modules.deepseek_v2.deepseek_configuration
DeepseekV2Config
Bases: EasyDeLPretrainedConfig
Source code in src/python/easydel/modules/deepseek_v2/deepseek_configuration.py
9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 |
|
add_jax_args(gradient_checkpointing='nothing_saveable', use_scan_mlp=False, scan_mlp_chunk_size=1024, bits=None, rope_scaling=None, **kwargs)
The add_jax_args function adds the following arguments to the model:
Parameters:
Name | Type | Description | Default |
---|---|---|---|
self |
Bind the attributes and methods of a class to an instance of that class |
required | |
gradient_checkpointing |
str
|
str: Determine whether to use gradient checkpointing |
'nothing_saveable'
|
use_scan_mlp |
bool
|
bool: Determine whether to use the scan_mlp function or not |
False
|
scan_mlp_chunk_size |
int
|
int: Chunk the input to the mlp |
1024
|
number_rep_kv |
int: Control the number of times that the key and value vectors are repeated |
required | |
bits |
Optional[int]
|
Optional[int]: Specify the number of bits to use for quantization |
None
|
attention_dropout |
float: Set the dropout rate for the attention layer |
required | |
attention_bias |
bool: when ever to use attention_bias |
required | |
initialization_of_moe |
bool: initialization of moe needs to disable some dynamic part's this boolean variable will turn them off. |
required | |
rope_scaling |
Dict[str, Union[str, float]]
|
Dict[str, Union[str, float]]: rope_scaling for rope |
None
|
Returns:
Type | Description |
---|---|
A tuple of the following: |
Source code in src/python/easydel/modules/deepseek_v2/deepseek_configuration.py
164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 |
|
get_partition_rules(fully_sharded_data_parallel=True)
The get_partition_rules function is used to define the partitioning scheme for a model. It returns a list of tuples, where each tuple contains two elements: 1) A regex string that matches the name of one or more parameters in the model. 2) A PartitionScheme object that defines how those parameters should be partitioned.
Parameters:
Name | Type | Description | Default |
---|---|---|---|
fully_sharded_data_parallel |
bool
|
bool: Determine whether to use the fully_sharded_data_parallel partitioning scheme or not |
True
|
Returns:
Type | Description |
---|---|
A list of tuples |
Source code in src/python/easydel/modules/deepseek_v2/deepseek_configuration.py
116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 |
|