|
| 1 | +from typing import Any, Dict, Optional, List, TYPE_CHECKING |
| 2 | +from aiconfig import ParameterizedModelParser, InferenceOptions |
| 3 | +from aiconfig.callback import CallbackEvent |
| 4 | +import torch |
| 5 | +from aiconfig.schema import Prompt, Output, ExecuteResult, Attachment |
| 6 | + |
| 7 | +from transformers import pipeline, Pipeline |
| 8 | + |
| 9 | +if TYPE_CHECKING: |
| 10 | + from aiconfig import AIConfigRuntime |
| 11 | + |
| 12 | + |
| 13 | +class HuggingFaceImage2TextTransformer(ParameterizedModelParser): |
| 14 | + def __init__(self): |
| 15 | + """ |
| 16 | + Returns: |
| 17 | + HuggingFaceImage2TextTransformer |
| 18 | +
|
| 19 | + Usage: |
| 20 | + 1. Create a new model parser object with the model ID of the model to use. |
| 21 | + parser = HuggingFaceImage2TextTransformer() |
| 22 | + 2. Add the model parser to the registry. |
| 23 | + config.register_model_parser(parser) |
| 24 | + """ |
| 25 | + super().__init__() |
| 26 | + self.pipelines: dict[str, Pipeline] = {} |
| 27 | + |
| 28 | + def id(self) -> str: |
| 29 | + """ |
| 30 | + Returns an identifier for the Model Parser |
| 31 | + """ |
| 32 | + return "HuggingFaceImage2TextTransformer" |
| 33 | + |
| 34 | + async def serialize( |
| 35 | + self, |
| 36 | + prompt_name: str, |
| 37 | + data: Any, |
| 38 | + ai_config: "AIConfigRuntime", |
| 39 | + parameters: Optional[Dict[str, Any]] = None, |
| 40 | + ) -> List[Prompt]: |
| 41 | + """ |
| 42 | + Defines how a prompt and model inference settings get serialized in the .aiconfig. |
| 43 | + Assume input in the form of input(s) being passed into an already constructed pipeline. |
| 44 | +
|
| 45 | + Args: |
| 46 | + prompt (str): The prompt to be serialized. |
| 47 | + data (Any): Model-specific inference settings to be serialized. |
| 48 | + ai_config (AIConfigRuntime): The AIConfig Runtime. |
| 49 | + parameters (Dict[str, Any], optional): Model-specific parameters. Defaults to None. |
| 50 | +
|
| 51 | + Returns: |
| 52 | + str: Serialized representation of the prompt and inference settings. |
| 53 | + """ |
| 54 | + await ai_config.callback_manager.run_callbacks( |
| 55 | + CallbackEvent( |
| 56 | + "on_serialize_start", |
| 57 | + __name__, |
| 58 | + { |
| 59 | + "prompt_name": prompt_name, |
| 60 | + "data": data, |
| 61 | + "parameters": parameters, |
| 62 | + }, |
| 63 | + ) |
| 64 | + ) |
| 65 | + |
| 66 | + prompts = [] |
| 67 | + |
| 68 | + if not isinstance(data, dict): |
| 69 | + raise ValueError("Invalid data type. Expected dict when serializing prompt data to aiconfig.") |
| 70 | + if data.get("inputs", None) is None: |
| 71 | + raise ValueError("Invalid data when serializing prompt to aiconfig. Input data must contain an inputs field.") |
| 72 | + |
| 73 | + prompt = Prompt( |
| 74 | + **{ |
| 75 | + "name": prompt_name, |
| 76 | + "input": {"attachments": [{"data": data["inputs"]}]}, |
| 77 | + "metadata": None, |
| 78 | + "outputs": None, |
| 79 | + } |
| 80 | + ) |
| 81 | + |
| 82 | + prompts.append(prompt) |
| 83 | + |
| 84 | + await ai_config.callback_manager.run_callbacks(CallbackEvent("on_serialize_complete", __name__, {"result": prompts})) |
| 85 | + return prompts |
| 86 | + |
| 87 | + async def deserialize( |
| 88 | + self, |
| 89 | + prompt: Prompt, |
| 90 | + aiconfig: "AIConfigRuntime", |
| 91 | + params: Optional[Dict[str, Any]] = {}, |
| 92 | + ) -> Dict[str, Any]: |
| 93 | + await aiconfig.callback_manager.run_callbacks(CallbackEvent("on_deserialize_start", __name__, {"prompt": prompt, "params": params})) |
