pyconverters-openai_vision 0.5.3__py3-none-any.whl → 0.5.4__py3-none-any.whl

pyconverters_openai_vision/__init__.py

@@ -1,2 +1,2 @@
 """OpenAIVision converter"""
-__version__ = "0.5.3"
+__version__ = "0.5.4"

pyconverters_openai_vision/openai_vision.py

@@ -1,7 +1,9 @@
 import base64
 import os
+import re
 from enum import Enum
 from logging import Logger
+from re import Pattern
 from typing import List, cast, Type, Dict, Any
 
 import filetype as filetype
@@ -24,12 +26,13 @@ class OpenAIVisionBaseParameters(ConverterParameters):
         None, extra="internal"
     )
     prompt: str = Field(
-        "Describe the image with a lot of details",
+        """If the attached file is an image: describe the image with a lot of details.",
+        If the attached file is a PDF document: convert the PDF document into Markdown format. The output must be just the markdown result without any explanation or introductory prefix.""",
         description="""Contains the prompt as a string""",
         extra="multiline",
     )
     max_tokens: int = Field(
-        256,
+        16384,
         description="""The maximum number of tokens to generate in the completion.
     The token count of your prompt plus max_tokens cannot exceed the model's context length.
     Most models have a context length of 2048 tokens (except for the newest models, which support 4096).""",
@@ -85,13 +88,14 @@ class OpenAIVisionBaseParameters(ConverterParameters):
 class OpenAIVisionModel(str, Enum):
     gpt_4o_mini = "gpt-4o-mini"
     gpt_4o = "gpt-4o"
+    o3_mini = "o3-mini"
 
 
 class OpenAIVisionParameters(OpenAIVisionBaseParameters):
     model: OpenAIVisionModel = Field(
         OpenAIVisionModel.gpt_4o_mini,
-        description="""The [OpenAI model](https://platform.openai.com/docs/models) used for speech to text transcription. Options currently available:</br>
-                        <li>`whisper-1` - state-of-the-art open source large-v2 Whisper model.
+        description="""The [OpenAI model](https://platform.openai.com/docs/models) used for vision. Options currently available:</br>
+
                         """, extra="pipeline-naming-hint"
     )
 
@@ -126,10 +130,24 @@ class OpenAIVisionConverterBase(ConverterBase):
     You input some text as a prompt, and the model will generate a text completion that attempts to match whatever context or pattern you gave it."""
     PREFIX: str = ""
 
-    def compute_args(self, params: OpenAIVisionBaseParameters, source: UploadFile
+    def compute_args(self, params: OpenAIVisionBaseParameters, source: UploadFile, kind
                      ) -> Dict[str, Any]:
         data = source.file.read()
         rv = base64.b64encode(data)
+        if kind.mime.startswith("image"):
+            binary_block = {
+                "type": "image_url",
+                "image_url": {
+                    "url": f"data:image/jpeg;base64,{rv.decode('utf-8')}"
+                }
+            }
+        else:
+            binary_block = {
+                "type": "file",
+                "file": {
+                    "filename": source.filename,
+                    "file_data": f"data:application/pdf;base64,{rv.decode('utf-8')}"}
+            }
         messages = [{"role": "system", "content": params.system_prompt}] if params.system_prompt is not None else []
         messages.append({"role": "user",
                          "content": [
@@ -137,12 +155,8 @@ class OpenAIVisionConverterBase(ConverterBase):
                                  "type": "text",
                                  "text": params.prompt
                              },
-                             {
-                                 "type": "image_url",
-                                 "image_url": {
-                                     "url": f"data:image/jpeg;base64,{rv.decode('utf-8')}"
-                                 }
-                             }]})
+                             binary_block
+                         ]})
         kwargs = {
             'model': params.model_str,
             'messages': messages,
@@ -156,11 +170,18 @@ class OpenAIVisionConverterBase(ConverterBase):
         return kwargs
 
