diskformer 0.0.1

package/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 Chirag Patnaik
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

package/README.md

@@ -0,0 +1,44 @@
+<h1 align="center">diskformer.js</h1>
+
+**Keep model weights on disk in the browser, and page into WebGPU only what each step needs.**
+A disk tier for in-browser inference — not another engine.
+
+Plain ES modules. Chromium with WebGPU. No server, no build step, no telemetry.
+
+> **Status: pre-release.** The code runs today inside [LocalMind](https://github.com/NakliTechie/LocalMind).
+> v0.1 extracts it here with no behaviour change. `0.0.x` on npm is a name placeholder.
+
+## Install
+
+| How | Command |
+|---|---|
+| npm | `npm install diskformer` *(from v0.1)* |
+| CDN | `import { … } from 'https://cdn.jsdelivr.net/npm/diskformer/+esm'` *(from v0.1)* |
+
+TODO(v0.1): the first call — copy a weights file into OPFS once, open a GPU row cache or expert pool on your
+`GPUDevice`, and read what a step needs.
+
+## Why
+
+Your model does not fit on the GPU, or it fits only by crowding out everything else. Browser engines load every
+weight onto the GPU at load time, including the ones a step never touches: Gemma 4's per-layer embedding table is
+read one row per token, and a mixture-of-experts model uses under a tenth of its weights per token.
+
+diskformer.js keeps those weights in OPFS — the browser's private file system, read through sync access handles in
+workers — and pages into a GPU cache only what each step needs. What it does today, inside LocalMind:
+
+- **Gemma 4 E2B:** the 1.2 GB per-layer embedding table on disk; Chrome's GPU process drops from 4.27 GB to 2.07 GB on
+  the live site, with identical output.
+- **Qwen3.6-35B-A3B (36.9 GB), experimental:** experts paged from OPFS on a 24 GB Mac; greedy output matches
+  llama.cpp's Metal backend 8/8 on 4- and 16-layer cuts of the same GGUF.
+
+## Context
+
+Extracted from LocalMind's SSD-streaming work (2026-10-02 to 2026-10-04): `opfs-reader.js` (OPFS reader pool and
+writer), `moe-expert-stream.js` (GPU slot pool with eviction, pins, prefetch and mapped staging-ring uploads) and
+`ple-opfs.js` (row file and GPU row cache). v0.1 scope is those two layers — store and residency — moved here with no
+behaviour change and verified by LocalMind's existing gates.
+
+## License
+
+MIT · LocalMind: [github.com/NakliTechie/LocalMind](https://github.com/NakliTechie/LocalMind)

package/package.json

@@ -0,0 +1,13 @@
+{
+  "name": "diskformer",
+  "version": "0.0.1",
+  "description": "A disk tier for in-browser inference: keep model weights in OPFS and page into WebGPU only what each step needs. Pre-release name placeholder; v0.1 is being extracted from LocalMind.",
+  "type": "module",
+  "license": "MIT",
+  "author": "Chirag Patnaik",
+  "homepage": "https://github.com/NakliTechie/diskformer.js#readme",
+  "repository": { "type": "git", "url": "git+https://github.com/NakliTechie/diskformer.js.git" },
+  "bugs": { "url": "https://github.com/NakliTechie/diskformer.js/issues" },
+  "keywords": ["webgpu", "opfs", "llm", "inference", "offload", "moe", "browser", "transformers"],
+  "files": ["README.md", "LICENSE"]
+}