File size: 3,130 Bytes
cb26a54 5515280 cb26a54 | 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 90 91 92 93 | """
Medical Park Makaleleri — Embedding Script
chunks.json icindeki her chunk_text icin embeddingmagibu-200m ile
768 boyutlu, L2-normalize chunk_vector uretir.
ONEMLI: Bu script CORPUS (dokuman) tarafi icindir -> encode_document() kullanilir.
Benchmarking asamasinda 30 test sorusunu embed ederken encode_query() kullanilmalidir.
Ikisini karistirmak benzerlik skorlarini bozar.
Kurulum:
pip install -U sentence-transformers transformers
Calistirma:
python embed_chunks.py
"""
import json
import os
import time
import torch
from sentence_transformers import SentenceTransformer
INPUT_PATH = "chunks.json"
OUTPUT_PATH = "chunks_with_vectors.json"
MODEL_NAME = "magibu/embeddingmagibu-200m"
BATCH_SIZE = 32 # GPU yoksa 8-16'ya dusurebilirsin
TEST_FIRST_N = 5 # once kucuk bir denemeyle dogrula, sonra tumunu calistir
def load_chunks(path: str) -> list[dict]:
with open(path, encoding="utf-8") as f:
return json.load(f)
def save_chunks(chunks: list[dict], path: str) -> None:
os.makedirs(os.path.dirname(path), exist_ok=True)
with open(path, "w", encoding="utf-8") as f:
json.dump(chunks, f, ensure_ascii=False)
def main():
# GPU gorunuyor mu kontrol et
cuda_ok = torch.cuda.is_available()
print(f"CUDA kullanilabilir mi: {cuda_ok}")
if cuda_ok:
print(f" GPU: {torch.cuda.get_device_name(0)}")
else:
print(" UYARI: GPU gorunmuyor, CPU'da calisacak (daha yavas olur).")
print(" torch'u CUDA destekli kurduğundan emin ol: https://pytorch.org/get-started/locally/")
chunks = load_chunks(INPUT_PATH)
print(f"\nYuklenen chunk sayisi: {len(chunks)}")
print(f"Model yukleniyor: {MODEL_NAME} (ilk calistirmada indirme suresi alabilir, ~yuzlerce MB)")
model = SentenceTransformer(MODEL_NAME, trust_remote_code=True)
print(f"Model cihazi: {model.device}")
texts = [c["chunk_text"] for c in chunks]
# --- Kucuk bir on-test: ilk birkac chunk uzerinde dogrula ---
print(f"\nOn test: ilk {TEST_FIRST_N} chunk encode ediliyor...")
sample_vecs = model.encode_document(texts[:TEST_FIRST_N])
print(f"Ornek vektor boyutu: {sample_vecs.shape}") # (TEST_FIRST_N, 768) olmali
assert sample_vecs.shape[1] == 768, "Beklenmeyen embedding boyutu!"
# --- Tum corpus'u encode et ---
print(f"\nTum chunk'lar encode ediliyor (batch_size={BATCH_SIZE})...")
t0 = time.time()
embeddings = model.encode_document(
texts,
batch_size=BATCH_SIZE,
show_progress_bar=True,
)
print(f"Bitti: {time.time() - t0:.1f} sn")
# --- Vektorleri chunk'lara ekle ---
for chunk, vec in zip(chunks, embeddings):
chunk["chunk_vector"] = vec.tolist()
save_chunks(chunks, OUTPUT_PATH)
print(f"\n[OK] Kaydedildi: {OUTPUT_PATH}")
print(f" Sutunlar: url, title, chunk_text, chunk_index, __source, chunk_vector")
print(f" Vektor boyutu: {len(chunks[0]['chunk_vector'])}")
# Dosya boyutu uyarisi
size_mb = os.path.getsize(OUTPUT_PATH) / (1024 * 1024)
print(f" Dosya boyutu: {size_mb:.1f} MB")
if __name__ == "__main__":
main()
|