-
Notifications
You must be signed in to change notification settings - Fork 1
Expand file tree
/
Copy pathrag_pipeline.py
More file actions
160 lines (122 loc) · 5.11 KB
/
rag_pipeline.py
File metadata and controls
160 lines (122 loc) · 5.11 KB
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
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
import os
from dotenv import load_dotenv
from haystack import Pipeline, component
from haystack.components.builders import PromptBuilder
from haystack.components.embedders import SentenceTransformersTextEmbedder
from haystack_integrations.document_stores.chroma import ChromaDocumentStore
from haystack_integrations.components.retrievers.chroma import ChromaEmbeddingRetriever
from google import genai
from google.genai import types
load_dotenv()
# Configuration
CHROMA_PERSIST_PATH = "chroma_db"
CHROMA_COLLECTION_NAME = "datacommit_all"
EMBEDDING_MODEL = "sentence-transformers/all-MiniLM-L6-v2"
GEMINI_MODEL = "gemini-3-flash-preview"
TEMPLATE = '''
Sen DataCommit podcast serisinin içeriklerinden sorulara cevap veren bir asistansın.
DataCommit, veri bilimi alanında deneyimli uzmanların kariyer yolculuklarını ve teknik bilgilerini paylaştığı bir platformdur.
Aşağıdaki bölümlerden gelen bilgilere dayanarak soruyu cevapla.
Context:
{% for doc in documents %}
---
[Bölüm {{ doc.meta.episode }} - Konuk: {{ doc.meta.guest }}]
{{ doc.content }}
---
{% endfor %}
Soru: {{question}}
Cevabını verirken:
1. İlgili bilgileri sentezle
2. Her önemli bilgi için kaynak göster: (Bölüm X, Konuk adı)
3. Birden fazla bölümden bilgi varsa hepsini belirt
Cevap:
'''
@component
class GeminiGenerator:
def __init__(self, model: str = "gemini-3-flash-preview", temperature: float = 0.5):
self.model = model
self.temperature = temperature
self.client = genai.Client(api_key=os.getenv("GEMINI_API_KEY"))
@component.output_types(replies=list)
def run(self, parts: str):
response = self.client.models.generate_content(
model=self.model,
contents=parts,
config=types.GenerateContentConfig(
temperature=self.temperature,
thinking_config=types.ThinkingConfig(thinking_budget=0)
)
)
return {"replies": [response.text]}
# Initialize components
document_store = ChromaDocumentStore(
persist_path=CHROMA_PERSIST_PATH,
collection_name=CHROMA_COLLECTION_NAME
)
query_embedder = SentenceTransformersTextEmbedder(model=EMBEDDING_MODEL)
query_embedder.warm_up()
retriever = ChromaEmbeddingRetriever(document_store=document_store, top_k=5)
prompt_builder = PromptBuilder(template=TEMPLATE, required_variables=["documents", "question"])
generator = GeminiGenerator(model=GEMINI_MODEL, temperature=0.5)
# Build pipeline
rag_pipeline = Pipeline()
rag_pipeline.add_component("query_embedder", query_embedder)
rag_pipeline.add_component("retriever", retriever)
rag_pipeline.add_component("prompt_builder", prompt_builder)
rag_pipeline.add_component("llm", generator)
rag_pipeline.connect("query_embedder.embedding", "retriever.query_embedding")
rag_pipeline.connect("retriever.documents", "prompt_builder.documents")
rag_pipeline.connect("prompt_builder.prompt", "llm.parts")
def show_retrieved_chunks(query: str, top_k: int = 5):
embedding = query_embedder.run(text=query)["embedding"]
docs = retriever.run(query_embedding=embedding, top_k=top_k)["documents"]
print(f"Query: {query}\n")
print(f"Retrieved {len(docs)} chunks:\n" + "=" * 80)
for i, doc in enumerate(docs, 1):
ep = doc.meta.get("episode", "?")
guest = doc.meta.get("guest", "Unknown")
print(f"\n[Chunk {i}] Bölüm {ep} - {guest} | Score: {doc.score:.4f}")
print("-" * 40)
print(doc.content[:400] + "..." if len(doc.content) > 400 else doc.content)
print("=" * 80)
return docs
def respond_with_sources(query: str, top_k: int = 5):
if not query.strip():
return "", []
embedding = query_embedder.run(text=query)["embedding"]
docs = retriever.run(query_embedding=embedding, top_k=top_k)["documents"]
prompt = prompt_builder.run(documents=docs, question=query)["prompt"]
response = generator.run(parts=prompt)["replies"][0]
sources = []
seen = set()
for doc in docs:
key = (doc.meta.get("episode"), doc.meta.get("guest"))
if key not in seen:
sources.append({
"episode": doc.meta.get("episode"),
"guest": doc.meta.get("guest"),
"score": doc.score
})
seen.add(key)
return response, sources
def query_datacommit(query: str, show_chunks: bool = False):
print("\n" + "=" * 80)
print(f"SORU: {query}")
print("=" * 80)
if show_chunks:
print("\n--- Retrieved Chunks ---")
show_retrieved_chunks(query)
response, sources = respond_with_sources(query)
print("\n--- CEVAP ---")
print(response)
print("\n--- KAYNAKLAR ---")
for src in sources:
print(f" • Bölüm {src['episode']}: {src['guest']} (relevance: {src['score']:.4f})")
return response, sources
if __name__ == "__main__":
doc_count = document_store.count_documents()
if doc_count == 0:
print("No documents found! Run 'python create_database.py' first.")
else:
print(f"Loaded {doc_count} chunks from all episodes\n")
query_datacommit("Jr lar nasıl iş bulur?", show_chunks=False)