Skip to content

VLM Engines Reference

vlm4ocr.vlm_engines.BasicVLMConfig 

BasicVLMConfig(
    max_new_tokens: int = 2048,
    temperature: float = 0.0,
    **kwargs
)

Bases: VLMConfig

The basic VLM configuration for most non-reasoning models.

Source code in packages/vlm4ocr/vlm4ocr/vlm_engines.py 
57
58
59
60
61
62
63
64
65
def __init__(self, max_new_tokens:int=2048, temperature:float=0.0, **kwargs):
    """
    The basic VLM configuration for most non-reasoning models.
    """
    super().__init__(**kwargs)
    self.max_new_tokens = max_new_tokens
    self.temperature = temperature
    self.params["max_new_tokens"] = self.max_new_tokens
    self.params["temperature"] = self.temperature

preprocess_messages 

preprocess_messages(
    messages: List[Dict[str, str]],
) -> List[Dict[str, str]]

This method preprocesses the input messages before passing them to the VLM.

Parameters:

messages : List[Dict[str,str]] a list of dict with role and content. role must be one of {"system", "user", "assistant"}

Returns:

messages : List[Dict[str,str]] a list of dict with role and content. role must be one of {"system", "user", "assistant"}

Source code in packages/vlm4ocr/vlm4ocr/vlm_engines.py 
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
def preprocess_messages(self, messages:List[Dict[str,str]]) -> List[Dict[str,str]]:
    """
    This method preprocesses the input messages before passing them to the VLM.

    Parameters:
    ----------
    messages : List[Dict[str,str]]
        a list of dict with role and content. role must be one of {"system", "user", "assistant"}

    Returns:
    -------
    messages : List[Dict[str,str]]
        a list of dict with role and content. role must be one of {"system", "user", "assistant"}
    """
    return messages

postprocess_response 

postprocess_response(
    response: Union[
        str, Dict[str, str], Generator[str, None, None]
    ],
) -> Union[
    Dict[str, str], Generator[Dict[str, str], None, None]
]

This method postprocesses the VLM response after it is generated.

Parameters:

response : Union[str, Generator[str, None, None]] the VLM response. Can be a string or a generator.

Returns: Union[str, Generator[Dict[str, str], None, None]] the postprocessed VLM response. if input is a generator, the output will be a generator {"type": "response", "data": }.

Source code in packages/vlm4ocr/vlm4ocr/vlm_engines.py 
 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
def postprocess_response(self, response:Union[str, Dict[str, str], Generator[str, None, None]]) -> Union[Dict[str, str], Generator[Dict[str, str], None, None]]:
    """
    This method postprocesses the VLM response after it is generated.

    Parameters:
    ----------
    response : Union[str, Generator[str, None, None]]
        the VLM response. Can be a string or a generator.

    Returns: Union[str, Generator[Dict[str, str], None, None]]
        the postprocessed VLM response. 
        if input is a generator, the output will be a generator {"type": "response", "data": <content>}.
    """
    if isinstance(response, str):
        return {"response": response}

    elif isinstance(response, dict):
        if "response" in response:
            return response
        else:
            warnings.warn(f"Invalid response dict keys: {response.keys()}. Returning default empty dict.", UserWarning)
            return {"response": ""}

    def _process_stream():
        for chunk in response:
            if isinstance(chunk, dict):
                yield chunk
            elif isinstance(chunk, str):
                yield {"type": "response", "data": chunk}

    return _process_stream()

vlm4ocr.vlm_engines.ReasoningVLMConfig 

ReasoningVLMConfig(
    thinking_token_start="<think>",
    thinking_token_end="</think>",
    **kwargs
)

Bases: VLMConfig

The general configuration for reasoning vision models.

Source code in packages/vlm4ocr/vlm4ocr/vlm_engines.py 
116
117
118
119
120
121
122
def __init__(self, thinking_token_start="<think>", thinking_token_end="</think>", **kwargs):
    """
    The general configuration for reasoning vision models.
    """
    super().__init__(**kwargs)
    self.thinking_token_start = thinking_token_start
    self.thinking_token_end = thinking_token_end

preprocess_messages 

preprocess_messages(
    messages: List[Dict[str, str]],
) -> List[Dict[str, str]]

This method preprocesses the input messages before passing them to the VLM.

Parameters:

messages : List[Dict[str,str]] a list of dict with role and content. role must be one of {"system", "user", "assistant"}

Returns:

messages : List[Dict[str,str]] a list of dict with role and content. role must be one of {"system", "user", "assistant"}

Source code in packages/vlm4ocr/vlm4ocr/vlm_engines.py 
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
def preprocess_messages(self, messages:List[Dict[str,str]]) -> List[Dict[str,str]]:
    """
    This method preprocesses the input messages before passing them to the VLM.

    Parameters:
    ----------
    messages : List[Dict[str,str]]
        a list of dict with role and content. role must be one of {"system", "user", "assistant"}

    Returns:
    -------
    messages : List[Dict[str,str]]
        a list of dict with role and content. role must be one of {"system", "user", "assistant"}
    """
    return messages.copy()

postprocess_response 

postprocess_response(
    response: Union[
        str, Dict[str, str], Generator[str, None, None]
    ],
) -> Union[
    Dict[str, str], Generator[Dict[str, str], None, None]
]

This method postprocesses the VLM response after it is generated. 1. If input is a string, it will extract the reasoning and response based on the thinking tokens. 2. If input is a dict, it should contain keys "reasoning" and "response". This is for inference engines that already parse reasoning and response. 3. If input is a generator, a. if the chunk is a dict, it should contain keys "type" and "data". This is for inference engines that already parse reasoning and response. b. if the chunk is a string, it will yield dicts with keys "type" and "data" based on the thinking tokens.

Parameters:

response : Union[str, Generator[str, None, None]] the VLM response. Can be a string or a generator.

Returns:

response : Union[str, Generator[str, None, None]] the postprocessed LLM response as a dict {"reasoning": , "response": } if input is a generator, the output will be a generator {"type": , "data": }.

