Llama 3.2 3B

What this model is

The architecture is dense. Every parameter is part of the active path, so memory and speed scale with the full model rather than a smaller expert subset.

Memory and quantization

Decimal GB, rounded from the Ollama tag page. Runtime memory still needs KV cache and overhead.

Estimate formula: published weight files + 0.9395 GB per 8K tokens × context / 8K + 1.5 GB runtime overhead. Sizes are decimal GB, not GiB. At 8K context, the smallest listed build is about 4.4 GB before OS and application reserve.

Conservative FP16 estimate from the standard Llama 3.2 GQA configuration; not a runtime measurement.

KV method source: https://ollama.com/library/llama3.2

Speed heuristic

Speed varies substantially between CPU and GPU; no single measured value is reported

Active path: 3.2B parameters. This is a relative speed signal only; it is not a tokens/sec benchmark.

Runtimes

• Ollama
• llama.cpp

ollama run llama3.2:3b

License restrictions

Llama 3.2 Community License and acceptable-use policy apply. Commercial deployments above the license's monthly-active-user threshold need a separate license.

This is a plain-language summary, not legal advice. Read the linked license before commercial deployment or redistribution.

Known issues

• Limited capability on complex reasoning tasks

Sources

• https://ollama.com/library/llama3.2
• https://huggingface.co/meta-llama/Llama-3.2-3B-Instruct

This page is an independent compatibility guide, not official documentation. Verify the primary source before downloading or deploying a model.