Removing pymupdf

gantry-requirements.txt

@@ -2,7 +2,6 @@ torchvision
 cached-path
 smart_open
 pypdf
-pymupdf
 pypdfium2
 lingua-language-detector
 Pillow

olmocr/prompts/anchor.py

@@ -4,7 +4,6 @@
 
 # pdftotext
 # pdfium
-# pymupdf
 # pypdf
 
 import random
@@ -16,7 +15,6 @@ from dataclasses import dataclass
 from typing import List, Literal
 
 import ftfy
-import pymupdf
 import pypdfium2 as pdfium
 from pypdf import PdfReader
 from pypdf.generic import RectangleObject
@@ -25,7 +23,7 @@ from olmocr.filter.coherency import get_document_coherency
 
 
 def get_anchor_text(
-    local_pdf_path: str, page: int, pdf_engine: Literal["pdftotext", "pdfium", "pymupdf", "pypdf", "topcoherency", "pdfreport"], target_length: int = 4000
+    local_pdf_path: str, page: int, pdf_engine: Literal["pdftotext", "pdfium", "pypdf", "topcoherency", "pdfreport"], target_length: int = 4000
 ) -> str:
     assert page > 0, "Pages are 1-indexed in pdf-land"
 
@@ -35,12 +33,9 @@ def get_anchor_text(
         return _get_pdfium(local_pdf_path, page)
     elif pdf_engine == "pypdf":
         return _get_pypdf_raw(local_pdf_path, page)
-    elif pdf_engine == "pymupdf":
-        return _get_pymupdf(local_pdf_path, page)
     elif pdf_engine == "topcoherency":
         options = {
             "pdftotext": _get_pdftotext(local_pdf_path, page),
-            "pymupdf": _get_pymupdf(local_pdf_path, page),
             "pdfium": _get_pdfium(local_pdf_path, page),
             "pypdf_raw": _get_pypdf_raw(local_pdf_path, page),
         }
@@ -70,11 +65,6 @@ def _get_pdftotext(local_pdf_path: str, page: int) -> str:
     return pdftotext_result.stdout.decode("utf-8")
 
 
-def _get_pymupdf(local_pdf_path: str, page: int) -> str:
-    pm_doc = pymupdf.open(local_pdf_path)
-    return pm_doc[page - 1].get_text()
-
-
 def _get_pypdf_raw(local_pdf_path: str, page: int) -> str:
     reader = PdfReader(local_pdf_path)
     pypage = reader.pages[page - 1]

pyproject.toml

@@ -22,7 +22,6 @@ dependencies = [
   "cached-path",
   "smart_open",
   "pypdf>=5.2.0",
-  "pymupdf",
   "pypdfium2",
   "cryptography",
   "lingua-language-detector",

tests/test_coherency.py

@@ -43,27 +43,17 @@ class TestCoherencyScores(unittest.TestCase):
             page=2,
             pdf_engine="pdftotext",
         )
-        pymupdf_text = get_anchor_text(
-            os.path.join(os.path.dirname(__file__), "gnarly_pdfs", "pdftotext_two_column_issue.pdf"),
-            page=2,
-            pdf_engine="pymupdf",
-        )
         pdfium_text = get_anchor_text(
             os.path.join(os.path.dirname(__file__), "gnarly_pdfs", "pdftotext_two_column_issue.pdf"),
             page=2,
             pdf_engine="pdfium",
         )
 
-        # pdftotext_text = get_document_text(os.path.join(os.path.dirname(__file__), "gnarly_pdfs", "pdftotext_two_column_issue.pdf"), pdf_engine="pdftotext")
-        # pymupdf_text = get_document_text(os.path.join(os.path.dirname(__file__), "gnarly_pdfs", "pdftotext_two_column_issue.pdf"), pdf_engine="pymupdf")
-
         print("pdftotext_text", pdftotext_score := get_document_coherency(pdftotext_text))
-        print("pymupdf_text", pymupdf_score := get_document_coherency(pymupdf_text))
         print("pdfium_text", pdfium_score := get_document_coherency(pdfium_text))
 
-        self.assertLess(pdftotext_score, pymupdf_score)
-        self.assertLess(pdfium_score, pymupdf_score)
+        self.assertLess(pdfium_score, pdftotext_score)
 
         anchor_text = get_anchor_text(os.path.join(os.path.dirname(__file__), "gnarly_pdfs", "pdftotext_two_column_issue.pdf"), 2, pdf_engine="topcoherency")
 
-        self.assertEqual(anchor_text, pymupdf_text)
+        self.assertEqual(anchor_text, pdfium_text)