modules.qwen2.qwen_configuration
Qwen2Config
Bases: EasyDeLPretrainedConfig
Source code in src/python/easydel/modules/qwen2/qwen_configuration.py
7 8 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 |
|
add_jax_args(resid_pdrop=0.0, embd_pdrop=0.0, attention_dropout=0.0, tie_word_embeddings=False, gradient_checkpointing='nothing_saveable', fcm_min_ratio=0.0, fcm_max_ratio=0.0, use_scan_mlp=False, scan_mlp_chunk_size=1024, number_rep_kv=1, bits=None, rope_theta=10000.0, hidden_act='silu', scan_layers=True, rope_scaling=None, **kwargs)
The add_jax_args function adds the following arguments to the Transformer class:
Parameters:
Name | Type | Description | Default |
---|---|---|---|
self |
Refer to the current object |
required | |
resid_pdrop |
float
|
float: Set the dropout rate for residual connections |
0.0
|
embd_pdrop |
float
|
float: Set the probability of dropping an embedding |
0.0
|
attention_dropout |
float
|
float: Set the probability of dropping out the attention layer |
0.0
|
tie_word_embeddings |
bool
|
bool: Tie the word embeddings to the decoder |
False
|
gradient_checkpointing |
str
|
str: Control the amount of memory used by jax |
'nothing_saveable'
|
fcm_min_ratio |
float
|
float: Control the minimum ratio of the number of chunks to be used in flash-based computation |
0.0
|
fcm_max_ratio |
float
|
float: Set the maximum ratio of the number of input tokens to output tokens |
0.0
|
use_scan_mlp |
bool
|
bool: Determine whether to use the scan_mlp function or not |
False
|
scan_mlp_chunk_size |
int
|
int: Set the chunk size for scan_mlp |
1024
|
number_rep_kv |
int
|
int: Determine how many times the key and value vectors are repeated |
1
|
bits |
Optional[int]
|
Optional[int]: Determine the number of bits used in the quantization |
None
|
rope_theta |
float
|
float : rope_theta for compute rope |
10000.0
|
hidden_act |
str
|
str : hidden_act for mlp |
'silu'
|
scan_layers |
bool
|
bool: Determine whether to use scan layers or not |
True
|
Returns:
Type | Description |
---|---|
The following: |
Source code in src/python/easydel/modules/qwen2/qwen_configuration.py
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 |
|
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 across devices.
Parameters:
Name | Type | Description | Default |
---|---|---|---|
fully_sharded_data_parallel |
bool
|
bool: Determine whether to partition the model fully or not |
True
|
Returns:
Type | Description |
---|---|
A list of tuples |
Source code in src/python/easydel/modules/qwen2/qwen_configuration.py
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 |
|