Update app.py
Browse files
app.py
CHANGED
|
@@ -13,10 +13,9 @@ chroma_client = chromadb.PersistentClient(path=CHROMA_DATA_PATH)
|
|
| 13 |
ef = embedding_functions.DefaultEmbeddingFunction()
|
| 14 |
collection = chroma_client.get_or_create_collection(name="yousef_vault", embedding_function=ef)
|
| 15 |
|
| 16 |
-
# النموذج الأساسي الافتراضي (يمكنك تغييره من هنا)
|
| 17 |
DEFAULT_MODEL = "huihui-ai/Qwen2.5-72B-Instruct-abliterated"
|
| 18 |
|
| 19 |
-
# --- 2. دالة استخراج النصوص
|
| 20 |
def process_document(file):
|
| 21 |
if file is None: return "⚠️ لم يتم اختيار ملف."
|
| 22 |
text = ""
|
|
@@ -34,95 +33,67 @@ def process_document(file):
|
|
| 34 |
text = f.read()
|
| 35 |
|
| 36 |
if not text.strip():
|
| 37 |
-
return "❌ تعذر استخراج نص.
|
| 38 |
|
| 39 |
-
# تقسيم النص لقطع وحقنها في الذاكرة
|
| 40 |
chunks = [text[i:i+2000] for i in range(0, len(text), 2000)]
|
| 41 |
ids = [str(uuid.uuid4()) for _ in chunks]
|
| 42 |
collection.add(documents=chunks, ids=ids)
|
| 43 |
|
| 44 |
-
return f"✅ تمت الأرشفة: {len(chunks)} شظية معرفية
|
| 45 |
except Exception as e:
|
| 46 |
-
return f"❌ خطأ
|
| 47 |
|
| 48 |
-
# --- 3. دالة الحوار
|
| 49 |
-
def predict(message, history, system_prompt, temperature, custom_model
|
| 50 |
-
# ا
|
| 51 |
-
|
|
|
|
| 52 |
|
| 53 |
if not api_key:
|
| 54 |
-
yield "⚠️ يرجى
|
| 55 |
return
|
| 56 |
|
| 57 |
-
client = OpenAI(
|
| 58 |
-
base_url="https://router.huggingface.co/v1",
|
| 59 |
-
api_key=api_key
|
| 60 |
-
)
|
| 61 |
-
|
| 62 |
-
# تحديد النموذج المستخدم
|
| 63 |
active_model = custom_model.strip() if custom_model and custom_model.strip() else DEFAULT_MODEL
|
| 64 |
|
| 65 |
-
# استرجاع سياق
|
| 66 |
-
results = collection.query(query_texts=[message], n_results=
|
| 67 |
context = "\n".join(results['documents'][0]) if results['documents'] else ""
|
| 68 |
|
| 69 |
-
# بناء قائمة الرسائل
|
| 70 |
messages = [{"role": "system", "content": f"{system_prompt}\n\n[CONTEXT]:\n{context}"}]
|
| 71 |
-
|
| 72 |
for msg in history:
|
| 73 |
-
messages.append({"role": msg[
|
| 74 |
-
|
| 75 |
messages.append({"role": "user", "content": message})
|
| 76 |
|
| 77 |
try:
|
| 78 |
response = client.chat.completions.create(
|
| 79 |
model=active_model,
|
| 80 |
messages=messages,
|
| 81 |
-
temperature=temperature,
|
| 82 |
stream=True
|
| 83 |
)
|
| 84 |
|
| 85 |
partial_message = ""
|
| 86 |
for chunk in response:
|
| 87 |
-
|
| 88 |
-
|
| 89 |
-
|
| 90 |
-
if content:
|
| 91 |
-
partial_message += content
|
| 92 |
-
yield partial_message
|
| 93 |
-
else:
|
| 94 |
-
# تجاهل الإطارات الفارغة التي تظهر في نهاية البث
|
| 95 |
-
continue
|
| 96 |
-
|
| 97 |
except Exception as e:
|
| 98 |
-
|
| 99 |
-
# إذا كان هناك نص تم توليده بالفعل، اعرضه ولا تظهر رسالة الخطأ
|
| 100 |
-
yield partial_message
|
| 101 |
-
else:
|
| 102 |
-
yield f"❌ فشل الاتصال بالنموذج ({active_model}): {str(e)}"
|
| 103 |
|
| 104 |
-
# --- 4. بناء الواجهة (
|
| 105 |
-
with gr.Blocks(fill_height=True
|
| 106 |
with gr.Sidebar():
|
| 107 |
gr.Markdown("# 🧬 دهليز يوسف")
|
| 108 |
-
gr.LoginButton("Sign in")
|
| 109 |
|
| 110 |
-
with gr.Accordion("⚙️ الضبط
|
| 111 |
-
model_input = gr.Textbox(
|
| 112 |
-
|
| 113 |
-
|
| 114 |
-
value="",
|
| 115 |
-
lines=1
|
| 116 |
-
)
|
| 117 |
-
system_input = gr.Textbox(
|
| 118 |
-
value="أنت رفيق حكيم يغوص في التيه مع المستخدم. أجب بصدق مستعيناً بالذاكرة.",
|
| 119 |
-
label="البرومبت النظامي"
|
| 120 |
-
)
|
| 121 |
-
temp_slider = gr.Slider(0.1, 1.5, 0.8, label="درجة التشظي (Temperature)")
|
| 122 |
|
| 123 |
gr.Markdown("---")
|
| 124 |
-
gr.
