Update app.py
Browse files
app.py
CHANGED
|
@@ -10,15 +10,10 @@ from openai import OpenAI
|
|
| 10 |
from docx import Document
|
| 11 |
from flask import Flask, request, jsonify, send_from_directory
|
| 12 |
|
| 13 |
-
# ---------- Flask App Setup ----------
|
| 14 |
app = Flask(__name__, static_folder='static', static_url_path='')
|
|
|
|
| 15 |
|
| 16 |
-
# ----------
|
| 17 |
-
OPENROUTER_KEY = os.environ["OPENROUTER_KEY"] # Hugging Face Secret se aaega
|
| 18 |
-
|
| 19 |
-
# ============================================================
|
| 20 |
-
# CUSTOM EMBEDDING MODEL (all-MiniLM-L6-v2 via transformers)
|
| 21 |
-
# ============================================================
|
| 22 |
model_name = "sentence-transformers/all-MiniLM-L6-v2"
|
| 23 |
tokenizer = AutoTokenizer.from_pretrained(model_name)
|
| 24 |
model = AutoModel.from_pretrained(model_name)
|
|
@@ -49,19 +44,15 @@ class CustomEmbedder:
|
|
| 49 |
|
| 50 |
embedder = CustomEmbedder()
|
| 51 |
|
| 52 |
-
#
|
| 53 |
-
# SIMPLE VECTOR STORE (NumPy + Cosine Similarity)
|
| 54 |
-
# ============================================================
|
| 55 |
class SimpleVectorStore:
|
| 56 |
def __init__(self):
|
| 57 |
self.documents = []
|
| 58 |
self.embeddings = []
|
| 59 |
-
|
| 60 |
def add(self, embeddings, documents, ids):
|
| 61 |
for emb, doc in zip(embeddings, documents):
|
| 62 |
self.embeddings.append(np.array(emb))
|
| 63 |
self.documents.append(doc)
|
| 64 |
-
|
| 65 |
def query(self, query_embedding, n_results=3):
|
| 66 |
query_vec = np.array(query_embedding).flatten()
|
| 67 |
if not self.embeddings:
|
|
@@ -75,9 +66,7 @@ class SimpleVectorStore:
|
|
| 75 |
|
| 76 |
store = SimpleVectorStore()
|
| 77 |
|
| 78 |
-
#
|
| 79 |
-
# TEXT EXTRACTION (PDF, DOCX, TXT, CSV – NO IMAGES)
|
| 80 |
-
# ============================================================
|
| 81 |
def extract_text_from_file(file_path):
|
| 82 |
fname = file_path.lower()
|
| 83 |
text = ""
|
|
@@ -125,16 +114,11 @@ def ingest_document(file_path):
|
|
| 125 |
store.add(emb, chunks, [f"doc_{i}" for i in range(len(chunks))])
|
| 126 |
return len(chunks)
|
| 127 |
|
| 128 |
-
# ============================================================
|
| 129 |
-
# RETRIEVAL FUNCTION
|
| 130 |
-
# ============================================================
|
| 131 |
def retrieve(query, top_k=3):
|
| 132 |
q_emb = embedder.encode([query]).tolist()
|
| 133 |
return store.query(q_emb[0], n_results=top_k)
|
| 134 |
|
| 135 |
-
#
|
| 136 |
-
# LLM FUNCTIONS (All via OpenRouter)
|
| 137 |
-
# ============================================================
|
| 138 |
def setup_gemini():
|
| 139 |
return OpenAI(base_url="https://openrouter.ai/api/v1", api_key=OPENROUTER_KEY)
|
| 140 |
|
|
@@ -200,9 +184,7 @@ def local_answer(model, tokenizer, context, question):
|
|
| 200 |
full = tokenizer.decode(outputs[0], skip_special_tokens=True)
|
| 201 |
return full.split("Answer:")[-1].strip()
|
| 202 |
|
| 203 |
-
#
|
| 204 |
-
# FLASK ROUTES
|
| 205 |
-
# ============================================================
|
| 206 |
@app.route('/')
|
| 207 |
def index():
|
| 208 |
return send_from_directory('static', 'index.html')
|
|
@@ -213,39 +195,29 @@ def upload():
|
|
| 213 |
file = request.files['file']
|
| 214 |
file_path = f"/tmp/{file.filename}"
|
| 215 |
file.save(file_path)
|
| 216 |
-
|
| 217 |
-
|
| 218 |
-
return jsonify({"message": f"✅ {num_chunks} chunks processed."})
|
| 219 |
-
else:
|
| 220 |
-
return jsonify({"message": "❌ No text could be extracted."})
|
| 221 |
except Exception as e:
|
| 222 |
return jsonify({"message": f"❌ Error: {str(e)}"})
|
| 223 |
|
| 224 |
@app.route('/chat', methods=['POST'])
|
| 225 |
def chat():
|
| 226 |
data = request.get_json()
|
| 227 |
-
query = data.get('query',
|
| 228 |
-
|
| 229 |
try:
|
| 230 |
chunks = retrieve(query, top_k=3)
|
| 231 |
ctx = "\n\n".join(chunks)
|
| 232 |
-
if
|
| 233 |
-
|
| 234 |
-
|
| 235 |
-
|
| 236 |
-
client = setup_openrouter()
|
| 237 |
-
answer = openrouter_answer(client, ctx, query)
|
| 238 |
