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
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
|
# type: ignore
import asyncio
import base64
import json
import logging
import string
import time
import unicodedata
from io import BytesIO
from typing import AsyncGenerator
from pdf2image import convert_from_bytes, convert_from_path
from pdf2image.exceptions import PDFInfoNotInstalledError
from PIL import Image
from pypdf import PdfReader
from core.base.abstractions import GenerationConfig
from core.base.parsers.base_parser import AsyncParser
from core.base.providers import (
CompletionProvider,
DatabaseProvider,
IngestionConfig,
)
from shared.abstractions import PDFParsingError, PopplerNotFoundError
logger = logging.getLogger()
class VLMPDFParser(AsyncParser[str | bytes]):
"""A parser for PDF documents using vision models for page processing."""
def __init__(
self,
config: IngestionConfig,
database_provider: DatabaseProvider,
llm_provider: CompletionProvider,
):
self.database_provider = database_provider
self.llm_provider = llm_provider
self.config = config
self.vision_prompt_text = None
async def convert_pdf_to_images(
self, data: str | bytes
) -> list[Image.Image]:
"""Convert PDF pages to images asynchronously using in-memory
conversion."""
logger.info("Starting PDF conversion to images.")
start_time = time.perf_counter()
options = {
"dpi": 300, # You can make this configurable via self.config if needed
"fmt": "jpeg",
"thread_count": 4,
"paths_only": False, # Return PIL Image objects instead of writing to disk
}
try:
if isinstance(data, bytes):
images = await asyncio.to_thread(
convert_from_bytes, data, **options
)
else:
images = await asyncio.to_thread(
convert_from_path, data, **options
)
elapsed = time.perf_counter() - start_time
logger.info(
f"PDF conversion completed in {elapsed:.2f} seconds, total pages: {len(images)}"
)
return images
except PDFInfoNotInstalledError as e:
logger.error(
"PDFInfoNotInstalledError encountered during PDF conversion."
)
raise PopplerNotFoundError() from e
except Exception as err:
logger.error(
f"Error converting PDF to images: {err} type: {type(err)}"
)
raise PDFParsingError(
f"Failed to process PDF: {str(err)}", err
) from err
async def process_page(
self, image: Image.Image, page_num: int
) -> dict[str, str]:
"""Process a single PDF page using the vision model."""
page_start = time.perf_counter()
try:
# Convert PIL image to JPEG bytes in-memory
buf = BytesIO()
image.save(buf, format="JPEG")
buf.seek(0)
image_data = buf.read()
image_base64 = base64.b64encode(image_data).decode("utf-8")
model = self.config.app.vlm
# Configure generation parameters
generation_config = GenerationConfig(
model=self.config.app.vlm,
stream=False,
)
is_anthropic = model and "anthropic/" in model
# FIXME: This is a hacky fix to handle the different formats
# that was causing an outage. This logic really needs to be refactored
# and cleaned up such that it handles providers more robustly.
# Prepare message with image content
if is_anthropic:
messages = [
{
"role": "user",
"content": [
{"type": "text", "text": self.vision_prompt_text},
{
"type": "image",
"source": {
"type": "base64",
"media_type": "image/jpeg",
"data": image_base64,
},
},
],
}
]
else:
# Use OpenAI format
messages = [
{
"role": "user",
"content": [
{"type": "text", "text": self.vision_prompt_text},
{
"type": "image_url",
"image_url": {
"url": f"data:image/jpeg;base64,{image_base64}"
},
},
],
}
]
logger.debug(f"Sending page {page_num} to vision model.")
req_start = time.perf_counter()
if is_anthropic:
response = await self.llm_provider.aget_completion(
messages=messages,
generation_config=generation_config,
tools=[
{
"name": "parse_pdf_page",
"description": "Parse text content from a PDF page",
"input_schema": {
"type": "object",
"properties": {
"page_content": {
"type": "string",
"description": "Extracted text from the PDF page",
},
},
"required": ["page_content"],
},
}
],
tool_choice={"type": "tool", "name": "parse_pdf_page"},
)
if (
response.choices
and response.choices[0].message
and response.choices[0].message.tool_calls
):
tool_call = response.choices[0].message.tool_calls[0]
args = json.loads(tool_call.function.arguments)
content = args.get("page_content", "")
page_elapsed = time.perf_counter() - page_start
logger.debug(
f"Processed page {page_num} in {page_elapsed:.2f} seconds."
)
return {"page": str(page_num), "content": content}
else:
logger.warning(
f"No valid tool call in response for page {page_num}, document might be missing text."
)
else:
response = await self.llm_provider.aget_completion(
messages=messages, generation_config=generation_config
)
req_elapsed = time.perf_counter() - req_start
logger.debug(
f"Vision model response for page {page_num} received in {req_elapsed:.2f} seconds."