|
| 125 |
-
file_box = gr.File(label="ارفع مسوداتك هنا", file_types=[".pdf", ".docx", ".txt"])
|
| 126 |
status_msg = gr.Markdown("*الذاكرة بانتظار الملفات...*")
|
| 127 |
|
| 128 |
file_box.change(process_document, inputs=file_box, outputs=status_msg)
|
|
@@ -130,17 +101,17 @@ with gr.Blocks(fill_height=True, theme=gr.themes.Soft(primary_hue="indigo")) as
|
|
| 130 |
clear_btn = gr.Button("🗑 مسح الذاكرة")
|
| 131 |
def clear_memory():
|
| 132 |
all_data = collection.get()
|
| 133 |
-
if all_data['ids']:
|
| 134 |
-
|
| 135 |
-
return "🗑 تم تصفير الذاكرة بالكامل."
|
| 136 |
clear_btn.click(clear_memory, outputs=status_msg)
|
| 137 |
|
|
|
|
| 138 |
gr.ChatInterface(
|
| 139 |
predict,
|
| 140 |
additional_inputs=[system_input, temp_slider, model_input],
|
| 141 |
-
fill_height=True
|
| 142 |
-
type="messages"
|
| 143 |
)
|
| 144 |
|
| 145 |
if __name__ == "__main__":
|
| 146 |
-
|
|
|
|
|
|
| 13 |
ef = embedding_functions.DefaultEmbeddingFunction()
|
| 14 |
collection = chroma_client.get_or_create_collection(name="yousef_vault", embedding_function=ef)
|
| 15 |
|
|
|
|
| 16 |
DEFAULT_MODEL = "huihui-ai/Qwen2.5-72B-Instruct-abliterated"
|
| 17 |
|
| 18 |
+
# --- 2. دالة استخراج النصوص ---
|
| 19 |
def process_document(file):
|
| 20 |
if file is None: return "⚠️ لم يتم اختيار ملف."
|
| 21 |
text = ""
|
|
|
|
| 33 |
text = f.read()
|
| 34 |
|
| 35 |
if not text.strip():
|
| 36 |
+
return "❌ تعذر استخراج نص."
|
| 37 |
|
|
|
|
| 38 |
chunks = [text[i:i+2000] for i in range(0, len(text), 2000)]
|
| 39 |
ids = [str(uuid.uuid4()) for _ in chunks]
|
| 40 |
collection.add(documents=chunks, ids=ids)
|
| 41 |
|
| 42 |
+
return f"✅ تمت الأرشفة: {len(chunks)} شظية معرفية."
|
| 43 |
except Exception as e:
|
| 44 |
+
return f"❌ خطأ: {str(e)}"
|
| 45 |
|
| 46 |
+
# --- 3. دالة الحوار المصححة ---
|
| 47 |
+
def predict(message, history, system_prompt, temperature, custom_model):
|
| 48 |
+
# ملاحظة: تم إزالة oauth_token من هنا لتبسيط الاتصال بالـ API
|
| 49 |
+
# تأكد من وضع HF_TOKEN في Secrets إذا كنت ترفعه على Space
|
| 50 |
+
api_key = os.getenv("HF_TOKEN")
|
| 51 |
|
| 52 |
if not api_key:
|
| 53 |
+
yield "⚠️ يرجى ضبط HF_TOKEN في الإعدادات."
