token ingeest

scripts/parser/file/rst_parser.py

@@ -29,7 +29,6 @@ class RstParser(BaseParser):
         remove_whitespaces_excess: bool = True,
         #Be carefull with remove_characters_excess, might cause data loss
         remove_characters_excess: bool = True,
-        # max_tokens: int = 2048,
         **kwargs: Any,
     ) -> None:
         """Init params."""
@@ -41,18 +40,6 @@ class RstParser(BaseParser):
         self._remove_directives = remove_directives
         self._remove_whitespaces_excess = remove_whitespaces_excess
         self._remove_characters_excess = remove_characters_excess
-        # self._max_tokens = max_tokens
-
-    # def tups_chunk_append(self, tups: List[Tuple[Optional[str], str]], current_header: Optional[str], current_text: str):
-    #     """Append to tups chunk."""
-    #     num_tokens = len(tiktoken.get_encoding("cl100k_base").encode(current_text))
-    #     if num_tokens > self._max_tokens:
-    #         chunks = [current_text[i:i + self._max_tokens] for i in range(0, len(current_text), self._max_tokens)]
-    #         for chunk in chunks:
-    #             tups.append((current_header, chunk))
-    #     else:
-    #         tups.append((current_header, current_text))
-    #     return tups
 
 
     def rst_to_tups(self, rst_text: str) -> List[Tuple[Optional[str], str]]:

scripts/parser/token_func.py

@@ -0,0 +1,70 @@
+import re
+import tiktoken
+
+from typing import List
+from parser.schema.base import Document
+from math import ceil
+
+
+def separate_header_and_body(text):
+    header_pattern = r"^(.*?\n){3}"
+    match = re.match(header_pattern, text)
+    header = match.group(0)
+    body = text[len(header):]
+    return header, body
+
+def group_documents(documents: List[Document], min_tokens: int = 200, max_tokens: int = 2000) -> List[Document]:
+    docs = []
+    current_group = None
+
+    for doc in documents:
+        doc_len = len(tiktoken.get_encoding("cl100k_base").encode(doc.text))
+
+        if current_group is None:
+            current_group = Document(text=doc.text, doc_id=doc.doc_id, embedding=doc.embedding,
+                                     extra_info=doc.extra_info)
+        elif len(tiktoken.get_encoding("cl100k_base").encode(current_group.text)) + doc_len < max_tokens and doc_len >= min_tokens:
+            current_group.text += " " + doc.text
+        else:
+            docs.append(current_group)
+            current_group = Document(text=doc.text, doc_id=doc.doc_id, embedding=doc.embedding,
+                                     extra_info=doc.extra_info)
+
+    if current_group is not None:
+        docs.append(current_group)
+
+    return docs
+
+def split_documents(documents: List[Document], max_tokens: int = 2000) -> List[Document]:
+    docs = []
+    for doc in documents:
+        token_length = len(tiktoken.get_encoding("cl100k_base").encode(doc.text))
+        if token_length <= max_tokens:
+            docs.append(doc)
+        else:
+            header, body = separate_header_and_body(doc.text)
+            num_body_parts = ceil(token_length / max_tokens)
+            part_length = ceil(len(body) / num_body_parts)
+            body_parts = [body[i:i + part_length] for i in range(0, len(body), part_length)]
+            for i, body_part in enumerate(body_parts):
+                new_doc = Document(text=header + body_part.strip(),
+                                   doc_id=f"{doc.doc_id}-{i}",
+                                   embedding=doc.embedding,
+                                   extra_info=doc.extra_info)
+                docs.append(new_doc)
+    return docs
+
+def group_split(documents: List[Document], max_tokens: int = 1500, min_tokens: int = 500, token_check: bool = True):
+    if token_check == False:
+        return documents
+    print("Grouping small documents")
+    try:
+        documents = group_documents(documents=documents, min_tokens=min_tokens, max_tokens=max_tokens)
+    except:
+        print("Grouping failed, try running without token_check")
+    print("Separating large documents")
+    try:
+        documents = split_documents(documents=documents, max_tokens=max_tokens)
+    except:
+        print("Grouping failed, try running without token_check")
+    return documents