OLMo-2-0425-1B

OLMo 2 1B (≈1 B params, Apache-2.0)

Open-science transformer you can run on a laptop.

• Spec sheet. 16-layer decoder-only stack, 2 048 hidden size, 16 attention heads, 4 096-token window; pre-trained on 4 T tokens of mixed web/code text.

• Punches above its class. OLMo 2 1B posts an average benchmark score of 42.7, edging out peer 1 B models like Gemma-3 and Llama-3.1 on MMLU, GSM-8K, and other tasks.

• Runs light. FP16 weights need ≈ 6 GB VRAM; 8-bit ≈ 3 GB; 4-bit ≈ 1 GB—laptops or single A10G are fine.

• ab-notebook release. AI2 ships the full training logs, intermediate checkpoints, and code so you can audit, rewind, or fine-tune without guesswork.

Why pick it for Norman AI?

Apache license, tiny VRAM needs, and full transparency make OLMo 2 1B a perfect sandbox: spin up edge demos, run interpretability experiments, or ship a “budget” inference tier without changing our Llama-friendly toolchain.

messages = [
    {"role": "system", "content": "You are a helpful AI assistant."},
    {"role": "user", "content": "Can you provide ways to eat combinations of bananas and dragonfruits?"},
    {"role": "assistant",
     "content": "Sure! Here are some ways to eat bananas and dragonfruits together"},
    {"role": "user", "content": "What about solving an 2x + 3 = 7 equation?"},
]

response = await norman.invoke(
    {
        "model_name": "phi-4",
        "inputs": [
            {
                "display_title": "Prompt",
                "data": messages
            }
        ]
    }
)