Source code in packages/vlm4ocr/vlm4ocr/vlm_engines.py 
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
def postprocess_response(self, response:Union[str, Dict[str, str], Generator[str, None, None]]) -> Union[Dict[str,str], Generator[Dict[str,str], None, None]]:
    """
    This method postprocesses the VLM response after it is generated.
    1. If input is a string, it will extract the reasoning and response based on the thinking tokens.
    2. If input is a dict, it should contain keys "reasoning" and "response". This is for inference engines that already parse reasoning and response.
    3. If input is a generator, 
        a. if the chunk is a dict, it should contain keys "type" and "data". This is for inference engines that already parse reasoning and response.
        b. if the chunk is a string, it will yield dicts with keys "type" and "data" based on the thinking tokens.

    Parameters:
    ----------
    response : Union[str, Generator[str, None, None]]
        the VLM response. Can be a string or a generator.

    Returns:
    -------
    response : Union[str, Generator[str, None, None]]
        the postprocessed LLM response as a dict {"reasoning": <reasoning>, "response": <content>}
        if input is a generator, the output will be a generator {"type": <reasoning or response>, "data": <content>}.
    """
    if isinstance(response, str):
        # get contents between thinking_token_start and thinking_token_end
        pattern = f"{re.escape(self.thinking_token_start)}(.*?){re.escape(self.thinking_token_end)}"
        match = re.search(pattern, response, re.DOTALL)
        reasoning = match.group(1) if match else ""
        # get response AFTER thinking_token_end
        response = re.sub(f".*?{self.thinking_token_end}", "", response, flags=re.DOTALL).strip()
        return {"reasoning": reasoning, "response": response}

    elif isinstance(response, dict):
        if "reasoning" in response and "response" in response:
            return response
        else:
            warnings.warn(f"Invalid response dict keys: {response.keys()}. Returning default empty dict.", UserWarning)
            return {"reasoning": "", "response": ""}

    elif isinstance(response, Generator):
        def _process_stream():
            think_flag = False
            buffer = ""
            for chunk in response:
                if isinstance(chunk, dict):
                    yield chunk

                elif isinstance(chunk, str):
                    buffer += chunk
                    # switch between reasoning and response
                    if self.thinking_token_start in buffer:
                        think_flag = True
                        buffer = buffer.replace(self.thinking_token_start, "")
                    elif self.thinking_token_end in buffer:
                        think_flag = False
                        buffer = buffer.replace(self.thinking_token_end, "")

                    # if chunk is in thinking block, tag it as reasoning; else tag it as response
                    if chunk not in [self.thinking_token_start, self.thinking_token_end]:
                        if think_flag:
                            yield {"type": "reasoning", "data": chunk}
                        else:
                            yield {"type": "response", "data": chunk}

        return _process_stream()

    else:
        warnings.warn(f"Invalid response type: {type(response)}. Returning default empty dict.", UserWarning)
        return {"reasoning": "", "response": ""}

vlm4ocr.vlm_engines.OpenAIReasoningVLMConfig 

OpenAIReasoningVLMConfig(
    reasoning_effort: str = "low", **kwargs
)

Bases: ReasoningVLMConfig

The OpenAI "o" series configuration. 1. The reasoning effort is set to "low" by default. 2. The temperature parameter is not supported and will be ignored. 3. The system prompt is not supported and will be concatenated to the next user prompt.

Parameters:

reasoning_effort : str, Optional the reasoning effort. Must be one of {"low", "medium", "high"}. Default is "low".

Source code in packages/vlm4ocr/vlm4ocr/vlm_engines.py 
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
def __init__(self, reasoning_effort:str="low", **kwargs):
    """
    The OpenAI "o" series configuration.
    1. The reasoning effort is set to "low" by default.
    2. The temperature parameter is not supported and will be ignored.
    3. The system prompt is not supported and will be concatenated to the next user prompt.

    Parameters:
    ----------
    reasoning_effort : str, Optional
        the reasoning effort. Must be one of {"low", "medium", "high"}. Default is "low".
    """
    super().__init__(**kwargs)
    if reasoning_effort not in ["low", "medium", "high"]:
        raise ValueError("reasoning_effort must be one of {'low', 'medium', 'high'}.")

    self.reasoning_effort = reasoning_effort
    self.params["reasoning_effort"] = self.reasoning_effort

    if "temperature" in self.params:
        warnings.warn("Reasoning models do not support temperature parameter. Will be ignored.", UserWarning)
        self.params.pop("temperature")

preprocess_messages 

preprocess_messages(
    messages: List[Dict[str, str]],
) -> List[Dict[str, str]]

Concatenate system prompts to the next user prompt.

Parameters:

messages : List[Dict[str,str]] a list of dict with role and content. role must be one of {"system", "user", "assistant"}

Returns:

messages : List[Dict[str,str]] a list of dict with role and content. role must be one of {"system", "user", "assistant"}

Source code in packages/vlm4ocr/vlm4ocr/vlm_engines.py 
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
def preprocess_messages(self, messages:List[Dict[str,str]]) -> List[Dict[str,str]]:
    """
    Concatenate system prompts to the next user prompt.

    Parameters:
    ----------
    messages : List[Dict[str,str]]
        a list of dict with role and content. role must be one of {"system", "user", "assistant"}

    Returns:
    -------
    messages : List[Dict[str,str]]
        a list of dict with role and content. role must be one of {"system", "user", "assistant"}
    """
    system_prompt_holder = ""
    new_messages = []
    for i, message in enumerate(messages):
        # if system prompt, store it in system_prompt_holder
        if message['role'] == 'system':
            system_prompt_holder = message['content']
        # if user prompt, concatenate it with system_prompt_holder
        elif message['role'] == 'user':
            if system_prompt_holder:
                new_message = {'role': message['role'], 'content': f"{system_prompt_holder} {message['content']}"}
                system_prompt_holder = ""
            else:
                new_message = {'role': message['role'], 'content': message['content']}

            new_messages.append(new_message)
        # if assistant/other prompt, do nothing
        else:
            new_message = {'role': message['role'], 'content': message['content']}
            new_messages.append(new_message)

    return new_messages

vlm4ocr.vlm_engines.OpenAICompatibleVLMEngine 

OpenAICompatibleVLMEngine(
    model: str,
    api_key: str,
    base_url: str,
    config: VLMConfig = None,
    **kwrs
)

Bases: VLMEngine

General OpenAI-compatible server inference engine. https://docs.vllm.ai/en/latest/serving/openai_compatible_server.html

For parameters and documentation, refer to https://platform.openai.com/docs/api-reference/introduction

Parameters:

model_name : str model name as shown in the vLLM server api_key : str the API key for the vLLM server. base_url : str the base url for the vLLM server. config : LLMConfig the LLM configuration.

