Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
|
@@ -1,6 +1,5 @@
|
|
| 1 |
# ==============================================================================
|
| 2 |
# Part 1: Core Classes from the Original Script
|
| 3 |
-
# All the necessary helper classes for the RAG system are defined here.
|
| 4 |
# ==============================================================================
|
| 5 |
import os
|
| 6 |
import re
|
|
@@ -13,30 +12,19 @@ import torch
|
|
| 13 |
import faiss
|
| 14 |
from PIL import Image, ImageOps
|
| 15 |
|
| 16 |
-
|
| 17 |
-
from transformers import (
|
| 18 |
-
CLIPVisionModel,
|
| 19 |
-
CLIPImageProcessor,
|
| 20 |
-
AutoTokenizer,
|
| 21 |
-
AutoModel,
|
| 22 |
-
)
|
| 23 |
from sentence_transformers import SentenceTransformer
|
| 24 |
-
|
| 25 |
-
# Google Generative AI
|
| 26 |
import google.generativeai as genai
|
| 27 |
from google.generativeai.types import GenerationConfig
|
| 28 |
-
|
| 29 |
-
# Gradio for Web UI
|
| 30 |
import gradio as gr
|
| 31 |
|
| 32 |
-
|
| 33 |
# --- CONFIGURATION CLASS ---
|
| 34 |
class Config:
|
| 35 |
per_option_ctx: int = 5
|
| 36 |
max_text_len: int = 512
|
| 37 |
docstore_path: str = "indexes/docstore.parquet"
|
| 38 |
-
glot_model_hf: str = "Arshiaizd/Glot500-
|
| 39 |
-
mclip_text_model_hf: str = "Arshiaizd/
|
| 40 |
clip_vision_model: str = "SajjadAyoubi/clip-fa-vision"
|
| 41 |
glot_index_out: str = "indexes/I_glot_text_fa.index"
|
| 42 |
clip_index_out: str = "indexes/I_clip_text_fa.index"
|
|
@@ -105,7 +93,7 @@ class BaseRetriever:
|
|
| 105 |
if os.path.isfile(self.index_path):
|
| 106 |
self.index = faiss.read_index(self.index_path)
|
| 107 |
else:
|
| 108 |
-
raise FileNotFoundError(f"Index file not found at {self.index_path}.
|
| 109 |
def search(self, query_vec: np.ndarray, k: int) -> List[Tuple[int, float]]:
|
| 110 |
D, I = self.index.search(query_vec[None, :].astype(np.float32), k)
|
| 111 |
return list(zip(I[0].tolist(), D[0].tolist()))
|
|
@@ -127,7 +115,7 @@ class TextIndexRetriever(BaseRetriever):
|
|
| 127 |
class VLM_GenAI:
|
| 128 |
def __init__(self, api_key: str, model_name: str, temperature: float = 0.1, max_output_tokens: int = 1024):
|
| 129 |
if not api_key or "YOUR" in api_key:
|
| 130 |
-
raise ValueError("Gemini API Key is missing
|
| 131 |
genai.configure(api_key=api_key)
|
| 132 |
self.model = genai.GenerativeModel(model_name)
|
| 133 |
self.generation_config = GenerationConfig(temperature=temperature, max_output_tokens=max_output_tokens)
|
|
@@ -150,7 +138,6 @@ class RAGSystem:
|
|
| 150 |
# Part 2: Gradio Web Application
|
| 151 |
# ==============================================================================
|
| 152 |
|
| 153 |
-
# --- 1. LOAD MODELS AND INDEXES (This runs only once when the app starts) ---
|
| 154 |
print("Initializing configuration...")
|
| 155 |
cfg = Config()
|
| 156 |
print("Loading RAG system...")
|
|
@@ -160,7 +147,6 @@ api_key = os.environ.get("GEMINI_API_KEY")
|
|
| 160 |
vlm = VLM_GenAI(api_key, model_name="models/gemini-1.5-flash")
|
| 161 |
print("System ready.")
