Add config.json and update README with metadata for HF Inference Endpoints

README.md

@@ -1,77 +1,33 @@
-# MultiTalk Hugging Face Endpoint Handler
+---
+tags:
+- custom
+- inference-endpoints
+- text-to-video
+- multitalk
+library_name: custom
+---
 
-This custom handler enables the MeiGen-AI/MeiGen-MultiTalk model to run on Hugging Face Inference Endpoints.
+# MultiTalk Handler for Hugging Face Inference Endpoints
 
-## Setup Instructions
+This is a custom handler for deploying the MeiGen-AI/MeiGen-MultiTalk model on Hugging Face Inference Endpoints.
 
-1. **Create a new Inference Endpoint** on Hugging Face:
-   - Go to https://huggingface.co/inference-endpoints
-   - Click "New endpoint"
+## Model Description
 
-2. **Configure the endpoint**:
-   - **Model repository**: `ajwestfield/multitalk-handler` (you'll need to upload this handler to your HF account)
-   - **Task**: Custom
-   - **Framework**: Custom
-   - **Instance type**: GPU · A100 · 1x GPU (80 GB)
-
-3. **Advanced Configuration**:
-   - **Container type**: Custom
-   - **Custom image**: `pytorch/pytorch:2.4.1-cuda12.1-cudnn9-runtime`
-   - **Autoscaling**:
-     - Min replicas: 0
-     - Max replicas: 1
-     - Scale to zero after: 300 seconds (5 minutes)
-
-4. **Environment Variables** (add these in Settings):
-   ```
-   PYTORCH_CUDA_ALLOC_CONF=max_split_size_mb:512
-   CUDA_VISIBLE_DEVICES=0
-   ```
-
-## Uploading the Handler
-
-1. Create a new model repository on Hugging Face:
-   ```bash
-   huggingface-cli repo create multitalk-handler --type model
-   ```
-
-2. Upload the handler files:
-   ```bash
-   cd huggingface-endpoint/multitalk-handler
-   git init
-   git add .
-   git commit -m "Add MultiTalk custom handler"
-   git remote add origin https://huggingface.co/ajwestfield/multitalk-handler
-   git push -u origin main
-   ```
+This handler wraps the MeiGen-AI/MeiGen-MultiTalk model for audio-driven multi-person conversational video generation.
 
 ## Usage
 
-Once deployed, you can call the endpoint with:
+This model should be used with Hugging Face Inference Endpoints with the following configuration:
+- GPU: A100 (80GB recommended)
+- Framework: Custom
+- Task: Custom
 
-```python
-import requests
-import json
+## Requirements
 
-API_URL = "https://YOUR-ENDPOINT-URL.endpoints.huggingface.cloud"
-headers = {
-    "Authorization": "Bearer YOUR_HF_TOKEN",
-    "Content-Type": "application/json"
-}
+- PyTorch 2.4.1
+- CUDA 12.1
+- Various dependencies listed in requirements.txt
 
-data = {
-    "inputs": {
-        "prompt": "A person speaking naturally",
-        "image": "base64_encoded_image_optional"
-    },
-    "parameters": {
-        "num_frames": 16,
-        "height": 480,
-        "width": 640,
-        "num_inference_steps": 25
-    }
-}
+## Handler Details
 
-response = requests.post(API_URL, headers=headers, json=data)
-result = response.json()
-```
+The custom handler (`handler.py`) implements the necessary interface for Hugging Face Inference Endpoints to run the MultiTalk model.

config.json

@@ -0,0 +1,6 @@
+{
+  "architectures": ["CustomHandler"],
+  "model_type": "custom",
+  "task": "text-to-video",
+  "custom_handler": true
+}