A VS Code extension for calculating memory requirements and parameter counts for Large Language Models.
LLM-Napkin helps you understand the memory footprint of transformer-based language models by providing accurate estimates based on model configuration files from Hugging Face.
Key capabilities:
- Load model configurations directly from Hugging Face repositories
- Calculate precise parameter counts with support for modern architectures (GQA, GLU)
- Estimate memory requirements for both inference and training
- Adjust calculations based on quantization level (FP16, INT8, INT4)
- Visualize how batch size and sequence length affect memory usage
- Click the LLM-Napkin icon in the VS Code Activity Bar
- Enter the Hugging Face model path (e.g.,
Qwen/Qwen3-4B) - Optionally enter your Hugging Face API key (required for private models)
- Adjust settings for quantization, sequence length, and batch size
- Click "Calculate Memory Usage" to see detailed results
LLM-Napkin uses precise formulas derived from transformer architecture analysis:
P = VD + L×[(2+r)D² + 3DF]
Where:
- V = Vocabulary size
- D = Hidden dimension
- L = Number of layers
- r = KV-to-Q head ratio (for GQA)
- F = Feed-forward dimension
- Weights memory:
Mₚ = P × b(bytes) - Inference activation memory:
Mₐ = B × T × D × b(bytes) - Training activation memory:
Mₐ = B × L × D × (T + 2D/h) × b(bytes)
Where:
- B = Batch size
- T = Sequence length
- b = Bytes per parameter (based on quantization)
- h = Number of attention heads
- VS Code 1.74.0 or higher
This extension doesn't add any VS Code settings yet.
- Some model architectures might require custom formula adjustments
- Initial release of LLM-Napkin
- Support for loading Hugging Face model configurations
- Parameter count calculation for transformer models
- Memory estimation for different quantization levels
- Support for GQA and GLU architecture variants
Ensure that you've read through the extensions guidelines and follow the best practices for creating your extension.
You can author your README using Visual Studio Code. Here are some useful editor keyboard shortcuts:
- Split the editor (
Cmd+\on macOS orCtrl+\on Windows and Linux). - Toggle preview (
Shift+Cmd+Von macOS orShift+Ctrl+Von Windows and Linux). - Press
Ctrl+Space(Windows, Linux, macOS) to see a list of Markdown snippets.
Enjoy!