Source code in packages/vlm4ocr/vlm4ocr/vlm_engines.py 
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
def __init__(self, model:str, api_key:str, base_url:str, config:VLMConfig=None, **kwrs):
    """
    General OpenAI-compatible server inference engine.
    https://docs.vllm.ai/en/latest/serving/openai_compatible_server.html

    For parameters and documentation, refer to https://platform.openai.com/docs/api-reference/introduction

    Parameters:
    ----------
    model_name : str
        model name as shown in the vLLM server
    api_key : str
        the API key for the vLLM server.
    base_url : str
        the base url for the vLLM server. 
    config : LLMConfig
        the LLM configuration.
    """
    if importlib.util.find_spec("openai") is None:
        raise ImportError("OpenAI Python API library not found. Please install OpanAI (```pip install openai```).")

    from openai import OpenAI, AsyncOpenAI
    from openai.types.chat import ChatCompletionChunk
    self.ChatCompletionChunk = ChatCompletionChunk
    super().__init__(config)
    self.client = OpenAI(api_key=api_key, base_url=base_url, **kwrs)
    self.async_client = AsyncOpenAI(api_key=api_key, base_url=base_url, **kwrs)
    self.model = model
    self.config = config if config else BasicVLMConfig()
    self.formatted_params = self._format_config()

chat 

chat(
    messages: List[Dict[str, str]],
    verbose: bool = False,
    stream: bool = False,
    messages_logger: MessagesLogger = None,
) -> Union[
    Dict[str, str], Generator[Dict[str, str], None, None]
]

This method inputs chat messages and outputs LLM generated text.

Parameters:

messages : List[Dict[str,str]] a list of dict with role and content. role must be one of {"system", "user", "assistant"} verbose : bool, Optional if True, VLM generated text will be printed in terminal in real-time. stream : bool, Optional if True, returns a generator that yields the output in real-time. messages_logger : MessagesLogger, Optional the message logger that logs the chat messages.

Returns:

response : Union[Dict[str,str], Generator[Dict[str, str], None, None]] a dict {"reasoning": , "response": } or Generator {"type": , "data": }

Source code in packages/vlm4ocr/vlm4ocr/vlm_engines.py 
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
def chat(self, messages:List[Dict[str,str]], verbose:bool=False, stream:bool=False, 
         messages_logger:MessagesLogger=None) -> Union[Dict[str, str], Generator[Dict[str, str], None, None]]:
    """
    This method inputs chat messages and outputs LLM generated text.

    Parameters:
    ----------
    messages : List[Dict[str,str]]
        a list of dict with role and content. role must be one of {"system", "user", "assistant"}
    verbose : bool, Optional
        if True, VLM generated text will be printed in terminal in real-time.
    stream : bool, Optional
        if True, returns a generator that yields the output in real-time.
    messages_logger : MessagesLogger, Optional
        the message logger that logs the chat messages.

    Returns:
    -------
    response : Union[Dict[str,str], Generator[Dict[str, str], None, None]]
        a dict {"reasoning": <reasoning>, "response": <response>} or Generator {"type": <reasoning or response>, "data": <content>}
    """
    processed_messages = self.config.preprocess_messages(messages)

    if stream:
        def _stream_generator():
            response_stream = self.client.chat.completions.create(
                                    model=self.model,
                                    messages=processed_messages,
                                    stream=True,
                                    **self.formatted_params
                                )
            res_text = ""
            for chunk in response_stream:
                if len(chunk.choices) > 0:
                    chunk_dict = self._format_response(chunk)
                    yield chunk_dict

                    res_text += chunk_dict["data"]
                    if chunk.choices[0].finish_reason == "length":
                        warnings.warn("Model stopped generating due to context length limit.", RuntimeWarning)

            # Postprocess response
            res_dict = self.config.postprocess_response(res_text)
            # Write to messages log
            if messages_logger:
                # replace images content with a placeholder "[image]" to save space
                for messages in processed_messages:
                    if "content" in messages and isinstance(messages["content"], list):
                        for content in messages["content"]:
                            if isinstance(content, dict) and content.get("type") == "image_url":
                                content["image_url"]["url"] = "[image]"

                processed_messages.append({"role": "assistant",
                                            "content": res_dict.get("response", ""),
                                            "reasoning": res_dict.get("reasoning", "")})
                messages_logger.log_messages(processed_messages)

        return self.config.postprocess_response(_stream_generator())

    elif verbose:
        response = self.client.chat.completions.create(
            model=self.model,
            messages=processed_messages,
            stream=True,
            **self.formatted_params
        )
        res = {"reasoning": "", "response": ""}
        phase = ""
        for chunk in response:
            if len(chunk.choices) > 0:
                chunk_dict = self._format_response(chunk)
                chunk_text = chunk_dict["data"]
                res[chunk_dict["type"]] += chunk_text
                if phase != chunk_dict["type"] and chunk_text != "":
                    print(f"\n--- {chunk_dict['type'].capitalize()} ---")
                    phase = chunk_dict["type"]

                print(chunk_text, end="", flush=True)
                if chunk.choices[0].finish_reason == "length":
                    warnings.warn("Model stopped generating due to context length limit.", RuntimeWarning)

        print('\n')

    else:
        response = self.client.chat.completions.create(
            model=self.model,
            messages=processed_messages,
            stream=False,
            **self.formatted_params
        )
        res = self._format_response(response)

        if response.choices[0].finish_reason == "length":
            warnings.warn("Model stopped generating due to context length limit.", RuntimeWarning)

