready
  • Output quality may be inconsistent
  • May underperform significantly on real-world tasks
  • Not fully validated for safety, robustness, or broad generalization
Downloads last month
888
Safetensors
Model size
5.84M params
Tensor type
F32
·
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support