     def compute_result(self, **kwargs):
+        pattern: Pattern = re.compile(r"```(?:markdown\s+)?(\W.*?)```", re.DOTALL)
+        """Regex pattern to parse the output."""
         response = openai_chat_completion(self.PREFIX, **kwargs)
         contents = []
         for choice in response.choices:
             if choice.message.content:
-                contents.append(choice.message.content)
+                if "```" in choice.message.content:
+                    action_match = pattern.search(choice.message.content)
+                    if action_match is not None:
+                        contents.append(action_match.group(1).strip())
+                else:
+                    contents.append(choice.message.content)
         if contents:
             result = "\n".join(contents)
         return result
@@ -178,9 +199,9 @@ class OpenAIVisionConverterBase(ConverterBase):
         try:
             kind = filetype.guess(source.file)
             source.file.seek(0)
-            if kind.mime.startswith("image"):
+            if kind.mime.startswith("image") or kind.mime.endswith("pdf"):
                 result = None
-                kwargs = self.compute_args(params, source)
+                kwargs = self.compute_args(params, source, kind)
                 if kwargs['model'] != NO_DEPLOYED_MODELS:
                     result = self.compute_result(**kwargs)
                 if result:
@@ -189,7 +210,7 @@ class OpenAIVisionConverterBase(ConverterBase):
         except BaseException as err:
             raise err
         if doc is None:
-            raise TypeError(f"Conversion of audio file {source.filename} failed")
+            raise TypeError(f"Conversion of file {source.filename} failed")
         return [doc]
 
     @classmethod

{pyconverters_openai_vision-0.5.3.dist-info → pyconverters_openai_vision-0.5.4.dist-info}/METADATA

@@ -1,6 +1,6 @@
-Metadata-Version: 2.1
+Metadata-Version: 2.4
 Name: pyconverters-openai_vision
-Version: 0.5.3
+Version: 0.5.4
 Summary: OpenAIVision converter
 Home-page: https://kairntech.com/
 Author: Olivier Terrier

pyconverters_openai_vision-0.5.4.dist-info/RECORD

@@ -0,0 +1,7 @@
+pyconverters_openai_vision/__init__.py,sha256=gg9uPdn4gUxA8hQ1hvO_tebG_dq6EzNtTSggy4MA9R0,51
+pyconverters_openai_vision/openai_utils.py,sha256=HRJ6sJg88en66gkQbOpQKh7cbwtfoAwVLNh7JQSA9ps,5014
+pyconverters_openai_vision/openai_vision.py,sha256=ACpeOEFNphqKceQqyWHwpP6PvuOOig8qTTHwImq34j8,10445
+pyconverters_openai_vision-0.5.4.dist-info/entry_points.txt,sha256=-DS1gRUTf08Fjb79S_8sqCaqxBifC3q3EJZqXXdcf7Q,197
+pyconverters_openai_vision-0.5.4.dist-info/WHEEL,sha256=G2gURzTEtmeR8nrdXUJfNiB3VYVxigPQ-bEQujpNiNs,82
+pyconverters_openai_vision-0.5.4.dist-info/METADATA,sha256=FMB-_MGpLRxKMKyjr5InU-P6VJRtB-4hixc12PF2Xyg,2635
+pyconverters_openai_vision-0.5.4.dist-info/RECORD,,

{pyconverters_openai_vision-0.5.3.dist-info → pyconverters_openai_vision-0.5.4.dist-info}/WHEEL

@@ -1,4 +1,4 @@
 Wheel-Version: 1.0
-Generator: flit 3.9.0
+Generator: flit 3.12.0
 Root-Is-Purelib: true
 Tag: py3-none-any

pyconverters_openai_vision-0.5.3.dist-info/RECORD

@@ -1,7 +0,0 @@
-pyconverters_openai_vision/__init__.py,sha256=3OKw8raUFiHJX-jh8L1RvKEFeD7DZW5-8hpnp21EK-A,51
-pyconverters_openai_vision/openai_utils.py,sha256=HRJ6sJg88en66gkQbOpQKh7cbwtfoAwVLNh7JQSA9ps,5014
-pyconverters_openai_vision/openai_vision.py,sha256=MoEyjYCZ75jWiMonaQdR1erpFBpWgYEiqabp9D8102Y,9562
-pyconverters_openai_vision-0.5.3.dist-info/entry_points.txt,sha256=-DS1gRUTf08Fjb79S_8sqCaqxBifC3q3EJZqXXdcf7Q,197
-pyconverters_openai_vision-0.5.3.dist-info/WHEEL,sha256=EZbGkh7Ie4PoZfRQ8I0ZuP9VklN_TvcZ6DSE5Uar4z4,81
-pyconverters_openai_vision-0.5.3.dist-info/METADATA,sha256=PJWwNjeSChCjjJQhgWxR1RIn9jA7-HvcPvKHWT7sJjk,2635
-pyconverters_openai_vision-0.5.3.dist-info/RECORD,,