    # Postprocess response
    res_dict = self.config.postprocess_response(res)
    # Write to messages log
    if messages_logger:
        # replace images content with a placeholder "[image]" to save space
        for messages in processed_messages:
            if "content" in messages and isinstance(messages["content"], list):
                for content in messages["content"]:
                    if isinstance(content, dict) and content.get("type") == "image_url":
                        content["image_url"]["url"] = "[image]"

        processed_messages.append({"role": "assistant", 
                                "content": res_dict.get("response", ""), 
                                "reasoning": res_dict.get("reasoning", "")})
        messages_logger.log_messages(processed_messages)

    return res_dict

chat_async async 

chat_async(
    messages: List[Dict[str, str]],
    messages_logger: MessagesLogger = None,
) -> Dict[str, str]

Async version of chat method. Streaming is not supported.

Source code in packages/vlm4ocr/vlm4ocr/vlm_engines.py 
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
async def chat_async(self, messages:List[Dict[str,str]], messages_logger:MessagesLogger=None) -> Dict[str,str]:
    """
    Async version of chat method. Streaming is not supported.
    """
    processed_messages = self.config.preprocess_messages(messages)

    response = await self.async_client.chat.completions.create(
        model=self.model,
        messages=processed_messages,
        stream=False,
        **self.formatted_params
    )

    if response.choices[0].finish_reason == "length":
        warnings.warn("Model stopped generating due to context length limit.", RuntimeWarning)

    res = self._format_response(response)

    # Postprocess response
    res_dict = self.config.postprocess_response(res)
    # Write to messages log
    if messages_logger:
        # replace images content with a placeholder "[image]" to save space
        for messages in processed_messages:
            if "content" in messages and isinstance(messages["content"], list):
                for content in messages["content"]:
                    if isinstance(content, dict) and content.get("type") == "image_url":
                        content["image_url"]["url"] = "[image]"

        processed_messages.append({"role": "assistant", 
                                    "content": res_dict.get("response", ""), 
                                    "reasoning": res_dict.get("reasoning", "")})
        messages_logger.log_messages(processed_messages)

    return res_dict

get_ocr_messages 

get_ocr_messages(
    system_prompt: str,
    user_prompt: str,
    image: Image,
    format: str = "png",
    detail: str = "high",
    few_shot_examples: List[FewShotExample] = None,
) -> List[Dict[str, str]]

This method inputs an image and returns the correesponding chat messages for the inference engine.

Parameters:

system_prompt : str the system prompt. user_prompt : str the user prompt. image : Image.Image the image for OCR. format : str, Optional the image format. detail : str, Optional the detail level of the image. Default is "high". few_shot_examples : List[FewShotExample], Optional list of few-shot examples.

Source code in packages/vlm4ocr/vlm4ocr/vlm_engines.py 
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
def get_ocr_messages(self, system_prompt:str, user_prompt:str, image:Image.Image, format:str='png', 
                     detail:str="high", few_shot_examples:List[FewShotExample]=None) -> List[Dict[str,str]]:
    """
    This method inputs an image and returns the correesponding chat messages for the inference engine.

    Parameters:
    ----------
    system_prompt : str
        the system prompt.
    user_prompt : str
        the user prompt.
    image : Image.Image
        the image for OCR.
    format : str, Optional
        the image format. 
    detail : str, Optional
        the detail level of the image. Default is "high". 
    few_shot_examples : List[FewShotExample], Optional
        list of few-shot examples.
    """
    base64_str = image_to_base64(image)
    output_messages = []
    # system message
    system_message = {"role": "system", "content": system_prompt}
    output_messages.append(system_message)

    # few-shot examples
    if few_shot_examples is not None:
        for example in few_shot_examples:
            if not isinstance(example, FewShotExample):
                raise ValueError("Few-shot example must be a FewShotExample object.")

            example_image_b64 = image_to_base64(example.image)
            example_user_message = {
                "role": "user",
                "content": [
                    {
                        "type": "image_url",
                        "image_url": {
                            "url": f"data:image/{format};base64,{example_image_b64}",
                            "detail": detail
                        },
                    },
                    {"type": "text", "text": user_prompt},
                ],
            }
            example_agent_message = {"role": "assistant", "content": example.text}
            output_messages.append(example_user_message)
            output_messages.append(example_agent_message)

    # user message
    user_message = {
        "role": "user",
        "content": [
            {
                "type": "image_url",
                "image_url": {
                    "url": f"data:image/{format};base64,{base64_str}",
                    "detail": detail
                },
            },
            {"type": "text", "text": user_prompt},
        ],
    }
    output_messages.append(user_message)
    return output_messages

vlm4ocr.vlm_engines.VLLMVLMEngine 

VLLMVLMEngine(
    model: str,
    api_key: str = "",
    base_url: str = "http://localhost:8000/v1",
    config: VLMConfig = None,
    **kwrs
)

Bases: OpenAICompatibleVLMEngine

vLLM OpenAI compatible server inference engine. https://docs.vllm.ai/en/latest/serving/openai_compatible_server.html

For parameters and documentation, refer to https://platform.openai.com/docs/api-reference/introduction

Parameters:

model_name : str model name as shown in the vLLM server api_key : str, Optional the API key for the vLLM server. base_url : str, Optional the base url for the vLLM server. config : LLMConfig the LLM configuration.

