def get_model_for_behavior()

in optimum/exporters/openvino/model_configs.py [0:0]

• 70 lines of code
• 16 McCabe index (conditional complexity)

    def get_model_for_behavior(model, behavior: Union[str, Phi4MMConfigBehavior]):
        if isinstance(behavior, str) and not isinstance(behavior, Phi4MMConfigBehavior):
            behavior = Phi4MMConfigBehavior(behavior)

        if behavior == Phi4MMConfigBehavior.LANGUAGE:
            return model

        if behavior == Phi4MMConfigBehavior.VISION_EMBEDDINGS:
            vision_embeddings = model.model.embed_tokens_extend.image_embed
            vision_embeddings.config = model.model.embed_tokens_extend.image_embed.img_processor.config
            return model.model.embed_tokens_extend.image_embed

        if behavior == Phi4MMConfigBehavior.VISION_PROJECTION:
            vision_model = model.model.embed_tokens_extend.image_embed
            if hasattr(vision_model, "img_projection"):
                projection = vision_model.img_projection
            else:
                import torch

                projection = torch.nn.Sequential(
                    *[vision_model.img_projection_up, torch.nn.GELU(), vision_model.img_projection_down]
                )
            projection.config = vision_model.img_processor.config
            return projection

        if behavior == Phi4MMConfigBehavior.TEXT_EMBEDDINGS:
            if hasattr(model.model, "_require_grads_hook"):
                model.model.disable_input_require_grads()
            text_embedding = model.model.embed_tokens
            text_embedding.config = model.config
            return text_embedding

        if behavior == Phi4MMConfigBehavior.AUDIO_EMBEDDINGS:
            audio_embeddings = model.model.embed_tokens_extend.audio_embed.encoder.encoder_embedding
            audio_embeddings.config = model.config
            return audio_embeddings

        if behavior == Phi4MMConfigBehavior.AUDIO_ENCODER:
            audio_encoder = model.model.embed_tokens_extend.audio_embed.encoder
            audio_encoder.config = model.config
            return audio_encoder

        if behavior == Phi4MMConfigBehavior.AUDIO_FORWARD_EMBEDDINGS:
            audio_encoder = model.model.embed_tokens_extend.audio_embed.encoder
            audio_encoder.config = model.config
            return audio_encoder

        if behavior == Phi4MMConfigBehavior.AUDIO_SPEECH_PROJECTION:
            if hasattr(model.model.embed_tokens_extend.audio_embed, "audio_projection"):
                audio_projection = model.model.embed_tokens_extend.audio_embed.audio_projection["speech"]
                audio_projection.config = model.config
                return audio_projection
            else:
                import torch

                audio_projection = torch.nn.Sequential(
                    *[
                        model.model.embed_tokens_extend.audio_embed.up_proj_for_speech,
                        torch.nn.GELU(),
                        model.model.embed_tokens_extend.audio_embed.down_proj_for_speech,
                    ]
                )
                audio_projection.config = model.config
                return audio_projection

        if behavior == Phi4MMConfigBehavior.AUDIO_VISION_PROJECTION:
            if hasattr(model.model.embed_tokens_extend.audio_embed, "audio_projection"):
                audio_projection = model.model.embed_tokens_extend.audio_embed.audio_projection["vision"]
                audio_projection.config = model.config
                return audio_projection
            else:
                import torch

                audio_projection = torch.nn.Sequential(
                    *[
                        model.model.embed_tokens_extend.audio_embed.up_proj_for_vision_speech,
                        torch.nn.GELU(),
                        model.model.embed_tokens_extend.audio_embed.down_proj_for_vision_speech,
                    ]
                )
                audio_projection.config = model.config
                return audio_projection