| --- |
| tags: |
| - drowse |
| - webgpu |
| - browser-models |
| - download-index |
| --- |
| |
| # drowse browser downloads |
|
|
| Eight browser-model bundles and four existing instrument repositories, published under the `logitsml` account. All listed revisions are public and ungated; artifact sizes and hashes were verified against the uploaded Git/LFS objects, and anonymous ranged file downloads were checked. |
|
|
| | Model | Status | Model + listed tools | Optional instruments | |
| | --- | --- | ---: | --- | |
| | [Gemma 3 1B Instruct](https://huggingface.co/logitsml/drowse-web-gemma3-1b) | chat mirror | 0.84 GB | [J-Lens + SAE](https://huggingface.co/logitsml/drowse-web-gemma3-1b-instruments) | |
| | [Gemma 3 4B Instruct](https://huggingface.co/logitsml/drowse-web-gemma3-4b) | chat mirror | 2.87 GB | [J-Lens + SAE](https://huggingface.co/logitsml/drowse-web-gemma3-4b-instruments) | |
| | [Qwen3 1.7B](https://huggingface.co/logitsml/drowse-web-qwen3-1.7b) | chat mirror | 1.19 GB | [J-Lens](https://huggingface.co/logitsml/drowse-web-qwen3-1.7b-instruments) | |
| | [Qwen3 4B](https://huggingface.co/logitsml/drowse-web-qwen3-4b) | chat mirror | 2.59 GB | [J-Lens](https://huggingface.co/logitsml/drowse-web-qwen3-4b-instruments) | |
| | [GPT-2 small](https://huggingface.co/logitsml/drowse-web-gpt2-base-candidate) | base candidate | 0.67 GB | Not packaged | |
| | [Pythia 70M deduped](https://huggingface.co/logitsml/drowse-web-pythia-70m-base-candidate) | base candidate | 0.29 GB | Not packaged | |
| | [Gemma 3 1B PT](https://huggingface.co/logitsml/drowse-web-gemma3-1b-pt-candidate) | base candidate | 0.63 GB | Not packaged | |
| | [Qwen 3.5 2B Base](https://huggingface.co/logitsml/drowse-web-qwen35-2b-base-candidate) | base candidate | 1.11 GB | Not packaged | |
|
|
| Total: **10.19 GB** for one copy of the listed files. Users download only the model and tools they choose. Sizes include documentation and repository-specific runtime files, but not development caches or older revisions. |
|
|
| ## Download |
|
|
| Open a model link, then Files and versions to download individual files. To download a complete bundle: |
|
|
| ```sh |
| hf download logitsml/drowse-web-gemma3-1b --local-dir ./drowse-gemma3-1b |
| ``` |
|
|
| [downloads.json](downloads.json) lists every exact repository revision, file size and SHA-256. It is an unsigned download index, **not the app's signed install catalog**. |
|
|
| ## Compatibility status |
|
|
| - **Chat mirrors:** copied from previously assembled browser bundles with original artifact bytes and provenance preserved. Their older runtime identities are retained; this publication is not a compatibility certification for the current app. |
| - **Base candidates:** local converted browser builds, intentionally labeled as candidates. GPT-2 and Pythia have narrow checkpoint-comparison evidence; Gemma PT still has a matched-weight numerical discrepancy, and Qwen 3.5 still needs matched-quantization validation. None is promoted to an install-ready release here. |
| - **App installation remains pending:** no signed `catalog.json` has been published for these revisions. Uploading these files does not by itself enable the app's download button. |
|
|
| The public repositories and download pages use the **drowse** name. Historical technical identifiers, binary symbols, filenames and upstream attributions are intentionally preserved to avoid changing artifact hashes or claiming a rebuild. Component licenses and model-specific restrictions are documented in each repository. |
|
|