Source code in packages/vlm4ocr/vlm4ocr/vlm_engines.py 
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
def __init__(self, model:str, api_key:str="", base_url:str="http://localhost:8000/v1", config:VLMConfig=None, **kwrs):
    """
    vLLM OpenAI compatible server inference engine.
    https://docs.vllm.ai/en/latest/serving/openai_compatible_server.html

    For parameters and documentation, refer to https://platform.openai.com/docs/api-reference/introduction

    Parameters:
    ----------
    model_name : str
        model name as shown in the vLLM server
    api_key : str, Optional
        the API key for the vLLM server.
    base_url : str, Optional
        the base url for the vLLM server. 
    config : LLMConfig
        the LLM configuration.
    """
    super().__init__(model, api_key, base_url, config, **kwrs)

vlm4ocr.vlm_engines.OpenRouterVLMEngine 

OpenRouterVLMEngine(
    model: str,
    api_key: str = None,
    base_url: str = "https://openrouter.ai/api/v1",
    config: VLMConfig = None,
    **kwrs
)

Bases: OpenAICompatibleVLMEngine

OpenRouter OpenAI-compatible server inference engine.

Parameters:

model_name : str model name as shown in the vLLM server api_key : str, Optional the API key for the vLLM server. If None, will use the key in os.environ['OPENROUTER_API_KEY']. base_url : str, Optional the base url for the vLLM server. config : LLMConfig the LLM configuration.

Source code in packages/vlm4ocr/vlm4ocr/vlm_engines.py 
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
def __init__(self, model:str, api_key:str=None, base_url:str="https://openrouter.ai/api/v1", config:VLMConfig=None, **kwrs):
    """
    OpenRouter OpenAI-compatible server inference engine.

    Parameters:
    ----------
    model_name : str
        model name as shown in the vLLM server
    api_key : str, Optional
        the API key for the vLLM server. If None, will use the key in os.environ['OPENROUTER_API_KEY'].
    base_url : str, Optional
        the base url for the vLLM server. 
    config : LLMConfig
        the LLM configuration.
    """
    self.api_key = api_key
    if self.api_key is None:
        self.api_key = os.getenv("OPENROUTER_API_KEY")
    super().__init__(model, self.api_key, base_url, config, **kwrs)

vlm4ocr.vlm_engines.OllamaVLMEngine 

OllamaVLMEngine(
    model_name: str,
    num_ctx: int = 8192,
    keep_alive: int = 300,
    config: VLMConfig = None,
    **kwrs
)

Bases: VLMEngine

The Ollama inference engine.

Parameters:

model_name : str the model name exactly as shown in >> ollama ls num_ctx : int, Optional context length that LLM will evaluate. keep_alive : int, Optional seconds to hold the LLM after the last API call. config : LLMConfig the LLM configuration.

Source code in packages/vlm4ocr/vlm4ocr/vlm_engines.py 
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
def __init__(self, model_name:str, num_ctx:int=8192, keep_alive:int=300, config:VLMConfig=None, **kwrs):
    """
    The Ollama inference engine.

    Parameters:
    ----------
    model_name : str
        the model name exactly as shown in >> ollama ls
    num_ctx : int, Optional
        context length that LLM will evaluate.
    keep_alive : int, Optional
        seconds to hold the LLM after the last API call.
    config : LLMConfig
        the LLM configuration. 
    """
    if importlib.util.find_spec("ollama") is None:
        raise ImportError("ollama-python not found. Please install ollama-python (```pip install ollama```).")

    from ollama import Client, AsyncClient
    self.client = Client(**kwrs)
    self.async_client = AsyncClient(**kwrs)
    self.model_name = model_name
    self.num_ctx = num_ctx
    self.keep_alive = keep_alive
    self.config = config if config else BasicVLMConfig()
    self.formatted_params = self._format_config()

chat 

chat(
    messages: List[Dict[str, str]],
    verbose: bool = False,
    stream: bool = False,
    messages_logger: MessagesLogger = None,
) -> Union[
    Dict[str, str], Generator[Dict[str, str], None, None]
]

This method inputs chat messages and outputs VLM generated text.

Parameters:

messages : List[Dict[str,str]] a list of dict with role and content. role must be one of {"system", "user", "assistant"} verbose : bool, Optional if True, VLM generated text will be printed in terminal in real-time. stream : bool, Optional if True, returns a generator that yields the output in real-time. Messages_logger : MessagesLogger, Optional the message logger that logs the chat messages.

Returns:

response : Union[Dict[str,str], Generator[Dict[str, str], None, None]] a dict {"reasoning": , "response": } or Generator {"type": , "data": }

Source code in packages/vlm4ocr/vlm4ocr/vlm_engines.py 
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
def chat(self, messages:List[Dict[str,str]], verbose:bool=False, stream:bool=False, 
         messages_logger:MessagesLogger=None) -> Union[Dict[str,str], Generator[Dict[str, str], None, None]]:
    """
    This method inputs chat messages and outputs VLM generated text.

    Parameters:
    ----------
    messages : List[Dict[str,str]]
        a list of dict with role and content. role must be one of {"system", "user", "assistant"}
    verbose : bool, Optional
        if True, VLM generated text will be printed in terminal in real-time.
    stream : bool, Optional
        if True, returns a generator that yields the output in real-time.
    Messages_logger : MessagesLogger, Optional
        the message logger that logs the chat messages.

    Returns:
    -------
    response : Union[Dict[str,str], Generator[Dict[str, str], None, None]]
        a dict {"reasoning": <reasoning>, "response": <response>} or Generator {"type": <reasoning or response>, "data": <content>}
    """
    processed_messages = self.config.preprocess_messages(messages)

    options={'num_ctx': self.num_ctx, **self.formatted_params}
    if stream:
        def _stream_generator():
            response_stream = self.client.chat(
                model=self.model_name, 
                messages=processed_messages, 
                options=options,
                stream=True, 
                keep_alive=self.keep_alive
            )
            res = {"reasoning": "", "response": ""}
            for chunk in response_stream:
                if hasattr(chunk.message, 'thinking') and chunk.message.thinking:
                    content_chunk = getattr(getattr(chunk, 'message', {}), 'thinking', '')
                    res["reasoning"] += content_chunk
                    yield {"type": "reasoning", "data": content_chunk}
                else:
                    content_chunk = getattr(getattr(chunk, 'message', {}), 'content', '')
                    res["response"] += content_chunk
                    yield {"type": "response", "data": content_chunk}

                if chunk.done_reason == "length":
                    warnings.warn("Model stopped generating due to context length limit.", RuntimeWarning)

