G0-nano-instruct

A 62M-parameter GPT pretrained and instruction-tuned entirely on a single 8GB-RAM NVIDIA Jetson device, without cloud infrastructure or multi-GPU setups. Chat-oriented checkpoint for short instruction-following interactions.

Overview

G0-nano-instruct is the instruction-tuned version of G0 Nano. It is a 62.1M-parameter decoder-only causal language model trained from scratch under a single 8GB-RAM NVIDIA Jetson.

The project focuses on making a complete model training workflow — tokenizer, pretraining, fine-tuning infrastructure and export — work on modest hardware.

Model variants

The raw pretrained version of the same model is available as azerdsq/g0-nano-base.

What this version adds

Compared with azerdsq/g0-nano-base, this checkpoint adds supervised instruction fine-tuning and a chat format for short, single-turn interactions.

The model is intentionally small: its purpose is to demonstrate a complete from-scratch training workflow on modest hardware, not to compete with much larger language models on broad factual knowledge.

Architecture

Llama-style decoder-only Transformer:

• Parameters: 62.1M
• Layers: 12
• Hidden size: 640
• Attention: Grouped-Query Attention, 10 query heads / 2 key-value heads, head dimension 64
• Position encoding: RoPE, θ=10000
• Feed-forward network: SwiGLU, hidden dimension 1728
• Normalization: RMSNorm
• Context length: 1024 tokens; 512 tokens during supervised fine-tuning
• Vocabulary: 16,388 tokens: 16,384 SentencePiece tokens plus 4 chat tokens

Training

• Pretraining data: approximately 1.5B tokens of English web and book text
• Sources: FineWeb-Edu, BookCorpus, OpenWebText, PG-19 and WikiHow
• Instruction tuning: cleaned Alpaca instruction/response data
• Conversation format: short, single-turn instruction/response interactions
• Training hardware: a single NVIDIA Jetson with 8GB of unified memory

Usage

ollama run azerdsq/g0-nano-instruct

The chat format is built into the model for simple instruction-following use.

Weights are also available on Hugging Face.

Benchmarks

Limitations

• 62M parameters impose a hard limit on factual knowledge; expect fluent but frequently incorrect answers on knowledge-intensive prompts.
• Maximum context length is 1024 tokens.
• English-only training data.
• Designed for short, single-turn exchanges; it is not a long-context conversational model.
• Single-sequence generation only; padded batched inference is not supported by the custom model code.

License

Apache 2.0. This release contains model weights and the code required to load them; it does not include the training data or private training infrastructure.