g-h-chen commited on
Commit
82bfac7
1 Parent(s): b440d28

upload configuration_phi.py

Browse files
Files changed (1) hide show
  1. configuration_phi.py +193 -0
configuration_phi.py ADDED
@@ -0,0 +1,193 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # coding=utf-8
2
+ # Copyright 2023 Microsoft and the HuggingFace Inc. team. All rights reserved.
3
+ #
4
+ # Licensed under the Apache License, Version 2.0 (the "License");
5
+ # you may not use this file except in compliance with the License.
6
+ # You may obtain a copy of the License at
7
+ #
8
+ # http://www.apache.org/licenses/LICENSE-2.0
9
+ #
10
+ # Unless required by applicable law or agreed to in writing, software
11
+ # distributed under the License is distributed on an "AS IS" BASIS,
12
+ # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13
+ # See the License for the specific language governing permissions and
14
+ # limitations under the License.
15
+
16
+ """ Phi model configuration"""
17
+
18
+
19
+ from transformers.configuration_utils import PretrainedConfig
20
+ from transformers.utils import logging
21
+
22
+
23
+ logger = logging.get_logger(__name__)
24
+
25
+ PHI_PRETRAINED_CONFIG_ARCHIVE_MAP = {
26
+ "microsoft/phi-2": "https://huggingface.co/microsoft/phi-2/resolve/main/config.json",
27
+ }
28
+
29
+
30
+ class PhiConfig(PretrainedConfig):
31
+ r"""
32
+ This is the configuration class to store the configuration of a [`PhiModel`]. It is used to instantiate an Phi
33
+ model according to the specified arguments, defining the model architecture. Instantiating a configuration with the
34
+ defaults will yield a similar configuration to that of the Phi
35
+ [microsoft/phi-1](https://huggingface.co/microsoft/phi-1).
36
+
37
+ Configuration objects inherit from [`PretrainedConfig`] and can be used to control the model outputs. Read the
38
+ documentation from [`PretrainedConfig`] for more information.
39
+
40
+ Args:
41
+ vocab_size (`int`, *optional*, defaults to 51200):
42
+ Vocabulary size of the Phi model. Defines the number of different tokens that can be represented by the
43
+ `inputs_ids` passed when calling [`PhiModel`].
44
+ hidden_size (`int`, *optional*, defaults to 2048):
45
+ Dimension of the hidden representations.
46
+ intermediate_size (`int`, *optional*, defaults to 8192):
47
+ Dimension of the MLP representations.
48
+ num_hidden_layers (`int`, *optional*, defaults to 24):
49
+ Number of hidden layers in the Transformer decoder.
50
+ num_attention_heads (`int`, *optional*, defaults to 32):
51
+ Number of attention heads for each attention layer in the Transformer decoder.
52
+ num_key_value_heads (`int`, *optional*):
53
+ This is the number of key_value heads that should be used to implement Grouped Query Attention. If
54
+ `num_key_value_heads=num_attention_heads`, the model will use Multi Head Attention (MHA), if
55
+ `num_key_value_heads=1 the model will use Multi Query Attention (MQA) otherwise GQA is used. When
56
+ converting a multi-head checkpoint to a GQA checkpoint, each group key and value head should be constructed
57
+ by meanpooling all the original heads within that group. For more details checkout [this
58
+ paper](https://arxiv.org/pdf/2305.13245.pdf). If it is not specified, will default to
59
+ `num_attention_heads`.
60
+ resid_pdrop (`float`, *optional*, defaults to 0.0):
61
+ Dropout probability for mlp outputs.
62
+ embd_pdrop (`int`, *optional*, defaults to 0.0):
63
+ The dropout ratio for the embeddings.
64
+ attention_dropout (`float`, *optional*, defaults to 0.0):
65
+ The dropout ratio after computing the attention scores.
66
+ hidden_act (`str` or `function`, *optional*, defaults to `"gelu_new"`):
67
+ The non-linear activation function (function or string) in the decoder.
68
+ max_position_embeddings (`int`, *optional*, defaults to 2048):
69
+ The maximum sequence length that this model might ever be used with. Phi-1 and Phi-1.5 supports up to 2048
70
+ tokens.
71
+ initializer_range (`float`, *optional*, defaults to 0.02):
72
+ The standard deviation of the truncated_normal_initializer for initializing all weight matrices.
73
+ layer_norm_eps (`float`, *optional*, defaults to 1e-05):
74
+ The epsilon used by the rms normalization layers.
75
+ use_cache (`bool`, *optional*, defaults to `True`):
76
+ Whether or not the model should return the last key/values attentions (not used by all models). Only
77
+ relevant if `config.is_decoder=True`. Whether to tie weight embeddings or not.
78
+ tie_word_embeddings (`bool`, *optional*, defaults to `False`):
79
+ Whether to tie weight embeddings
80
+ rope_theta (`float`, *optional*, defaults to 10000.0):
81
+ The base period of the RoPE embeddings.
