| |
|
|
| from transformers import PretrainedConfig |
|
|
| class PaddleOCRVLConfig(PretrainedConfig): |
| """ |
| Configuration class for PaddleOCR-VL MLX model. |
| |
| This is the configuration class to store the configuration of a PaddleOCRVLForConditionalGeneration. |
| It is used to instantiate a PaddleOCR-VL model according to the specified arguments. |
| """ |
| |
| model_type = "paddleocr_vl" |
| |
| def __init__( |
| self, |
| |
| vision_config=None, |
| |
| hidden_size=1024, |
| intermediate_size=3072, |
| num_hidden_layers=18, |
| num_attention_heads=16, |
| num_key_value_heads=2, |
| head_dim=128, |
| hidden_act="silu", |
| max_position_embeddings=131072, |
| rms_norm_eps=1e-5, |
| rope_theta=500000, |
| rope_is_neox_style=True, |
| use_3d_rope=True, |
| |
| vocab_size=103424, |
| image_token_id=100295, |
| vision_start_token_id=101305, |
| vision_end_token_id=101306, |
| video_token_id=101307, |
| |
| tie_word_embeddings=False, |
| use_cache=False, |
| use_flash_attention=False, |
| attention_probs_dropout_prob=0.0, |
| hidden_dropout_prob=0.0, |
| **kwargs |
| ): |
| self.vision_config = vision_config or {} |
| self.hidden_size = hidden_size |
| self.intermediate_size = intermediate_size |
| self.num_hidden_layers = num_hidden_layers |
| self.num_attention_heads = num_attention_heads |
| self.num_key_value_heads = num_key_value_heads |
| self.head_dim = head_dim |
| self.hidden_act = hidden_act |
| self.max_position_embeddings = max_position_embeddings |
| self.rms_norm_eps = rms_norm_eps |
| self.rope_theta = rope_theta |
| self.rope_is_neox_style = rope_is_neox_style |
| self.use_3d_rope = use_3d_rope |
| self.vocab_size = vocab_size |
| self.image_token_id = image_token_id |
| self.vision_start_token_id = vision_start_token_id |
| self.vision_end_token_id = vision_end_token_id |
| self.video_token_id = video_token_id |
| self.tie_word_embeddings = tie_word_embeddings |
| self.use_cache = use_cache |
| self.use_flash_attention = use_flash_attention |
| self.attention_probs_dropout_prob = attention_probs_dropout_prob |
| self.hidden_dropout_prob = hidden_dropout_prob |
| |
| super().__init__(**kwargs) |
|
|
