lighthouse

/

mdeberta-v3-base-kor-further

like 13

Follow

KPMG Lighthouse KR 5

안녕하세요? 아래와 같이 코드를 이용해서 사용해볼려고 하는데

input_ids = tf.keras.Input(shape=(MAX_TOKEN_LENGTH,), dtype=tf.int32)
token_type_ids = tf.keras.Input(shape=(MAX_TOKEN_LENGTH,), dtype=tf.int32)
attention_mask = tf.keras.Input(shape=(MAX_TOKEN_LENGTH,), dtype=tf.int32)

mdeberta_model = TFAutoModel.from_pretrained("lighthouse/mdeberta-v3-base-kor-further", from_pt=True, output_hidden_states=True)
x = mdeberta_model({"input_ids": input_ids, "token_type_ids": token_type_ids, "attention_mask": attention_mask}, training=False)
outputs = x["pooler_output"]

model = tf.keras.Model([input_ids, token_type_ids, attention_mask], outputs)

ValueError: Tried to convert 'shape' to a tensor and failed. Error: Cannot convert a partially known TensorShape (None, 512, 512) to a Tensor.

                Call arguments received by layer "self" (type TFDebertaV2DisentangledSelfAttention):
                  • hidden_states=tf.Tensor(shape=(None, 512, 768), dtype=float32)
                  • attention_mask=tf.Tensor(shape=(None, 1, 512, 512), dtype=uint8)
                  • query_states=None
                  • relative_pos=tf.Tensor(shape=(1, 512, 512), dtype=int64)
                  • rel_embeddings=tf.Tensor(shape=(512, 768), dtype=float32)
                  • output_attentions=False
                  • training=False
            
            
            Call arguments received by layer "attention" (type TFDebertaV2Attention):
              • input_tensor=tf.Tensor(shape=(None, 512, 768), dtype=float32)
              • attention_mask=tf.Tensor(shape=(None, 1, 512, 512), dtype=uint8)
              • query_states=None
              • relative_pos=tf.Tensor(shape=(1, 512, 512), dtype=int64)
              • rel_embeddings=tf.Tensor(shape=(512, 768), dtype=float32)
              • output_attentions=False
              • training=False
        
        
        Call arguments received by layer "layer_._0" (type TFDebertaV2Layer):
          • hidden_states=tf.Tensor(shape=(None, 512, 768), dtype=float32)
          • attention_mask=tf.Tensor(shape=(None, 1, 512, 512), dtype=uint8)
          • query_states=None
          • relative_pos=tf.Tensor(shape=(1, 512, 512), dtype=int64)
          • rel_embeddings=tf.Tensor(shape=(512, 768), dtype=float32)
          • output_attentions=False
          • training=False
    
    
    Call arguments received by layer "encoder" (type TFDebertaV2Encoder):
      • hidden_states=tf.Tensor(shape=(None, 512, 768), dtype=float32)
      • attention_mask=tf.Tensor(shape=(None, 512), dtype=int32)
      • query_states=None
      • relative_pos=None
      • output_attentions=False
      • output_hidden_states=True
      • return_dict=True
      • training=False


Call arguments received by layer "deberta" (type TFDebertaV2MainLayer):
  • self=tf.Tensor(shape=(None, 512), dtype=int32)
  • input_ids=None
  • attention_mask=tf.Tensor(shape=(None, 512), dtype=int32)
  • token_type_ids=tf.Tensor(shape=(None, 512), dtype=int32)
  • position_ids=None
  • inputs_embeds=None
  • output_attentions=False
  • output_hidden_states=True
  • return_dict=True
  • training=False

위와 같이 에러가 납니다. 다른 모델들은 오류가 없는데.. input_ids = None으로 값이 들어가서 나는 오류로 보이는데 혹시 PyTorch 코드 예제나 원인을 알수 없는지 궁금합니다.