vllm.prompt_adapter.models
LRUCachePromptAdapterModelManager
¶
Bases: PromptAdapterModelManager
A model manager that manages multiple prompt_adapters with LRU cache.
Source code in vllm/prompt_adapter/models.py
_active_adapters
instance-attribute
¶
_active_adapters = PromptAdapterLRUCache(
prompt_adapter_slots, _deactivate_adapter
)
_registered_adapters
instance-attribute
¶
_registered_adapters = PromptAdapterLRUCache(
capacity, deactivate_adapter
)
__init__
¶
__init__(
model: Module,
max_num_seqs: int,
max_num_batched_tokens: int,
prompt_adapter_config: PromptAdapterConfig,
)
Source code in vllm/prompt_adapter/models.py
_pin_prompt_adapter_in_cpu_cache
¶
_pin_prompt_adapter_in_cpu_cache(prompt_adapter_id: int)
Source code in vllm/prompt_adapter/models.py
_pin_prompt_adapter_in_gpu_cache
¶
_pin_prompt_adapter_in_gpu_cache(prompt_adapter_id: int)
Source code in vllm/prompt_adapter/models.py
activate_adapter
¶
Source code in vllm/prompt_adapter/models.py
add_adapter
¶
add_adapter(prompt_adapter: PromptAdapterModel) -> bool
Add a PromptAdapterModel to the manager.
Source code in vllm/prompt_adapter/models.py
list_adapters
¶
list_adapters() -> Dict[int, PromptAdapterModel]
pin_adapter
¶
Pin a PromptAdapterModel in the manager cache.
PromptAdapterLRUCache
¶
Bases: AdapterLRUCache[PromptAdapterModel]
Source code in vllm/prompt_adapter/models.py
PromptAdapterModel
¶
Bases: AdapterModel
Source code in vllm/prompt_adapter/models.py
__init__
¶
from_local_checkpoint
classmethod
¶
from_local_checkpoint(
adapter_model_path: str,
prompt_adapter_id: int,
num_virtual_tokens: int,
config: PromptAdapterConfig,
device: str = "cuda",
) -> PromptAdapterModel
Source code in vllm/prompt_adapter/models.py
PromptAdapterModelManager
¶
Bases: AdapterModelManager
A manager that manages multiple Prompt Adapter models.
Source code in vllm/prompt_adapter/models.py
110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 |
|
max_num_batched_tokens
instance-attribute
¶
max_num_batched_tokens = (
ceil(max_num_batched_tokens / 8) * 8
)
prompt_adapter_index_to_id
instance-attribute
¶
__init__
¶
__init__(
model: Module,
max_num_seqs: int,
max_num_batched_tokens: int,
prompt_adapter_config: PromptAdapterConfig,
)
Create a PromptAdapterModel and adapter for a given model.
Parameters:
Name | Type | Description | Default |
---|---|---|---|
model
|
Module
|
the model to be adapted. |
required |
max_num_seqs
|
int
|
the maximum number of sequences model can run in a single batch. |
required |
max_num_batched_tokens
|
int
|
the maximum number of tokens model can run in a single batch. |
required |
prompt_adapter_config
|
PromptAdapterConfig
|
the PromptAdapter config, |
required |
Source code in vllm/prompt_adapter/models.py
_add_adapter
¶
_add_adapter(prompt_adapter: PromptAdapterModel)
_create_prompt_adapter_modules
¶
Source code in vllm/prompt_adapter/models.py
_deactivate_adapter
¶
_deactivate_adapter(prompt_adapter_id: int)
Source code in vllm/prompt_adapter/models.py
_set_adapter_mapping
¶
_set_adapter_mapping(mapping: PromptAdapterMapping) -> None
Source code in vllm/prompt_adapter/models.py
activate_adapter
¶
Move PromptAdapter into a GPU buffer to be used in the forward pass.
Source code in vllm/prompt_adapter/models.py
add_adapter
¶
add_adapter(adapter: PromptAdapterModel) -> bool
deactivate_adapter
¶
get_adapter
¶
list_adapters
¶
pin_adapter
¶
Pin a PromptAdapterModel in the manager cache.
Source code in vllm/prompt_adapter/models.py
register_module
¶
remove_adapter
¶
remove_all_adapters
¶
Remove all PromptAdapterModel from the manager.
replace_submodule
¶
Replace a submodule in a model with a new module.
Source code in vllm/prompt_adapter/models.py
set_adapter_mapping
¶
set_adapter_mapping(mapping: PromptAdapterMapping) -> None
convert_mapping
¶
convert_mapping(
mapping: PromptAdapterMapping,
prompt_adapter_index_to_id: List[Optional[int]],
) -> Tensor
Converts PromptAdapterMapping to index tensors.
Parameters:
Name | Type | Description | Default |
---|---|---|---|
mapping
|
PromptAdapterMapping
|
PromptAdapterMapping mapping rows in a batch to PromptAdapter ids. |
required |
prompt_adapter_index_to_id
|
List[Optional[int]]
|
List mapping PromptAdapter ids to PromptAdapter indices. |
required |
Returns:
Name | Type | Description |
---|---|---|
pa_indices |
Tensor
|
Tensor of shape [batch_size] mapping batch rows to PromptAdapter indices. |
Source code in vllm/prompt_adapter/models.py
convert_to_embedding_indices
¶
Source code in vllm/prompt_adapter/models.py
create_prompt_adapter_manager
¶
create_prompt_adapter_manager(
model: Module,
max_num_seqs: int,
max_num_batched_tokens: int,
prompt_adapter_config: PromptAdapterConfig,
prompt_adapter_manager_cls: Type[
PromptAdapterModelManager
] = PromptAdapterModelManager,
**kwargs,
) -> PromptAdapterModelManager
Create a PromptAdapterModel for a given model.