|
| 162 |
|
| 163 |
-
# --- 2. DEFINE THE FUNCTION TO HANDLE USER INPUT ---
|
| 164 |
def run_rag_query(question_text: str, question_image: Optional[Image.Image]) -> Tuple[str, str]:
|
| 165 |
if not question_text.strip():
|
| 166 |
return "Please ask a question.", ""
|
|
@@ -193,47 +179,29 @@ def run_rag_query(question_text: str, question_image: Optional[Image.Image]) ->
|
|
| 193 |
|
| 194 |
# --- 3. CREATE THE GRADIO INTERFACE ---
|
| 195 |
|
| 196 |
-
#
|
| 197 |
custom_css = """
|
| 198 |
-
|
| 199 |
-
|
| 200 |
-
background-image: url('/
|
| 201 |
-
|
| 202 |
-
background-
|
| 203 |
-
|
| 204 |
-
background-
|
| 205 |
-
/* Fix the background image so it doesn't scroll with content */
|
| 206 |
-
background-attachment: fixed;
|
| 207 |
-
/* Center the background image */
|
| 208 |
-
background-position: center;
|
| 209 |
-
color: white; /* Set default text color to white for readability */
|
| 210 |
-
}
|
| 211 |
-
|
| 212 |
-
/* Add a semi-transparent overlay to make text more readable */
|
| 213 |
-
body::before {
|
| 214 |
-
content: "";
|
| 215 |
-
position: absolute;
|
| 216 |
-
top: 0; left: 0; right: 0; bottom: 0;
|
| 217 |
-
background-color: rgba(0, 0, 0, 0.5); /* Black overlay with 50% opacity */
|
| 218 |
-
z-index: -1; /* Place it behind the content */
|
| 219 |
}
|
| 220 |
|
| 221 |
-
/*
|
| 222 |
.gradio-container {
|
| 223 |
-
background: rgba(0, 0, 0, 0.6) !important;
|
| 224 |
-
border-radius: 20px !important;
|
| 225 |
-
border: 1px solid rgba(255, 255, 255, 0.2);
|
| 226 |
}
|
| 227 |
|
| 228 |
-
/*
|
| 229 |
-
|
| 230 |
-
background-color: rgba(255, 255, 255, 0.1) !important;
|
| 231 |
color: white !important;
|
| 232 |
-
border: 1px solid rgba(255, 255, 255, 0.3) !important;
|
| 233 |
}
|
| 234 |
"""
|
| 235 |
|
| 236 |
-
with gr.Blocks(css=custom_css, title="Persian Culinary RAG") as demo:
|
| 237 |
gr.Markdown("# 🍲 Persian Culinary RAG Demo")
|
| 238 |
gr.Markdown("Ask a question about Iranian food, with or without an image, to see the RAG system in action.")
|
| 239 |
|
|
|
|
| 1 |
# ==============================================================================
|
| 2 |
# Part 1: Core Classes from the Original Script
|
|
|
|
| 3 |
# ==============================================================================
|
| 4 |
import os
|
| 5 |
import re
|
|
|
|
| 12 |
import faiss
|
| 13 |
from PIL import Image, ImageOps
|
| 14 |
|
| 15 |
+
from transformers import (CLIPVisionModel, CLIPImageProcessor, AutoTokenizer, AutoModel)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 16 |
from sentence_transformers import SentenceTransformer
|
|
|
|
|
|
|
| 17 |
import google.generativeai as genai
|
| 18 |
from google.generativeai.types import GenerationConfig
|
|
|
|
|
|
|
| 19 |
import gradio as gr
|
| 20 |
|
|
|
|
| 21 |
# --- CONFIGURATION CLASS ---
|
| 22 |
class Config:
|
| 23 |
per_option_ctx: int = 5
|
| 24 |
max_text_len: int = 512
|
| 25 |
docstore_path: str = "indexes/docstore.parquet"
|
| 26 |
+
glot_model_hf: str = "Arshiaizd/Glot500-Fine-Tuned"
|
| 27 |
+
mclip_text_model_hf: str = "Arshiaizd/MCLIP_FA-FineTuned"
|
| 28 |
clip_vision_model: str = "SajjadAyoubi/clip-fa-vision"
|
| 29 |
glot_index_out: str = "indexes/I_glot_text_fa.index"
|
| 30 |
clip_index_out: str = "indexes/I_clip_text_fa.index"
|
|
|
|
| 93 |
if os.path.isfile(self.index_path):
|
| 94 |
self.index = faiss.read_index(self.index_path)
|
| 95 |
else:
|
| 96 |
+
raise FileNotFoundError(f"Index file not found at {self.index_path}.")
