Toivo0 commited on
Commit
cb26a54
·
verified ·
1 Parent(s): 44ed5c4

Upload embed_chunks.py with huggingface_hub

Browse files
Files changed (1) hide show
  1. embed_chunks.py +92 -0
embed_chunks.py ADDED
@@ -0,0 +1,92 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ """
2
+ Medical Park Makaleleri — Embedding Script
3
+ chunks.json icindeki her chunk_text icin embeddingmagibu-200m ile
4
+ 768 boyutlu, L2-normalize chunk_vector uretir.
5
+
6
+ ONEMLI: Bu script CORPUS (dokuman) tarafi icindir -> encode_document() kullanilir.
7
+ Benchmarking asamasinda 30 test sorusunu embed ederken encode_query() kullanilmalidir.
8
+ Ikisini karistirmak benzerlik skorlarini bozar.
9
+
10
+ Kurulum:
11
+ pip install -U sentence-transformers transformers
12
+
13
+ Calistirma:
14
+ python embed_chunks.py
15
+ """
16
+
17
+ import json
18
+ import os
19
+ import time
20
+
21
+ import torch
22
+ from sentence_transformers import SentenceTransformer
23
+
24
+ INPUT_PATH = "data/chunks.json"
25
+ OUTPUT_PATH = "data/chunks_with_vectors.json"
26
+ MODEL_NAME = "magibu/embeddingmagibu-200m"
27
+ BATCH_SIZE = 32 # GPU yoksa 8-16'ya dusurebilirsin
28
+ TEST_FIRST_N = 5 # once kucuk bir denemeyle dogrula, sonra tumunu calistir
29
+
30
+
31
+ def load_chunks(path: str) -> list[dict]:
32
+ with open(path, encoding="utf-8") as f:
33
+ return json.load(f)
34
+
35
+
36
+ def save_chunks(chunks: list[dict], path: str) -> None:
37
+ os.makedirs(os.path.dirname(path), exist_ok=True)
38
+ with open(path, "w", encoding="utf-8") as f:
39
+ json.dump(chunks, f, ensure_ascii=False)
40
+
41
+
42
+ def main():
43
+ # GPU gorunuyor mu kontrol et
44
+ cuda_ok = torch.cuda.is_available()
45
+ print(f"CUDA kullanilabilir mi: {cuda_ok}")
46
+ if cuda_ok:
47
+ print(f" GPU: {torch.cuda.get_device_name(0)}")
48
+ else:
49
+ print(" UYARI: GPU gorunmuyor, CPU'da calisacak (daha yavas olur).")
50
+ print(" torch'u CUDA destekli kurduğundan emin ol: https://pytorch.org/get-started/locally/")
51
+
52
+ chunks = load_chunks(INPUT_PATH)
53
+ print(f"\nYuklenen chunk sayisi: {len(chunks)}")
54
+
55
+ print(f"Model yukleniyor: {MODEL_NAME} (ilk calistirmada indirme suresi alabilir, ~yuzlerce MB)")
56
+ model = SentenceTransformer(MODEL_NAME, trust_remote_code=True)
57
+ print(f"Model cihazi: {model.device}")
58
+
59
+ texts = [c["chunk_text"] for c in chunks]
60
+
61
+ # --- Kucuk bir on-test: ilk birkac chunk uzerinde dogrula ---
62
+ print(f"\nOn test: ilk {TEST_FIRST_N} chunk encode ediliyor...")
63
+ sample_vecs = model.encode_document(texts[:TEST_FIRST_N])
64
+ print(f"Ornek vektor boyutu: {sample_vecs.shape}") # (TEST_FIRST_N, 768) olmali
65
+ assert sample_vecs.shape[1] == 768, "Beklenmeyen embedding boyutu!"
66
+
67
+ # --- Tum corpus'u encode et ---
68
+ print(f"\nTum chunk'lar encode ediliyor (batch_size={BATCH_SIZE})...")
69
+ t0 = time.time()
70
+ embeddings = model.encode_document(
71
+ texts,
72
+ batch_size=BATCH_SIZE,
73
+ show_progress_bar=True,
74
+ )
75
+ print(f"Bitti: {time.time() - t0:.1f} sn")
76
+
77
+ # --- Vektorleri chunk'lara ekle ---
78
+ for chunk, vec in zip(chunks, embeddings):
79
+ chunk["chunk_vector"] = vec.tolist()
80
+
81
+ save_chunks(chunks, OUTPUT_PATH)
82
+ print(f"\n[OK] Kaydedildi: {OUTPUT_PATH}")
83
+ print(f" Sutunlar: url, title, chunk_text, chunk_index, __source, chunk_vector")
84
+ print(f" Vektor boyutu: {len(chunks[0]['chunk_vector'])}")
85
+
86
+ # Dosya boyutu uyarisi
87
+ size_mb = os.path.getsize(OUTPUT_PATH) / (1024 * 1024)
88
+ print(f" Dosya boyutu: {size_mb:.1f} MB")
89
+
90
+
91
+ if __name__ == "__main__":
92
+ main()