[1/N] Initial prototype for multi-modal processor (#10044)

Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
This commit is contained in:
Cyrus Leung
2024-11-13 12:39:03 +00:00
committed by GitHub
parent bb7991aa29
commit 0b8bb86bf1
48 changed files with 1132 additions and 436 deletions
+11 -5
View File
@@ -62,7 +62,6 @@ class EngineClient(ABC):
async def beam_search(
self,
prompt: PromptType,
model_config: ModelConfig,
request_id: str,
params: BeamSearchParams,
) -> AsyncGenerator[RequestOutput, None]:
@@ -74,13 +73,14 @@ class EngineClient(ABC):
length_penalty = params.length_penalty
include_stop_str_in_output = params.include_stop_str_in_output
tokenizer = await self.get_tokenizer()
input_preprocessor = InputPreprocessor(model_config, tokenizer)
preprocessor = await self.get_input_preprocessor()
tokenizer_group = preprocessor.get_tokenizer_group()
tokenizer = await tokenizer_group.get_lora_tokenizer_async()
if is_explicit_encoder_decoder_prompt(prompt):
raise NotImplementedError
else:
processed_inputs = input_preprocessor._prompt_to_llm_inputs(
processed_inputs = preprocessor._prompt_to_llm_inputs(
prompt,
request_id=request_id,
)
@@ -220,6 +220,7 @@ class EngineClient(ABC):
Args:
request_id: The unique id of the request.
"""
...
@abstractmethod
async def get_model_config(self) -> ModelConfig:
@@ -228,8 +229,13 @@ class EngineClient(ABC):
@abstractmethod
async def get_decoding_config(self) -> DecodingConfig:
...
"""Get the decoding configuration of the vLLM engine."""
...
@abstractmethod
async def get_input_preprocessor(self) -> InputPreprocessor:
"""Get the input processor of the vLLM engine."""
...
@abstractmethod
async def get_tokenizer(