            # Postprocess response
            res_dict = self.config.postprocess_response(res)
            # Write to messages log
            if messages_logger:
                # replace images content with a placeholder "[image]" to save space
                for messages in processed_messages:
                    if "images" in messages:
                        messages["images"] = ["[image]" for _ in messages["images"]]

                processed_messages.append({"role": "assistant",
                                            "content": res_dict.get("response", ""),
                                            "reasoning": res_dict.get("reasoning", "")})
                messages_logger.log_messages(processed_messages)

        return self.config.postprocess_response(_stream_generator())

    elif verbose:
        response = self.client.chat(
                        model=self.model_name, 
                        messages=processed_messages, 
                        options=options,
                        stream=True,
                        keep_alive=self.keep_alive
                    )

        res = {"reasoning": "", "response": ""}
        phase = ""
        for chunk in response:
            if hasattr(chunk.message, 'thinking') and chunk.message.thinking:
                if phase != "reasoning":
                    print("\n--- Reasoning ---")
                    phase = "reasoning"

                content_chunk = getattr(getattr(chunk, 'message', {}), 'thinking', '')
                res["reasoning"] += content_chunk
            else:
                if phase != "response":
                    print("\n--- Response ---")
                    phase = "response"
                content_chunk = getattr(getattr(chunk, 'message', {}), 'content', '')
                res["response"] += content_chunk

            print(content_chunk, end='', flush=True)

            if chunk.done_reason == "length":
                warnings.warn("Model stopped generating due to context length limit.", RuntimeWarning)
        print('\n')

    else:
        response = self.client.chat(
                            model=self.model_name, 
                            messages=processed_messages, 
                            options=options,
                            stream=False,
                            keep_alive=self.keep_alive
                        )
        res = {"reasoning": getattr(getattr(response, 'message', {}), 'thinking', ''),
               "response": getattr(getattr(response, 'message', {}), 'content', '')}

        if response.done_reason == "length":
            warnings.warn("Model stopped generating due to context length limit.", RuntimeWarning)

    # Postprocess response
    res_dict = self.config.postprocess_response(res)
    # Write to messages log
    if messages_logger:
        # replace images content with a placeholder "[image]" to save space
        for messages in processed_messages:
            if "images" in messages:
                messages["images"] = ["[image]" for _ in messages["images"]]

        processed_messages.append({"role": "assistant", 
                                "content": res_dict.get("response", ""), 
                                "reasoning": res_dict.get("reasoning", "")})
        messages_logger.log_messages(processed_messages)

    return res_dict

chat_async async 

chat_async(
    messages: List[Dict[str, str]],
    messages_logger: MessagesLogger = None,
) -> Dict[str, str]

Async version of chat method. Streaming is not supported.

Source code in packages/vlm4ocr/vlm4ocr/vlm_engines.py 
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
async def chat_async(self, messages:List[Dict[str,str]], messages_logger:MessagesLogger=None) -> Dict[str,str]:
    """
    Async version of chat method. Streaming is not supported.
    """
    processed_messages = self.config.preprocess_messages(messages)

    response = await self.async_client.chat(
                        model=self.model_name, 
                        messages=processed_messages, 
                        options={'num_ctx': self.num_ctx, **self.formatted_params},
                        stream=False,
                        keep_alive=self.keep_alive
                    )

    res = {"reasoning": getattr(getattr(response, 'message', {}), 'thinking', ''),
           "response": getattr(getattr(response, 'message', {}), 'content', '')}

    if response.done_reason == "length":
        warnings.warn("Model stopped generating due to context length limit.", RuntimeWarning)
    # Postprocess response
    res_dict = self.config.postprocess_response(res)
    # Write to messages log
    if messages_logger:
        # replace images content with a placeholder "[image]" to save space
        for messages in processed_messages:
            if "images" in messages:
                messages["images"] = ["[image]" for _ in messages["images"]]

        processed_messages.append({"role": "assistant", 
                                    "content": res_dict.get("response", ""), 
                                    "reasoning": res_dict.get("reasoning", "")})
        messages_logger.log_messages(processed_messages)

    return res_dict

get_ocr_messages 

get_ocr_messages(
    system_prompt: str,
    user_prompt: str,
    image: Image,
    few_shot_examples: List[FewShotExample] = None,
) -> List[Dict[str, str]]

This method inputs an image and returns the correesponding chat messages for the inference engine.

Parameters:

system_prompt : str the system prompt. user_prompt : str the user prompt. image : Image.Image the image for OCR. few_shot_examples : List[FewShotExample], Optional list of few-shot examples.

Source code in packages/vlm4ocr/vlm4ocr/vlm_engines.py 
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
def get_ocr_messages(self, system_prompt:str, user_prompt:str, image:Image.Image, few_shot_examples:List[FewShotExample]=None) -> List[Dict[str,str]]:
    """
    This method inputs an image and returns the correesponding chat messages for the inference engine.

    Parameters:
    ----------
    system_prompt : str
        the system prompt.
    user_prompt : str
        the user prompt.
    image : Image.Image
        the image for OCR.
    few_shot_examples : List[FewShotExample], Optional
        list of few-shot examples. 
    """
    base64_str = image_to_base64(image)
    output_messages = []
    # system message
    system_message = {"role": "system", "content": system_prompt}
    output_messages.append(system_message)

    # few-shot examples
    if few_shot_examples is not None:
        for example in few_shot_examples:
            if not isinstance(example, FewShotExample):
                raise ValueError("Few-shot example must be a FewShotExample object.")

            example_image_b64 = image_to_base64(example.image)
            example_user_message = {"role": "user", "content": user_prompt, "images": [example_image_b64]}
            example_agent_message = {"role": "assistant", "content": example.text}
            output_messages.append(example_user_message)
            output_messages.append(example_agent_message)

    # user message
    user_message = {"role": "user", "content": user_prompt, "images": [base64_str]}
    output_messages.append(user_message)

    return output_messages

vlm4ocr.vlm_engines.OpenAIVLMEngine 

OpenAIVLMEngine(
    model: str, config: VLMConfig = None, **kwrs
)

Bases: VLMEngine

The OpenAI API inference engine. Supports OpenAI models and OpenAI compatible servers: - vLLM OpenAI compatible server (https://docs.vllm.ai/en/latest/serving/openai_compatible_server.html)

For parameters and documentation, refer to https://platform.openai.com/docs/api-reference/introduction

Parameters:

model_name : str model name as described in https://platform.openai.com/docs/models config : VLMConfig, Optional the VLM configuration. Must be a child class of VLMConfig.

