Spaces:
Build error
Build error
File size: 3,275 Bytes
c0b2a4d |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 |
import dspy, gradio as gr
import chromadb
from chromadb.config import Settings
import fitz # PyMuPDF
from sentence_transformers import SentenceTransformer
import json
from dspy import Example, MIPROv2, Evaluate, evaluate
# إعداد LLM
dspy.settings.configure(lm=dspy.OpenAI(model="gpt-4"))
# إعداد قاعدة البيانات
client = chromadb.Client(Settings(chroma_db_impl="duckdb+parquet", persist_directory="./chroma_db"))
col = client.get_or_create_collection(name="arabic_docs", metadata={"hnsw:space": "cosine"})
embedder = SentenceTransformer("sentence-transformers/LaBSE")
# استيراد وتقطيع PDF
def process_pdf(pdf_bytes):
doc = fitz.open(stream=pdf_bytes, filetype="pdf")
texts = []
for p in doc:
text = p.get_text()
for chunk in text.split("\n\n"):
if len(chunk) > 50:
texts.append(chunk.strip())
return texts
def ingest(pdf_bytes):
texts = process_pdf(pdf_bytes)
embeddings = embedder.encode(texts, show_progress_bar=True)
for i, (chunk, emb) in enumerate(zip(texts, embeddings)):
col.add(ids=[f"chunk_{i}"], embeddings=[emb.tolist()], metadatas=[{"text": chunk}])
return f"تمت إضافة {len(texts)} مقطعاً"
retriever = dspy.Retrieve(lambda q: [m["text"] for m in col.query(q, n_results=3)["metadatas"]], k=1)
class RagSig(dspy.Signature):
question: str
context: str
answer: str
class RagMod(dspy.Module):
def __init__(self):
super().__init__()
self.predictor = dspy.Predict(RagSig)
def forward(self, question):
context = retriever(question)[0]
return self.predictor(question=question, context=context)
model = RagMod()
def answer(question):
out = model(question)
return out.answer
def load_dataset(path):
with open(path, "r", encoding="utf-8") as f:
return [Example(**json.loads(l)).with_inputs("question") for l in f]
def optimize(train_file, val_file):
trainset = load_dataset(train_file.name)
valset = load_dataset(val_file.name)
tp = MIPROv2(metric=evaluate.answer_exact_match, auto="light", num_threads=4)
optimized = tp.compile(model, trainset=trainset, valset=valset)
global model
model = optimized
return "✅ تم تحسين النموذج!"
with gr.Blocks() as demo:
gr.Markdown("## 🧠 نظام RAG عربي باستخدام DSPy")
with gr.Tab("📥 تحميل وتخزين"):
pdf_input = gr.File(label="ارفع ملف PDF")
ingest_btn = gr.Button("إضافة إلى قاعدة البيانات")
ingest_btn.click(ingest, inputs=pdf_input, outputs=gr.Textbox())
with gr.Tab("❓ سؤال"):
q = gr.Textbox(label="اكتب سؤالك")
answer_btn = gr.Button("احصل على الإجابة")
out = gr.Textbox(label="الإجابة")
answer_btn.click(answer, inputs=q, outputs=out)
with gr.Tab("⚙️ تحسين النموذج"):
train_file = gr.File(label="trainset.jsonl")
val_file = gr.File(label="valset.jsonl")
opt_btn = gr.Button("ابدأ التحسين")
result = gr.Textbox(label="نتيجة التحسين")
opt_btn.click(optimize, inputs=[train_file, val_file], outputs=result)
demo.launch()
|