Partition pdf

backend/app/config.py

@@ -13,6 +13,7 @@ class EmbeddingConfig(TypedDict):
     model_id: str
     chunk_size: int
     chunk_overlap: int
+    enable_partition_pdf: bool
 
 
 # Configure generation parameter for Claude chat response.
@@ -42,6 +43,7 @@ class EmbeddingConfig(TypedDict):
     # NOTE: consider that cohere allows up to 2048 tokens per request
     "chunk_size": 1000,
     "chunk_overlap": 200,
+    "enable_partition_pdf": False,
 }
 
 # Configure search parameter to fetch relevant documents from vector store.

backend/app/repositories/custom_bot.py

@@ -334,6 +334,12 @@ def find_private_bot_by_id(user_id: str, bot_id: str) -> BotModel:
                 and "chunk_overlap" in item["EmbeddingParams"]
                 else 200
             ),
+            enable_partition_pdf=(
+                item["EmbeddingParams"]["enable_partition_pdf"]
+                if "EmbeddingParams" in item
+                and "enable_partition_pdf" in item["EmbeddingParams"]
+                else False
+            ),
         ),
         generation_params=GenerationParamsModel(
             **(
@@ -407,6 +413,12 @@ def find_public_bot_by_id(bot_id: str) -> BotModel:
                 and "chunk_overlap" in item["EmbeddingParams"]
                 else 200
             ),
+            enable_partition_pdf=(
+                item["EmbeddingParams"]["enable_partition_pdf"]
+                if "EmbeddingParams" in item
+                and "enable_partition_pdf" in item["EmbeddingParams"]
+                else False
+            ),
         ),
         generation_params=GenerationParamsModel(
             **(

backend/app/repositories/models/custom_bot.py

@@ -6,6 +6,7 @@
 class EmbeddingParamsModel(BaseModel):
     chunk_size: int
     chunk_overlap: int
+    enable_partition_pdf: bool
 
 
 class KnowledgeModel(BaseModel):

backend/app/routes/bot.py

@@ -129,6 +129,7 @@ def get_private_bot(request: Request, bot_id: str):
         embedding_params=EmbeddingParams(
             chunk_size=bot.embedding_params.chunk_size,
             chunk_overlap=bot.embedding_params.chunk_overlap,
+            enable_partition_pdf=bot.embedding_params.enable_partition_pdf,
         ),
         knowledge=Knowledge(
             source_urls=bot.knowledge.source_urls,

backend/app/routes/schemas/bot.py

@@ -16,6 +16,7 @@
 class EmbeddingParams(BaseSchema):
     chunk_size: int
     chunk_overlap: int
+    enable_partition_pdf: bool
 
 
 class GenerationParams(BaseSchema):
@@ -94,6 +95,8 @@ def is_embedding_required(self, current_bot_model: BotModel) -> bool:
                 == current_bot_model.embedding_params.chunk_size
                 and self.embedding_params.chunk_overlap
                 == current_bot_model.embedding_params.chunk_overlap
+                and self.embedding_params.enable_partition_pdf
+                == current_bot_model.embedding_params.enable_partition_pdf
             ):
                 pass
             else:

backend/app/usecases/bot.py

@@ -131,6 +131,12 @@ def create_new_bot(user_id: str, bot_input: BotInput) -> BotOutput:
         else DEFAULT_EMBEDDING_CONFIG["chunk_overlap"]
     )
 
