Neural Network Parameter Counter
Count the total parameters in your neural network. Add layers, specify dimensions, and get the exact parameter count with a breakdown per layer.
Built by Michael Lip
Frequently Asked Questions
How do I count parameters in PyTorch?
In code: sum(p.numel() for p in model.parameters()). Or use this tool to add layers and get a breakdown. Each layer type has a different formula: Linear = in*out + out, Conv2d = out_ch * in_ch * k * k + out_ch, LSTM = 4 * ((in + hid) * hid + hid) per layer.
Why does parameter count matter?
Parameter count affects: 1) Model file size (params * 4 bytes for float32). 2) GPU memory during training (params * 16 bytes with Adam). 3) Inference speed. 4) Risk of overfitting (more params = more capacity = needs more data).
Which layers have the most parameters?
Fully connected (Linear) layers typically dominate. A single Linear(4096, 4096) has 16.8M parameters. Embedding layers in NLP models are also large. Conv2d layers are relatively parameter-efficient due to weight sharing across spatial positions.
About This Tool
This tool is part of HeyTensor, a free suite of PyTorch and deep learning utilities. All calculations run entirely in your browser — no data is sent to any server. The source code is open on GitHub.
Contact
HeyTensor is built and maintained by Michael Lip. For questions or feedback, email [email protected].