biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[Core] Rename input data types (#8688)

Browse Source
This commit is contained in:
Cyrus Leung
2024-10-16 18:49:37 +08:00
committed by GitHub
parent 1de76a0e55
commit cee711fdbb
32 changed files with 438 additions and 340 deletions
Download Patch File Download Diff File Expand all files Collapse all files

19
vllm/model_executor/models/fuyu.py
View File

@@ -27,7 +27,8 @@ from transformers import FuyuConfig, FuyuImageProcessor
from vllm.attention import AttentionMetadata
from vllm.config import CacheConfig, MultiModalConfig
from vllm.inputs import INPUT_REGISTRY, InputContext, LLMInputs
from vllm.inputs import (INPUT_REGISTRY, DecoderOnlyInputs, InputContext,
token_inputs)
from vllm.model_executor.layers.linear import ColumnParallelLinear
from vllm.model_executor.layers.quantization import QuantizationConfig
from vllm.model_executor.layers.sampler import SamplerOutput
@@ -149,10 +150,10 @@ def _fuyu_image_preprocess(image_processor: FuyuImageProcessor,
return model_image_input
def input_processor_for_fuyu(ctx: InputContext, llm_inputs: LLMInputs):
multi_modal_data = llm_inputs.get("multi_modal_data")
def input_processor_for_fuyu(ctx: InputContext, inputs: DecoderOnlyInputs):
multi_modal_data = inputs.get("multi_modal_data")
if multi_modal_data is None or "image" not in multi_modal_data:
return llm_inputs
return inputs
model_config = ctx.model_config
image_data = multi_modal_data["image"]
@@ -176,8 +177,8 @@ def input_processor_for_fuyu(ctx: InputContext, llm_inputs: LLMInputs):
raise TypeError(f"Invalid image type: {type(image_data)}")
# process prompts
prompt = llm_inputs.get("prompt")
prompt_token_ids = llm_inputs["prompt_token_ids"]
prompt = inputs.get("prompt")
prompt_token_ids = inputs["prompt_token_ids"]
tokenizer = cached_get_tokenizer(model_config.model)
# dim0 is batch_size, dim1 is subseq_size which will always be 1
image_input_ids: List[List[
@@ -190,9 +191,9 @@ def input_processor_for_fuyu(ctx: InputContext, llm_inputs: LLMInputs):
new_prompt_token_ids = image_input_ids + bos_token + prompt_token_ids[
1:] + boa_token
return LLMInputs(prompt=new_prompt,
prompt_token_ids=new_prompt_token_ids,
multi_modal_data=new_multi_modal_data)
return token_inputs(prompt=new_prompt,
prompt_token_ids=new_prompt_token_ids,
multi_modal_data=new_multi_modal_data)
def input_mapper_for_fuyu(ctx: InputContext, data: object):