Skip to content
| Marketplace
Sign in
Visual Studio Code>Data Science>WeightlessNew to Visual Studio Code? Get it now.
Weightless

Weightless

kipngeno koech

|
108 installs
| (0) | Free
Inspect open-weight models in VS Code. Explore module graphs, transformer schematics, tensor shapes, parameter counts and model configuration. Header-only reads support local safetensors and GGUF checkpoints and the Hugging Face Hub.
Installation
Launch VS Code Quick Open (Ctrl+P), paste the following command, and press enter.
Copied to clipboard
More Info

Weightless

Inspect open-weight models in VS Code.

Open a .safetensors or .gguf file, browse the model gallery, or load a Hugging Face model by ID. Weightless reads tensor headers and available configuration without downloading the full weights.

VS Code Marketplace · Open VSX · Browser viewer · Hugging Face Space

ResNet-18 module hierarchy, graph and selected convolution in Weightless

Inspect a model

  • Architecture workspace. Search the module tree, select a module in the graph, drill into its children, and inspect tensor shapes, parameter counts, storage and dtypes. Selection stays synchronized across views.
  • Paper diagrams. Recognized transformer stacks offer a schematic with attention, feed-forward, normalization and residual stages. Other architectures retain the module graph and tensor inventory.
  • Input, hidden layers and output. Recognized model families show their stages, dimensions and parameter distribution. Parameter and memory controls update interactively.
  • Tensors and model data. Expand the scoped, searchable tensor table when needed. Open Model data for loaded configuration, metadata and available training, generation, adapter or tokenizer sidecars.
  • Training exploration. Plan trainable and frozen parameter groups and explore a one-weight learning example with an adjustable learning rate.

The interface follows the active VS Code theme. Tensor details and configuration stay collapsed until opened.

Formats and sources

Single-file and sharded .safetensors checkpoints are supported, including adjacent model.safetensors.index.json files. .gguf support includes quantized dtypes, tensor storage sizes and architecture metadata.

Run Weightless: Browse Open Models for the gallery or Weightless: Open Model from Hugging Face Hub to enter a repository ID. Hub loading uses HTTP range requests and supports HF_TOKEN for gated repositories.

The module tree and graph describe checkpoint structure. Paper diagrams use recognized architecture conventions. Recorded execution, live activations and actual per-parameter training flags require information beyond a weight header. Trainable/frozen controls represent the selected training plan.

Develop

git clone https://github.com/kkipngenokoech/weightless.git
cd weightless
npm test

Press F5 in VS Code to launch an Extension Development Host. Open a model file or run Weightless: Browse Open Models.

Run the DOM interaction checks with a temporary test dependency.

npm install --prefix /tmp/weightless-test --no-save --ignore-scripts jsdom@26.1.0
NODE_PATH=/tmp/weightless-test/node_modules node test/explain.js
NODE_PATH=/tmp/weightless-test/node_modules node test/workspace.js

Package the extension with npx @vscode/vsce package.

License

MIT. See LICENSE.

  • Contact us
  • Jobs
  • Privacy
  • Manage cookies
  • Terms of use
  • Trademarks
  • Your Privacy Choices
  • Consumer Health Privacy
© 2026 Microsoft