|
| 97 |
def search(self, query_vec: np.ndarray, k: int) -> List[Tuple[int, float]]:
|
| 98 |
D, I = self.index.search(query_vec[None, :].astype(np.float32), k)
|
| 99 |
return list(zip(I[0].tolist(), D[0].tolist()))
|
|
|
|
| 115 |
class VLM_GenAI:
|
| 116 |
def __init__(self, api_key: str, model_name: str, temperature: float = 0.1, max_output_tokens: int = 1024):
|
| 117 |
if not api_key or "YOUR" in api_key:
|
| 118 |
+
raise ValueError("Gemini API Key is missing. Add it to Space secrets.")
|
| 119 |
genai.configure(api_key=api_key)
|
| 120 |
self.model = genai.GenerativeModel(model_name)
|
| 121 |
self.generation_config = GenerationConfig(temperature=temperature, max_output_tokens=max_output_tokens)
|
|
|
|
| 138 |
# Part 2: Gradio Web Application
|
| 139 |
# ==============================================================================
|
| 140 |
|
|
|
|
| 141 |
print("Initializing configuration...")
|
| 142 |
cfg = Config()
|
| 143 |
print("Loading RAG system...")
|
|
|
|
| 147 |
vlm = VLM_GenAI(api_key, model_name="models/gemini-1.5-flash")
|
| 148 |
print("System ready.")
|
| 149 |
|
|
|
|
| 150 |
def run_rag_query(question_text: str, question_image: Optional[Image.Image]) -> Tuple[str, str]:
|
| 151 |
if not question_text.strip():
|
| 152 |
return "Please ask a question.", ""
|
|
|
|
| 179 |
|
| 180 |
# --- 3. CREATE THE GRADIO INTERFACE ---
|
| 181 |
|
| 182 |
+
# CORRECTED AND IMPROVED CSS
|
| 183 |
custom_css = """
|
| 184 |
+
/* The #gradio-app selector is more specific and reliable */
|
| 185 |
+
#gradio-app {
|
| 186 |
+
background-image: url('file/background/back.jpg') !important;
|
| 187 |
+
background-size: cover !important;
|
| 188 |
+
background-repeat: no-repeat !important;
|
| 189 |
+
background-attachment: fixed !important;
|
| 190 |
+
background-position: center !important;
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 191 |
}
|
| 192 |
|
| 193 |
+
/* Add a semi-transparent overlay to the main container for better text readability */
|
| 194 |
.gradio-container {
|
| 195 |
+
background-color: rgba(0, 0, 0, 0.6) !important;
|
|
|
|
|
|
|
| 196 |
}
|
| 197 |
|
| 198 |
+
/* Style labels to be white so they are visible */
|
| 199 |
+
.gradio-container .gr-label {
|
|
|
|
| 200 |
color: white !important;
|
|
|
|
| 201 |
}
|
| 202 |
"""
|
| 203 |
|
| 204 |
+
with gr.Blocks(css=custom_css, theme=gr.themes.Soft(), title="Persian Culinary RAG") as demo:
|
| 205 |
gr.Markdown("# 🍲 Persian Culinary RAG Demo")
|
| 206 |
gr.Markdown("Ask a question about Iranian food, with or without an image, to see the RAG system in action.")
|
| 207 |
|