|
| 54 |
return
|
| 55 |
|
| 56 |
+
client = OpenAI(base_url="https://router.huggingface.co/v1", api_key=api_key)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 57 |
active_model = custom_model.strip() if custom_model and custom_model.strip() else DEFAULT_MODEL
|
| 58 |
|
| 59 |
+
# RAG: استرجاع السياق
|
| 60 |
+
results = collection.query(query_texts=[message], n_results=5)
|
| 61 |
context = "\n".join(results['documents'][0]) if results['documents'] else ""
|
| 62 |
|
|
|
|
| 63 |
messages = [{"role": "system", "content": f"{system_prompt}\n\n[CONTEXT]:\n{context}"}]
|
| 64 |
+
# تحويل التاريخ إلى صيغة OpenAI المتوافقة
|
| 65 |
for msg in history:
|
| 66 |
+
messages.append({"role": msg["role"], "content": msg["content"]})
|
|
|
|
| 67 |
messages.append({"role": "user", "content": message})
|
| 68 |
|
| 69 |
try:
|
| 70 |
response = client.chat.completions.create(
|
| 71 |
model=active_model,
|
| 72 |
messages=messages,
|
| 73 |
+
temperature=float(temperature),
|
| 74 |
stream=True
|
| 75 |
)
|
| 76 |
|
| 77 |
partial_message = ""
|
| 78 |
for chunk in response:
|
| 79 |
+
if chunk.choices and chunk.choices[0].delta.content:
|
| 80 |
+
partial_message += chunk.choices[0].delta.content
|
| 81 |
+
yield partial_message
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 82 |
except Exception as e:
|
| 83 |
+
yield f"❌ فشل: {str(e)}"
|
|
|
|
|
|
|
|
|
|
|
|
|
| 84 |
|
| 85 |
+
# --- 4. بناء الواجهة (التحديث الجديد) ---
|
| 86 |
+
with gr.Blocks(fill_height=True) as demo:
|
| 87 |
with gr.Sidebar():
|
| 88 |
gr.Markdown("# 🧬 دهليز يوسف")
|
|
|
|
| 89 |
|
| 90 |
+
with gr.Accordion("⚙️ الضبط", open=True):
|
| 91 |
+
model_input = gr.Textbox(label="المحرك المستهدف", placeholder="Default Qwen", value="")
|
| 92 |
+
system_input = gr.Textbox(value="أنت رفيق حكيم.", label="البرومبت النظامي")
|
| 93 |
+
temp_slider = gr.Slider(0.1, 1.5, 0.8, label="Temperature")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 94 |
|
| 95 |
gr.Markdown("---")
|
| 96 |
+
file_box = gr.File(label="ارفع مسوداتك", file_types=[".pdf", ".docx", ".txt"])
|
|
|
|
| 97 |
status_msg = gr.Markdown("*الذاكرة بانتظار الملفات...*")
|
| 98 |
|
| 99 |
file_box.change(process_document, inputs=file_box, outputs=status_msg)
|
|
|
|
| 101 |
clear_btn = gr.Button("🗑 مسح الذاكرة")
|
| 102 |
def clear_memory():
|
| 103 |
all_data = collection.get()
|
| 104 |
+
if all_data['ids']: collection.delete(ids=all_data['ids'])
|
| 105 |
+
return "🗑 تم تصفير الذاكرة."
|
|
|
|
| 106 |
clear_btn.click(clear_memory, outputs=status_msg)
|
| 107 |
|
| 108 |
+
# التعديل هنا: حذف type="messages" لأنه أصبح افتراضياً أو غير مطلوب بهذا الشكل
|
| 109 |
gr.ChatInterface(
|
| 110 |
predict,
|
| 111 |
additional_inputs=[system_input, temp_slider, model_input],
|
| 112 |
+
fill_height=True
|
|
|
|
| 113 |
)
|
| 114 |
|
| 115 |
if __name__ == "__main__":
|
| 116 |
+
# تم نقل theme إلى هنا حسب تحذير Gradio 6
|
| 117 |
+
demo.launch(theme=gr.themes.Soft(primary_hue="indigo"))
|