)
if response.choices and response.choices[0].message:
content = response.choices[0].message.content
page_elapsed = time.perf_counter() - page_start
logger.debug(
f"Processed page {page_num} in {page_elapsed:.2f} seconds."
)
return {"page": str(page_num), "content": content}
else:
msg = f"No response content for page {page_num}"
logger.error(msg)
raise ValueError(msg)
except Exception as e:
logger.error(
f"Error processing page {page_num} with vision model: {str(e)}"
)
raise
async def ingest(
self, data: str | bytes, maintain_order: bool = True, **kwargs
) -> AsyncGenerator[dict[str, str | int], None]:
"""Ingest PDF data and yield the text description for each page using
the vision model.
(This version yields a string per page rather than a dictionary.)
"""
ingest_start = time.perf_counter()
logger.info("Starting PDF ingestion using VLMPDFParser.")
if not self.vision_prompt_text:
self.vision_prompt_text = (
await self.database_provider.prompts_handler.get_cached_prompt(
prompt_name=self.config.vision_pdf_prompt_name
)
)
logger.info("Retrieved vision prompt text from database.")
try:
# Convert PDF to images (in-memory)
images = await self.convert_pdf_to_images(data)
# Create asynchronous tasks for processing each page
tasks = {
asyncio.create_task(
self.process_page(image, page_num)
): page_num
for page_num, image in enumerate(images, 1)
}
if maintain_order:
pending = set(tasks.keys())
results = {}
next_page = 1
while pending:
done, pending = await asyncio.wait(
pending, return_when=asyncio.FIRST_COMPLETED
)
for task in done:
result = await task
page_num = int(result["page"])
results[page_num] = result
while next_page in results:
yield {
"content": results[next_page]["content"] or "",
"page_number": next_page,
}
results.pop(next_page)
next_page += 1
else:
# Yield results as tasks complete
for coro in asyncio.as_completed(tasks.keys()):
result = await coro
yield {
"content": result["content"],
"page_number": int(result["page"]),
}
total_elapsed = time.perf_counter() - ingest_start
logger.info(
f"Completed PDF ingestion in {total_elapsed:.2f} seconds using VLMPDFParser."
)
except Exception as e:
logger.error(f"Error processing PDF: {str(e)}")
raise
class BasicPDFParser(AsyncParser[str | bytes]):
"""A parser for PDF data."""
def __init__(
self,
config: IngestionConfig,
database_provider: DatabaseProvider,
llm_provider: CompletionProvider,
):
self.database_provider = database_provider
self.llm_provider = llm_provider
self.config = config
self.PdfReader = PdfReader
async def ingest(
self, data: str | bytes, **kwargs
) -> AsyncGenerator[str, None]:
"""Ingest PDF data and yield text from each page."""
if isinstance(data, str):
raise ValueError("PDF data must be in bytes format.")
pdf = self.PdfReader(BytesIO(data))
for page in pdf.pages:
page_text = page.extract_text()
if page_text is not None:
page_text = "".join(
filter(
lambda x: (
unicodedata.category(x)
in [
"Ll",
"Lu",
"Lt",
"Lm",
"Lo",
"Nl",
"No",
] # Keep letters and numbers
or "\u4e00" <= x <= "\u9fff" # Chinese characters
or "\u0600" <= x <= "\u06ff" # Arabic characters
or "\u0400" <= x <= "\u04ff" # Cyrillic letters
or "\u0370" <= x <= "\u03ff" # Greek letters
or "\u0e00" <= x <= "\u0e7f" # Thai
or "\u3040" <= x <= "\u309f" # Japanese Hiragana
or "\u30a0" <= x <= "\u30ff" # Katakana
or x in string.printable
),
page_text,
)
) # Keep characters in common languages ; # Filter out non-printable characters
yield page_text
class PDFParserUnstructured(AsyncParser[str | bytes]):
def __init__(
self,
config: IngestionConfig,
database_provider: DatabaseProvider,
llm_provider: CompletionProvider,
):
self.database_provider = database_provider
self.llm_provider = llm_provider
self.config = config
try:
from unstructured.partition.pdf import partition_pdf
self.partition_pdf = partition_pdf
except ImportError as e:
logger.error("PDFParserUnstructured ImportError : ", e)
async def ingest(
self,
data: str | bytes,
partition_strategy: str = "hi_res",
chunking_strategy="by_title",
) -> AsyncGenerator[str, None]:
# partition the pdf
elements = self.partition_pdf(
file=BytesIO(data),
partition_strategy=partition_strategy,
chunking_strategy=chunking_strategy,
)
for element in elements:
yield element.text
|