fxed some bugs

README.md

@@ -10,6 +10,13 @@ app_file: app.py
 pinned: true
 license: mit
 short_description: Turn anyone into a living RPG character
+full_width: true
+tags:
+  - zerogpu
+  - gradio-server
+  - image-to-text
+  - rpg
+  - character-generation
 ---
 
 <!--
@@ -52,12 +59,17 @@ NPCverse uses [`openbmb/MiniCPM-V-4.6`](https://huggingface.co/openbmb/MiniCPM-V
 git clone https://huggingface.co/spaces/build-small-hackathon/NPCverse
 cd NPCverse
 python -m venv .venv
-.venv\Scripts\activate
+.venv\Scripts\activate      # Windows
+# source .venv/bin/activate  # macOS/Linux
 pip install -r requirements.txt
+
+# Optional: create a .env file if you hit Hub rate limits locally
+# echo HF_TOKEN=hf_... > .env
+
 python app.py
 ```
 
-Open `http://localhost:7860` after the server starts.
+Open `http://localhost:7860` after the server starts. The model (`openbmb/MiniCPM-V-4.6`, ~3 GB) will be downloaded from the Hub on first run.
 
 ## Environment Variables
 

app.py

@@ -8,7 +8,6 @@ import json
 import os
 from typing import Any
 
-import spaces
 from dotenv import load_dotenv
 from fastapi.responses import HTMLResponse, JSONResponse
 from gradio import Server
@@ -46,7 +45,6 @@ async def health() -> JSONResponse:
 
 
 @app.api(name="summon_npc")
-@spaces.GPU
 def summon_npc(image_path: Any) -> dict:
     """Analyze an uploaded image and summon a complete NPC profile."""
     try:
@@ -59,7 +57,6 @@ def summon_npc(image_path: Any) -> dict:
 
 
 @app.api(name="chat_with_npc")
-@spaces.GPU
 def chat_with_npc_api(
     npc_json: str,
     history_json: str,

index.html

@@ -709,7 +709,7 @@
 
     function getClient() {
       if (!clientPromise) {
-        clientPromise = Client.connect(`${window.location.origin}/gradio`);
+        clientPromise = Client.connect(window.location.origin);
       }
       return clientPromise;
     }

model_engine.py

@@ -22,7 +22,7 @@ model = AutoModelForImageTextToText.from_pretrained(
     MODEL_ID,
     trust_remote_code=True,
     dtype="auto",
-).eval().to("cuda")
+).eval()
 
 FRIENDSHIP_THRESHOLDS = [10, 20, 35, 55]
 SECRET_THRESHOLDS = [8, 18, 35]
@@ -122,6 +122,10 @@ def _generate_from_messages(
     downsample_mode: str = "16x",
 ) -> str:
     """Generate text with MiniCPM-V-4.6 from chat-template messages."""
+    device = "cuda" if torch.cuda.is_available() else "cpu"
+    if model.device.type != device:
+        model.to(device)
+
     inputs = processor.apply_chat_template(
         messages,
         tokenize=True,

requirements.txt

@@ -1,4 +1,5 @@
-gradio @ https://gradio-builds.s3.amazonaws.com/72f78a255c31e038b3c8074619cfd304ac94d54b/gradio-6.16.0-py3-none-any.whl
+gradio==6.16.0
+fastapi>=0.115.0
 transformers>=5.7.0
 torch
 torchvision