Source code in packages/vlm4ocr/vlm4ocr/vlm_engines.py 
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
def __init__(self, model:str, config:VLMConfig=None, **kwrs):
    """
    The OpenAI API inference engine. Supports OpenAI models and OpenAI compatible servers:
    - vLLM OpenAI compatible server (https://docs.vllm.ai/en/latest/serving/openai_compatible_server.html)

    For parameters and documentation, refer to https://platform.openai.com/docs/api-reference/introduction

    Parameters:
    ----------
    model_name : str
        model name as described in https://platform.openai.com/docs/models
    config : VLMConfig, Optional
        the VLM configuration. Must be a child class of VLMConfig.
    """
    if importlib.util.find_spec("openai") is None:
        raise ImportError("OpenAI Python API library not found. Please install OpanAI (```pip install openai```).")

    from openai import OpenAI, AsyncOpenAI
    self.client = OpenAI(**kwrs)
    self.async_client = AsyncOpenAI(**kwrs)
    self.model = model
    self.config = config if config else BasicVLMConfig()
    self.formatted_params = self._format_config()

chat 

chat(
    messages: List[Dict[str, str]],
    verbose: bool = False,
    stream: bool = False,
    messages_logger: MessagesLogger = None,
) -> Union[
    Dict[str, str], Generator[Dict[str, str], None, None]
]

This method inputs chat messages and outputs LLM generated text.

Parameters:

messages : List[Dict[str,str]] a list of dict with role and content. role must be one of {"system", "user", "assistant"} verbose : bool, Optional if True, VLM generated text will be printed in terminal in real-time. stream : bool, Optional if True, returns a generator that yields the output in real-time. messages_logger : MessagesLogger, Optional the message logger that logs the chat messages.

Returns:

response : Union[Dict[str,str], Generator[Dict[str, str], None, None]] a dict {"reasoning": , "response": } or Generator {"type": , "data": }

Source code in packages/vlm4ocr/vlm4ocr/vlm_engines.py 
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
def chat(self, messages:List[Dict[str,str]], verbose:bool=False, stream:bool=False, messages_logger:MessagesLogger=None) -> Union[Dict[str, str], Generator[Dict[str, str], None, None]]:
    """
    This method inputs chat messages and outputs LLM generated text.

    Parameters:
    ----------
    messages : List[Dict[str,str]]
        a list of dict with role and content. role must be one of {"system", "user", "assistant"}
    verbose : bool, Optional
        if True, VLM generated text will be printed in terminal in real-time.
    stream : bool, Optional
        if True, returns a generator that yields the output in real-time.
    messages_logger : MessagesLogger, Optional
        the message logger that logs the chat messages.

    Returns:
    -------
    response : Union[Dict[str,str], Generator[Dict[str, str], None, None]]
        a dict {"reasoning": <reasoning>, "response": <response>} or Generator {"type": <reasoning or response>, "data": <content>}
    """
    processed_messages = self.config.preprocess_messages(messages)

    if stream:
        def _stream_generator():
            response_stream = self.client.chat.completions.create(
                                    model=self.model,
                                    messages=processed_messages,
                                    stream=True,
                                    **self.formatted_params
                                )
            res_text = ""
            for chunk in response_stream:
                if len(chunk.choices) > 0:
                    chunk_text = chunk.choices[0].delta.content
                    if chunk_text is not None:
                        res_text += chunk_text
                        yield chunk_text
                    if chunk.choices[0].finish_reason == "length":
                        warnings.warn("Model stopped generating due to context length limit.", RuntimeWarning)

            # Postprocess response
            res_dict = self.config.postprocess_response(res_text)
            # Write to messages log
            if messages_logger:
                # replace images content with a placeholder "[image]" to save space
                for messages in processed_messages:
                    if "content" in messages and isinstance(messages["content"], list):
                        for content in messages["content"]:
                            if isinstance(content, dict) and content.get("type") == "image_url":
                                content["image_url"]["url"] = "[image]"

                processed_messages.append({"role": "assistant",
                                            "content": res_dict.get("response", ""),
                                            "reasoning": res_dict.get("reasoning", "")})
                messages_logger.log_messages(processed_messages)

        return self.config.postprocess_response(_stream_generator())

    elif verbose:
        response = self.client.chat.completions.create(
            model=self.model,
            messages=processed_messages,
            stream=True,
            **self.formatted_params
        )
        res = ''
        for chunk in response:
            if len(chunk.choices) > 0:
                if chunk.choices[0].delta.content is not None:
                    res += chunk.choices[0].delta.content
                    print(chunk.choices[0].delta.content, end="", flush=True)
                if chunk.choices[0].finish_reason == "length":
                    warnings.warn("Model stopped generating due to context length limit.", RuntimeWarning)

        print('\n')

    else:
        response = self.client.chat.completions.create(
            model=self.model,
            messages=processed_messages,
            stream=False,
            **self.formatted_params
        )
        res = response.choices[0].message.content

    # Postprocess response
    res_dict = self.config.postprocess_response(res)
    # Write to messages log
    if messages_logger:
        # replace images content with a placeholder "[image]" to save space
        for messages in processed_messages:
            if "content" in messages and isinstance(messages["content"], list):
                for content in messages["content"]:
                    if isinstance(content, dict) and content.get("type") == "image_url":
                        content["image_url"]["url"] = "[image]"

        processed_messages.append({"role": "assistant", 
                                "content": res_dict.get("response", ""), 
                                "reasoning": res_dict.get("reasoning", "")})
        messages_logger.log_messages(processed_messages)

    return res_dict

chat_async async 

chat_async(
    messages: List[Dict[str, str]],
    messages_logger: MessagesLogger = None,
) -> Dict[str, str]

Async version of chat method. Streaming is not supported.