| 94 | + |
| 95 | + # Build Completion data |
| 96 | + completion_params = self.get_model_settings(prompt, aiconfig) |
| 97 | + |
| 98 | + inputs = validate_and_retrieve_image_from_attachments(prompt) |
| 99 | + |
| 100 | + completion_params["inputs"] = inputs |
| 101 | + |
| 102 | + await aiconfig.callback_manager.run_callbacks(CallbackEvent("on_deserialize_complete", __name__, {"output": completion_params})) |
| 103 | + return completion_params |
| 104 | + |
| 105 | + async def run_inference(self, prompt: Prompt, aiconfig: "AIConfigRuntime", options: InferenceOptions, parameters: Dict[str, Any]) -> list[Output]: |
| 106 | + await aiconfig.callback_manager.run_callbacks( |
| 107 | + CallbackEvent( |
| 108 | + "on_run_start", |
| 109 | + __name__, |
| 110 | + {"prompt": prompt, "options": options, "parameters": parameters}, |
| 111 | + ) |
| 112 | + ) |
| 113 | + model_name = aiconfig.get_model_name(prompt) |
| 114 | + |
| 115 | + self.pipelines[model_name] = pipeline(task="image-to-text", model=model_name) |
| 116 | + |
| 117 | + captioner = self.pipelines[model_name] |
| 118 | + completion_data = await self.deserialize(prompt, aiconfig, parameters) |
| 119 | + inputs = completion_data.pop("inputs") |
| 120 | + model = completion_data.pop("model") |
| 121 | + response = captioner(inputs, **completion_data) |
| 122 | + |
| 123 | + output = ExecuteResult(output_type="execute_result", data=response, metadata={}) |
| 124 | + |
| 125 | + prompt.outputs = [output] |
| 126 | + await aiconfig.callback_manager.run_callbacks(CallbackEvent("on_run_complete", __name__, {"result": prompt.outputs})) |
| 127 | + return prompt.outputs |
| 128 | + |
| 129 | + def get_output_text(self, response: dict[str, Any]) -> str: |
| 130 | + raise NotImplementedError("get_output_text is not implemented for HuggingFaceImage2TextTransformer") |
| 131 | + |
| 132 | + |
| 133 | +def validate_attachment_type_is_image(attachment: Attachment): |
| 134 | + if not hasattr(attachment, "mime_type"): |
| 135 | + raise ValueError(f"Attachment has no mime type. Specify the image mimetype in the aiconfig") |
| 136 | + |
| 137 | + if not attachment.mime_type.startswith("image/"): |
| 138 | + raise ValueError(f"Invalid attachment mimetype {attachment.mime_type}. Expected image mimetype.") |
| 139 | + |
| 140 | + |
| 141 | +def validate_and_retrieve_image_from_attachments(prompt: Prompt) -> list[str]: |
| 142 | + """ |
| 143 | + Retrieves the image uri's from each attachment in the prompt input. |
| 144 | +
|
| 145 | + Throws an exception if |
| 146 | + - attachment is not image |
| 147 | + - attachment data is not a uri |
| 148 | + - no attachments are found |
| 149 | + - operation fails for any reason |
| 150 | + """ |
| 151 | + |
| 152 | + if not hasattr(prompt.input, "attachments") or len(prompt.input.attachments) == 0: |
| 153 | + raise ValueError(f"No attachments found in input for prompt {prompt.name}. Please add an image attachment to the prompt input.") |
| 154 | + |
| 155 | + image_uris: list[str] = [] |
| 156 | + |
| 157 | + for i, attachment in enumerate(prompt.input.attachments): |
| 158 | + validate_attachment_type_is_image(attachment) |
| 159 | + |
| 160 | + if not isinstance(attachment.data, str): |
| 161 | + # See todo above, but for now only support uri's |
| 162 | + raise ValueError(f"Attachment #{i} data is not a uri. Please specify a uri for the image attachment in prompt {prompt.name}.") |
| 163 | + |
| 164 | + image_uris.append(attachment.data) |
| 165 | + |
| 166 | + return image_uris |
0 commit comments