| --- |
| license: gemma |
| base_model: |
| - google/gemma-3-1b-it |
| base_model_relation: quantized |
| library_name: executorch |
| pipeline_tag: text-generation |
| inference: false |
| tags: |
| - executorch |
| - webgpu |
| - on-device |
| - browser |
| --- |
| # Gemma 3 1B Instruct for ExecuTorch WebGPU |
|
|
| This repository contains an [ExecuTorch](https://github.com/pytorch/executorch) conversion of |
| [google/gemma-3-1b-it](https://huggingface.co/google/gemma-3-1b-it) for the ExecuTorch **WebGPU** backend, so the model runs entirely in a web browser on |
| the visitor's GPU. It powers the corresponding demo in the [ExecuTorch WebGPU Playground](https://executorch.ai/playground/). |
|
|
| - **Base model:** [google/gemma-3-1b-it](https://huggingface.co/google/gemma-3-1b-it) |
| - **Format:** ExecuTorch `.pte` programs delegated to the WebGPU backend |
| - **Quantization:** 4-bit groupwise weights (q4gsw, group size 64) for all linear layers and the tied token-embedding/output head; fp32 activations and KV cache |
| - **Total size:** 643.6 MB |
|
|
| ## About the original model |
|
|
| The 1B instruction-tuned, text-only model of the Gemma 3 family. |
|
|
| - **Developed by:** Google DeepMind |
| - **Paper:** [Gemma 3 Technical Report](https://arxiv.org/abs/2503.19786) |
| - 1.0B parameters (302M embedding + 698M non-embedding) |
| - Trained on 2T tokens of web documents in over 140 languages, code and mathematics |
| - Context length 32K tokens |
| - 262K-token vocabulary |
|
|
| For training data, evaluation results and full details, see the original model card: [google/gemma-3-1b-it](https://huggingface.co/google/gemma-3-1b-it). |
|
|
| ## Intended use and limitations |
|
|
| Text generation, chat and conversational assistants, summarization, and research and education. Use must comply with the Gemma Terms of Use and the Gemma Prohibited Use Policy. |
|
|
| Quality depends on the training data, which can carry biases and gaps. It does better on clearly framed tasks than open-ended or very complex ones, can miss nuance, sarcasm or figurative language, and is not a knowledge base: it can state incorrect or outdated facts. See the Ethical Considerations and Risks section of the upstream card. |
|
|
| ## What changed in this conversion |
|
|
| - Exported with a Gemma 3 decoder written for ExecuTorch; the sliding-window and global attention layers are exact |
| - All linear layers and the tied embedding/output head quantized to 4-bit groupwise weights (group 64); the embedding/head data is stored once |
| - Context capped at 8,960 tokens (upstream: 32K); at most 512 tokens per call |
| - Quantization changes outputs: answers are fluent but can be worded differently from the original model |
|
|
| ## Inputs and outputs |
|
|
| - `forward(tokens int64 [1, 1..512], input_pos int64 [1]) -> logits float32 [1, 262144]` (last position only) |
| - The KV cache lives inside the program; call it with up to 512 new tokens at a time, then one token per step |
| - Tokenize with the files in `tokenizer/` (the official Gemma tokenizer and chat template); stop on token 1 (`<eos>`) or 106 (`<end_of_turn>`) |
|
|
| ## Files |
|
|
| | File | Size | SHA-256 | |
| |---|---|---| |
| | `gemma3_1b_it_q4gsw.pte` | 610.6 MB | `6056492708430182…` | |
| | `tokenizer/config.json` | 899 B | `19cb5d28c9777827…` | |
| | `tokenizer/generation_config.json` | 215 B | `fd9324becc53c4be…` | |
| | `tokenizer/special_tokens_map.json` | 662 B | `2f7b0adf4fb46977…` | |
| | `tokenizer/tokenizer.json` | 31.8 MB | `4667f2089529e8e7…` | |
| | `tokenizer/tokenizer_config.json` | 1.1 MB | `bfe25c2735e39540…` | |
|
|
| Full checksums are in `manifest.json`. The playground verifies them after download. |
|
|
| ## Running it |
|
|
| These programs were exported with the ExecuTorch WebGPU backend |
| ([backends/webgpu](https://github.com/pytorch/executorch/tree/main/backends/webgpu)) and run in a browser (headless Chromium, WebGPU on SwiftShader) with the WebAssembly runtime build used by the [ExecuTorch WebGPU Playground](https://executorch.ai/playground/), where its prefill logits and greedy output matched the CPU reference token for token. That runtime build is published in |
| [executorch-community/ExecuTorch-WebGPU-Runtime](https://huggingface.co/executorch-community/ExecuTorch-WebGPU-Runtime); the playground is the easiest way to run them. It downloads these files once, caches them in the browser, |
| and runs everything locally. |
|
|
| Requirements: a browser with WebGPU enabled. |
|
|
| ## Provenance |
|
|
| - **Upstream:** `google/gemma-3-1b-it` at revision `dcc83ea841ab6100d6b47a070329e1ba4cf78752` (official) |
| - **Exported with:** executorch 1.5.1+cpu, torch 2.14.1+cpu, torchao 0.18.0+cpu, transformers 5.19.0 (executorch git `3b6068392324`) |
| - **Reproduce:** `python recipes/gemma-3-1b-it.py --out out/gemma-3-1b-it` (recipe in `export/`) |
| - Re-running the recipe produces byte-identical files. |
|
|
| ## Citation |
|
|
| If you use this model, please cite the original work: |
|
|
| ```bibtex |
| @article{gemma_2025, |
| title={Gemma 3}, |
| url={https://goo.gle/Gemma3Report}, |
| publisher={Kaggle}, |
| author={Gemma Team}, |
| year={2025} |
| } |
| ``` |
|
|
| ## License |
|
|
| This is a derivative of the upstream model and is distributed under the same terms |
| (`gemma`); see `LICENSE` and `PROHIBITED_USE_POLICY.md`. |
| Modifications (conversion to the ExecuTorch format and quantization) by Meta Platforms, Inc. |
|
|
| Gemma is provided under and subject to the Gemma Terms of Use found at ai.google.dev/gemma/terms. |
| By using this model you agree to the Gemma Terms of Use, including the use restrictions in its |
| Section 3.2 and the Gemma Prohibited Use Policy. |
|
|
