Spaces:
Sleeping
Sleeping
ADD: implement text chunking functionality for policy content processing
Browse files- rag/chunker.py +27 -0
rag/chunker.py
ADDED
|
@@ -0,0 +1,27 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
from langchain_text_splitters import RecursiveCharacterTextSplitter
|
| 2 |
+
|
| 3 |
+
|
| 4 |
+
splitter = RecursiveCharacterTextSplitter(
|
| 5 |
+
chunk_size=500,
|
| 6 |
+
chunk_overlap=50,
|
| 7 |
+
separators=["\n\n", "\n", ".", "!", "?", ",", " "],
|
| 8 |
+
length_function=len,
|
| 9 |
+
)
|
| 10 |
+
|
| 11 |
+
|
| 12 |
+
def prepare_policy_chunks(policy: dict) -> list[dict]:
|
| 13 |
+
chunks = splitter.split_text(policy["content"])
|
| 14 |
+
return [
|
| 15 |
+
{
|
| 16 |
+
"workspace_id": policy["workspace_id"],
|
| 17 |
+
"source_type": "policy",
|
| 18 |
+
"source_id": policy["id"],
|
| 19 |
+
"chunk_index": i,
|
| 20 |
+
"content": chunk,
|
| 21 |
+
"metadata": {
|
| 22 |
+
"policy_type": policy["type"],
|
| 23 |
+
"policy_title": policy["title"],
|
| 24 |
+
},
|
| 25 |
+
}
|
| 26 |
+
for i, chunk in enumerate(chunks)
|
| 27 |
+
]
|