| 101 | model_type = "chatglm" |
| 102 | |
| 103 | def __init__( |
| 104 | self, |
| 105 | vocab_size=150528, |
| 106 | hidden_size=4096, |
| 107 | num_layers=28, |
| 108 | num_attention_heads=32, |
| 109 | layernorm_epsilon=1e-5, |
| 110 | use_cache=False, |
| 111 | bos_token_id=150004, |
| 112 | eos_token_id=150005, |
| 113 | pad_token_id=0, |
| 114 | max_sequence_length=2048, |
| 115 | inner_hidden_size=16384, |
| 116 | position_encoding_2d=True, |
| 117 | **kwargs |
| 118 | ): |
| 119 | self.num_layers = num_layers |
| 120 | self.vocab_size = vocab_size |
| 121 | self.hidden_size = hidden_size |
| 122 | self.num_attention_heads = num_attention_heads |
| 123 | self.max_sequence_length = max_sequence_length |
| 124 | self.layernorm_epsilon = layernorm_epsilon |
| 125 | self.inner_hidden_size = inner_hidden_size |
| 126 | self.use_cache = use_cache |
| 127 | self.bos_token_id = bos_token_id |
| 128 | self.eos_token_id = eos_token_id |
| 129 | self.pad_token_id = pad_token_id |
| 130 | self.position_encoding_2d = position_encoding_2d |
| 131 | super().__init__( |
| 132 | pad_token_id=pad_token_id, |
| 133 | bos_token_id=bos_token_id, |
| 134 | eos_token_id=eos_token_id, |
| 135 | **kwargs |
| 136 | ) |
| 137 | |
| 138 | class InvalidScoreLogitsProcessor(LogitsProcessor): |
| 139 | def __call__(self, input_ids: torch.LongTensor, scores: torch.FloatTensor) -> torch.FloatTensor: |