-
elif model_choice == "local":
|
| 239 |
-
m, t = setup_local_model()
|
| 240 |
-
answer = local_answer(m, t, ctx, query)
|
| 241 |
else:
|
| 242 |
-
|
| 243 |
-
|
|
|
|
| 244 |
except Exception as e:
|
| 245 |
return jsonify({"answer": f"Error: {str(e)}", "sources": []})
|
| 246 |
|
| 247 |
-
# ============================================================
|
| 248 |
-
# APP RUN
|
| 249 |
-
# ============================================================
|
| 250 |
if __name__ == "__main__":
|
| 251 |
app.run(host="0.0.0.0", port=7860)
|
|
|
|
| 10 |
from docx import Document
|
| 11 |
from flask import Flask, request, jsonify, send_from_directory
|
| 12 |
|
|
|
|
| 13 |
app = Flask(__name__, static_folder='static', static_url_path='')
|
| 14 |
+
OPENROUTER_KEY = os.environ["OPENROUTER_KEY"]
|
| 15 |
|
| 16 |
+
# ---------- Embedding Model ----------
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 17 |
model_name = "sentence-transformers/all-MiniLM-L6-v2"
|
| 18 |
tokenizer = AutoTokenizer.from_pretrained(model_name)
|
| 19 |
model = AutoModel.from_pretrained(model_name)
|
|
|
|
| 44 |
|
| 45 |
embedder = CustomEmbedder()
|
| 46 |
|
| 47 |
+
# ---------- Vector Store ----------
|
|
|
|
|
|
|
| 48 |
class SimpleVectorStore:
|
| 49 |
def __init__(self):
|
| 50 |
self.documents = []
|
| 51 |
self.embeddings = []
|
|
|
|
| 52 |
def add(self, embeddings, documents, ids):
|
| 53 |
for emb, doc in zip(embeddings, documents):
|
| 54 |
self.embeddings.append(np.array(emb))
|
| 55 |
self.documents.append(doc)
|
|
|
|
| 56 |
def query(self, query_embedding, n_results=3):
|
| 57 |
query_vec = np.array(query_embedding).flatten()
|
| 58 |
if not self.embeddings:
|
|
|
|
| 66 |
|
| 67 |
store = SimpleVectorStore()
|
| 68 |
|
| 69 |
+
# ---------- Text Extraction (NO IMAGES) ----------
|
|
|
|
|
|
|
| 70 |
def extract_text_from_file(file_path):
|
| 71 |
fname = file_path.lower()
|
| 72 |
text = ""
|
|
|
|
| 114 |
store.add(emb, chunks, [f"doc_{i}" for i in range(len(chunks))])
|
| 115 |
return len(chunks)
|
| 116 |
|
|
|
|
|
|
|
|
|
|
| 117 |
def retrieve(query, top_k=3):
|
| 118 |
q_emb = embedder.encode([query]).tolist()
|
| 119 |
return store.query(q_emb[0], n_results=top_k)
|
| 120 |
|
| 121 |
+
# ---------- LLM Functions ----------
|
|
|
|
|
|
|
| 122 |
def setup_gemini():
|
| 123 |
return OpenAI(base_url="https://openrouter.ai/api/v1", api_key=OPENROUTER_KEY)
|
| 124 |
|
|
|
|
| 184 |
full = tokenizer.decode(outputs[0], skip_special_tokens=True)
|
| 185 |
return full.split("Answer:")[-1].strip()
|
| 186 |
|
| 187 |
+
# ---------- Routes ----------
|
|
|
|
|
|
|
| 188 |
@app.route('/')
|
| 189 |
def index():
|
| 190 |
return send_from_directory('static', 'index.html')
|
|
|
|
| 195 |
file = request.files['file']
|
| 196 |
file_path = f"/tmp/{file.filename}"
|
| 197 |
file.save(file_path)
|
| 198 |
+
num = ingest_document(file_path)
|
| 199 |
+
return jsonify({"message": f"✅ {num} chunks processed." if num else "❌ No text could be extracted."})
|
|
|
|
|
|
|
|
|
|
| 200 |
except Exception as e:
|
| 201 |
return jsonify({"message": f"❌ Error: {str(e)}"})
|
| 202 |
|
| 203 |
@app.route('/chat', methods=['POST'])
|
| 204 |
def chat():
|
| 205 |
data = request.get_json()
|
| 206 |
+
query = data.get('query','')
|
| 207 |
+
choice = data.get('model_choice','local')
|
| 208 |
try:
|
| 209 |
chunks = retrieve(query, top_k=3)
|
| 210 |
ctx = "\n\n".join(chunks)
|
| 211 |
+
if choice == "gemini":
|
| 212 |
+
ans = gemini_answer(setup_gemini(), ctx, query)
|
| 213 |
+
elif choice == "openrouter":
|
| 214 |
+
ans = openrouter_answer(setup_openrouter(), ctx, query)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 215 |
else:
|
| 216 |
+
m, t = setup_local_model()
|
| 217 |
+
ans = local_answer(m, t, ctx, query)
|
| 218 |
+
return jsonify({"answer": ans, "sources": chunks})
|
| 219 |
except Exception as e:
|
| 220 |
return jsonify({"answer": f"Error: {str(e)}", "sources": []})
|
| 221 |
|
|
|
|
|
|
|
|
|
|
| 222 |
if __name__ == "__main__":
|
| 223 |
app.run(host="0.0.0.0", port=7860)
|