Source code in packages/vlm4ocr/vlm4ocr/vlm_engines.py 
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
async def chat_async(self, messages:List[Dict[str,str]], messages_logger:MessagesLogger=None) -> Dict[str,str]:
    """
    Async version of chat method. Streaming is not supported.
    """
    processed_messages = self.config.preprocess_messages(messages)

    response = await self.async_client.chat.completions.create(
        model=self.model,
        messages=processed_messages,
        stream=False,
        **self.formatted_params
    )

    if response.choices[0].finish_reason == "length":
        warnings.warn("Model stopped generating due to context length limit.", RuntimeWarning)

    res = response.choices[0].message.content
    # Postprocess response
    res_dict = self.config.postprocess_response(res)
    # Write to messages log
    if messages_logger:
        # replace images content with a placeholder "[image]" to save space
        for messages in processed_messages:
            if "content" in messages and isinstance(messages["content"], list):
                for content in messages["content"]:
                    if isinstance(content, dict) and content.get("type") == "image_url":
                        content["image_url"]["url"] = "[image]"

        processed_messages.append({"role": "assistant", 
                                "content": res_dict.get("response", ""), 
                                "reasoning": res_dict.get("reasoning", "")})
        messages_logger.log_messages(processed_messages)

    return res_dict

get_ocr_messages 

get_ocr_messages(
    system_prompt: str,
    user_prompt: str,
    image: Image,
    format: str = "png",
    detail: str = "high",
    few_shot_examples: List[FewShotExample] = None,
) -> List[Dict[str, str]]

This method inputs an image and returns the correesponding chat messages for the inference engine.

Parameters:

system_prompt : str the system prompt. user_prompt : str the user prompt. image : Image.Image the image for OCR. format : str, Optional the image format. detail : str, Optional the detail level of the image. Default is "high". few_shot_examples : List[FewShotExample], Optional list of few-shot examples. Each example is a dict with keys "image" (PIL.Image.Image) and "text" (str).

Source code in packages/vlm4ocr/vlm4ocr/vlm_engines.py 
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
def get_ocr_messages(self, system_prompt:str, user_prompt:str, image:Image.Image, format:str='png', 
                     detail:str="high", few_shot_examples:List[FewShotExample]=None) -> List[Dict[str,str]]:
    """
    This method inputs an image and returns the correesponding chat messages for the inference engine.

    Parameters:
    ----------
    system_prompt : str
        the system prompt.
    user_prompt : str
        the user prompt.
    image : Image.Image
        the image for OCR.
    format : str, Optional
        the image format. 
    detail : str, Optional
        the detail level of the image. Default is "high". 
    few_shot_examples : List[FewShotExample], Optional
        list of few-shot examples. Each example is a dict with keys "image" (PIL.Image.Image) and "text" (str).
    """
    base64_str = image_to_base64(image)
    output_messages = []
    # system message
    system_message = {"role": "system", "content": system_prompt}
    output_messages.append(system_message)

    # few-shot examples
    if few_shot_examples is not None:
        for example in few_shot_examples:
            if not isinstance(example, FewShotExample):
                raise ValueError("Few-shot example must be a FewShotExample object.")

            example_image_b64 = image_to_base64(example.image)
            example_user_message = {
                "role": "user",
                "content": [
                    {
                        "type": "image_url",
                        "image_url": {
                            "url": f"data:image/{format};base64,{example_image_b64}",
                            "detail": detail
                        },
                    },
                    {"type": "text", "text": user_prompt},
                ],
            }
            example_agent_message = {"role": "assistant", "content": example.text}
            output_messages.append(example_user_message)
            output_messages.append(example_agent_message)

    # user message
    user_message = {
        "role": "user",
        "content": [
            {
                "type": "image_url",
                "image_url": {
                    "url": f"data:image/{format};base64,{base64_str}",
                    "detail": detail
                },
            },
            {"type": "text", "text": user_prompt},
        ],
    }
    output_messages.append(user_message)
    return output_messages

vlm4ocr.vlm_engines.AzureOpenAIVLMEngine 

AzureOpenAIVLMEngine(
    model: str,
    api_version: str,
    config: VLMConfig = None,
    **kwrs
)

Bases: OpenAIVLMEngine

The Azure OpenAI API inference engine. For parameters and documentation, refer to - https://azure.microsoft.com/en-us/products/ai-services/openai-service - https://learn.microsoft.com/en-us/azure/ai-services/openai/quickstart

Parameters:

model : str model name as described in https://platform.openai.com/docs/models api_version : str the Azure OpenAI API version config : LLMConfig the LLM configuration.

Source code in packages/vlm4ocr/vlm4ocr/vlm_engines.py 
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
def __init__(self, model:str, api_version:str, config:VLMConfig=None, **kwrs):
    """
    The Azure OpenAI API inference engine.
    For parameters and documentation, refer to 
    - https://azure.microsoft.com/en-us/products/ai-services/openai-service
    - https://learn.microsoft.com/en-us/azure/ai-services/openai/quickstart

    Parameters:
    ----------
    model : str
        model name as described in https://platform.openai.com/docs/models
    api_version : str
        the Azure OpenAI API version
    config : LLMConfig
        the LLM configuration.
    """
    if importlib.util.find_spec("openai") is None:
        raise ImportError("OpenAI Python API library not found. Please install OpanAI (```pip install openai```).")

    from openai import AzureOpenAI, AsyncAzureOpenAI
    self.model = model
    self.api_version = api_version
    self.client = AzureOpenAI(api_version=self.api_version, 
                              **kwrs)
    self.async_client = AsyncAzureOpenAI(api_version=self.api_version, 
                                         **kwrs)
    self.config = config if config else BasicVLMConfig()
    self.formatted_params = self._format_config()