Fix config

README.md

@@ -1,20 +1,10 @@
-# Run a fine-tuned Gemma 3 270M model in the browser with MediaPipe LLM Inference API
-
-This app demonstrates how to generate emojis from a text input using a fine-tuned Gemma 3 270M model running directly in the browser. For this demo you'll just change one line of code to point to your MediaPipe Task model bundle.
-
-## Run the demo
-1. Download the app files in this directory and include your .task model bundle in the local app folder.
-2. In the worker.js file, update the `modelPath` to point to the .task file.
-3. Open terminal on your computer and navigate (`cd`) to the app folder.
-4. Run `npx serve` to start the local server.
-5. Open the provided `localhost` address in your browser to run the app.
-
-## How it works
-This demo sets up a simple web server to host a frontend where users can enter a text prompt. This starts a generation process in a web worker to avoid blocking the main UI thread. The worker uses a bundled version of the MediaPipe Tasks GenAI package ([@mediapipe/tasks-genai](https://www.npmjs.com/package/@mediapipe/tasks-genai)) to generate a response from the model and send it back to the user.
-
-**Requirements:** Browser with [WebGPU support](https://caniuse.com/webgpu)
-
-## Resources
-* [Notebook: Fine-tune Gemma 3 270M](https://github.com/google-gemini/gemma-cookbook/blob/main/Demos/Emoji-Gemma-on-Web/resources/Fine_tune_Gemma_3_270M_for_emoji_generation.ipynb)
-* [Notebook: Convert Gemma 3 270M for use with MediaPipe](https://github.com/google-gemini/gemma-cookbook/blob/main/Demos/Emoji-Gemma-on-Web/resources/Convert_Gemma_3_270M_to_LiteRT_for_MediaPipe_LLM_Inference_API.ipynb)
-* [MediaPipe LLM Inference Web documentation](https://ai.google.dev/edge/mediapipe/solutions/genai/llm_inference/web_js)
+---
+title: Emoji Generator
+emoji: ✨
+colorFrom: blue
+colorTo: yellow
+sdk: static
+short_description: Translate text to emojis with Gemma 3 270M running on-device
+---
+
+This demo runs a fine-tuned Gemma 3 270M model client-side using MediaPipe LLM Inference API. After the initial download, the model is cached by the browser, allowing for offline inference and instant loading and on future visits.

index.html

@@ -40,8 +40,8 @@
         </div>
     </div>
     <div class="footer">
-        <p>Powered by <a href="https://huggingface.co/google/gemma-3-270m-it">Gemma 3 270M</a> and <a href="https://ai.google.dev/edge/mediapipe/">MediaPipe</a></p>
-        <a href="https://github.com/google-gemini/gemma-cookbook/tree/main/Demos/Emoji-Gemma-on-Web" target="_blank">View on GitHub<svg fill="none" xmlns="http://www.w3.org/2000/svg"
+        <p>Powered by <a href="https://goo.gle/hf-gemma-3-270m-it">Gemma 3 270M</a> and <a href="https://ai.google.dev/edge/mediapipe/">MediaPipe</a></p>
+        <a href="https://goo.gle/emoji-gemma-cookbook" target="_blank">View on GitHub<svg fill="none" xmlns="http://www.w3.org/2000/svg"
                 viewBox="0 0 7 12">
                 <path d="m.06 10.4.94.93L6.33 6 1 .67l-.94.94L4.45 6" fill="#327AC3" />
             </svg></a>

myemoji-gemma-3-270m-it.task

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b1df7985bc5ef447755836c8cfb4ff54d7249ffac27a64231a13b37ed4bc374a
+size 284366083