+    enable_partition_pdf = (
+        bot_input.embedding_params.enable_partition_pdf
+        if bot_input.embedding_params
+        else DEFAULT_EMBEDDING_CONFIG["enable_partition_pdf"]
+    )
+
     generation_params = (
         bot_input.generation_params.model_dump()
         if bot_input.generation_params
@@ -158,6 +164,7 @@ def create_new_bot(user_id: str, bot_input: BotInput) -> BotOutput:
             embedding_params=EmbeddingParamsModel(
                 chunk_size=chunk_size,
                 chunk_overlap=chunk_overlap,
+                enable_partition_pdf=enable_partition_pdf,
             ),
             generation_params=GenerationParamsModel(**generation_params),
             search_params=SearchParamsModel(**search_params),
@@ -185,6 +192,7 @@ def create_new_bot(user_id: str, bot_input: BotInput) -> BotOutput:
         embedding_params=EmbeddingParams(
             chunk_size=chunk_size,
             chunk_overlap=chunk_overlap,
+            enable_partition_pdf=enable_partition_pdf,
         ),
         generation_params=GenerationParams(**generation_params),
         search_params=SearchParams(**search_params),
@@ -239,6 +247,12 @@ def modify_owned_bot(
         else DEFAULT_EMBEDDING_CONFIG["chunk_overlap"]
     )
 
+    enable_partition_pdf = (
+        modify_input.embedding_params.enable_partition_pdf
+        if modify_input.embedding_params
+        else DEFAULT_EMBEDDING_CONFIG["enable_partition_pdf"]
+    )
+
     generation_params = (
         modify_input.generation_params.model_dump()
         if modify_input.generation_params
@@ -266,6 +280,7 @@ def modify_owned_bot(
         embedding_params=EmbeddingParamsModel(
             chunk_size=chunk_size,
             chunk_overlap=chunk_overlap,
+            enable_partition_pdf=enable_partition_pdf,
         ),
         generation_params=GenerationParamsModel(**generation_params),
         search_params=SearchParamsModel(**search_params),
@@ -286,6 +301,7 @@ def modify_owned_bot(
         embedding_params=EmbeddingParams(
             chunk_size=chunk_size,
             chunk_overlap=chunk_overlap,
+            enable_partition_pdf=enable_partition_pdf,
         ),
         generation_params=GenerationParams(**generation_params),
         search_params=SearchParams(**search_params),

backend/embedding.requirements.txt

@@ -15,3 +15,5 @@ unstructured[docx]==0.12.6
 unstructured[xlsx]==0.12.6
 unstructured[pptx]==0.12.6
 unstructured[md]==0.12.6
+retry==0.9.2
+types-retry==0.9.9.4

backend/embedding/loaders/s3.py

@@ -1,20 +1,32 @@
 import os
 import tempfile
-
+import logging
 import boto3
+from distutils.util import strtobool
 from embedding.loaders.base import BaseLoader, Document
 from unstructured.partition.auto import partition
+from unstructured.partition.pdf import partition_pdf
+
+logger = logging.getLogger(__name__)
+logger.setLevel(logging.DEBUG)
 
 
 class S3FileLoader(BaseLoader):
     """Loads a document from a file in S3.
     Reference: `langchain_community.document_loaders.S3FileLoader` class
     """
 
-    def __init__(self, bucket: str, key: str, mode: str = "single"):
+    def __init__(
+        self,
+        bucket: str,
+        key: str,
+        mode: str = "single",
+        enable_partition_pdf: bool = False,
+    ):
         self.bucket = bucket
         self.key = key
         self.mode = mode
+        self.enable_partition_pdf = enable_partition_pdf
 
     def _get_elements(self) -> list:
         """Get elements."""
@@ -23,7 +35,19 @@ def _get_elements(self) -> list:
             file_path = f"{temp_dir}/{self.key}"
             os.makedirs(os.path.dirname(file_path), exist_ok=True)
             s3.download_file(self.bucket, self.key, file_path)
-            return partition(filename=file_path)
+            extension = os.path.splitext(file_path)[1]
+
+            if extension == ".pdf" and self.enable_partition_pdf == True:
+                logger.info(f"Start partitioning using hi-resolution mode: {file_path}")
+                return partition_pdf(
+                    filename=file_path,
+                    strategy="hi_res",
+                    infer_table_structure=True,
+                    extract_images_in_pdf=False,
+                )
+            else:
+                logger.info(f"Start partitioning using auto mode: {file_path}")
+                return partition(filename=file_path)
 
     def _get_metadata(self) -> dict:
         return {"source": f"s3://{self.bucket}/{self.key}"}