82
+ rope_scaling (`Dict`, *optional*):
83
+ Dictionary containing the scaling configuration for the RoPE embeddings. Currently supports two scaling
84
+ strategies: linear and dynamic. Their scaling factor must be an float greater than 1. The expected format
85
+ is `{"type": strategy name, "factor": scaling factor}`. When using this flag, don't update
86
+ `max_position_embeddings` to the expected new maximum. See the following thread for more information on how
87
+ these scaling strategies behave:
88
+ https://www.reddit.com/r/LocalPersimmon/comments/14mrgpr/dynamically_scaled_rope_further_increases/. This
89
+ is an experimental feature, subject to breaking API changes in future versions.
90
+ partial_rotary_factor (`float`, *optional*, defaults to 0.5):
91
+ Percentage of the query and keys which will have rotary embedding.
92
+ qk_layernorm (`bool`, *optional*, defaults to `False`):
93
+ Whether or not to normalize the Queries and Keys after projecting the hidden states.
94
+ bos_token_id (`int`, *optional*, defaults to 1):
95
+ Denotes beginning of sequences token id.
96
+ eos_token_id (`int`, *optional*, defaults to 2):
97
+ Denotes end of sequences token id.
98
+
99
+ Example:
100
+
101
+ ```python
102
+ >>> from transformers import PhiModel, PhiConfig
103
+
104
+ >>> # Initializing a Phi-1 style configuration
105
+ >>> configuration = PhiConfig.from_pretrained("microsoft/phi-1")
106
+
107
+ >>> # Initializing a model from the configuration
108
+ >>> model = PhiModel(configuration)
109
+
110
+ >>> # Accessing the model configuration
111
+ >>> configuration = model.config
112
+ ```"""
113
+
114
+ model_type = "phi"
115
+ keys_to_ignore_at_inference = ["past_key_values"]
116
+
117
+ def __init__(
118
+ self,
119
+ vocab_size=51200,
120
+ hidden_size=2048,
121
+ intermediate_size=8192,
122
+ num_hidden_layers=24,
123
+ num_attention_heads=32,
124
+ num_key_value_heads=None,
125
+ resid_pdrop=0.0,
126
+ embd_pdrop=0.0,
127
+ attention_dropout=0.0,
128
+ hidden_act="gelu_new",
129
+ max_position_embeddings=2048,
130
+ initializer_range=0.02,
131
+ layer_norm_eps=1e-5,
132
+ use_cache=True,
133
+ tie_word_embeddings=False,
134
+ rope_theta=10000.0,
135
+ rope_scaling=None,
136
+ partial_rotary_factor=0.5,
137
+ qk_layernorm=False,
138
+ bos_token_id=1,
139
+ eos_token_id=2,
140
+ **kwargs,
141
+ ):
142
+ self.vocab_size = vocab_size
143
+ self.hidden_size = hidden_size
144
+ self.intermediate_size = intermediate_size
145
+ self.num_hidden_layers = num_hidden_layers
146
+ self.num_attention_heads = num_attention_heads
147
+
148
+ if num_key_value_heads is None:
149
+ num_key_value_heads = num_attention_heads
150
+
151
+ self.num_key_value_heads = num_key_value_heads
152
+ self.resid_pdrop = resid_pdrop
153
+ self.embd_pdrop = embd_pdrop
154
+ self.attention_dropout = attention_dropout
155
+ self.hidden_act = hidden_act
156
+ self.max_position_embeddings = max_position_embeddings
157
+ self.initializer_range = initializer_range
158
+ self.layer_norm_eps = layer_norm_eps
159
+ self.use_cache = use_cache
160
+ self.rope_theta = rope_theta
161
+ self.rope_scaling = rope_scaling
162
+ self.partial_rotary_factor = partial_rotary_factor
163
+ self.qk_layernorm = qk_layernorm
164
+ self._rope_scaling_validation()
165
+
166
+ super().__init__(
167
+ bos_token_id=bos_token_id,
168
+ eos_token_id=eos_token_id,
169
+ tie_word_embeddings=tie_word_embeddings,
170
+ **kwargs,
171
+ )
172
+
173
+ # Copied from transformers.models.llama.configuration_llama.LlamaConfig._rope_scaling_validation
174
+ def _rope_scaling_validation(self):
175
+ """
176
+ Validate the `rope_scaling` configuration.
177
+ """
178
+ if self.rope_scaling is None:
179
+ return
180
+
181
+ if not isinstance(self.rope_scaling, dict) or len(self.rope_scaling) != 2:
182
+ raise ValueError(
183
+ "`rope_scaling` must be a dictionary with with two fields, `type` and `factor`, "
184
+ f"got {self.rope_scaling}"
185
+ )
186
+ rope_scaling_type = self.rope_scaling.get("type", None)
187
+ rope_scaling_factor = self.rope_scaling.get("factor", None)
188
+ if rope_scaling_type is None or rope_scaling_type not in ["linear", "dynamic"]:
189
+ raise ValueError(
190
+ f"`rope_scaling`'s type field must be one of ['linear', 'dynamic'], got {rope_scaling_type}"
191
+ )
192
+ if rope_scaling_factor is None or not isinstance(rope_scaling_factor, float) or rope_scaling_factor <= 1.0:
193
+ raise ValueError(f"`rope_scaling`'s factor field must be a float > 1, got {rope_scaling_factor}")