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/llava_next_video.py
View File

@@ -11,7 +11,8 @@ from transformers import (CLIPVisionConfig, LlavaNextVideoConfig,
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.activation import get_act_fn
from vllm.model_executor.layers.quantization import QuantizationConfig
from vllm.model_executor.layers.sampler import Sampler, SamplerOutput
@@ -139,10 +140,10 @@ def dummy_data_for_llava_next_video(ctx: InputContext, seq_len: int,
def input_processor_for_llava_next_video(ctx: InputContext,
llm_inputs: LLMInputs):
multi_modal_data = llm_inputs.get("multi_modal_data")
inputs: DecoderOnlyInputs):
multi_modal_data = inputs.get("multi_modal_data")
if multi_modal_data is None or "video" not in multi_modal_data:
return llm_inputs
return inputs
video_data = multi_modal_data["video"]
model_config = ctx.model_config
@@ -160,15 +161,15 @@ def input_processor_for_llava_next_video(ctx: InputContext,
new_prompt, new_token_ids = repeat_and_pad_placeholder_tokens(
tokenizer,
llm_inputs.get("prompt"),
llm_inputs["prompt_token_ids"],
inputs.get("prompt"),
inputs["prompt_token_ids"],
placeholder_token_id=hf_config.video_token_index,
repeat_count=video_feature_size,
)
return LLMInputs(prompt_token_ids=new_token_ids,
prompt=new_prompt,
multi_modal_data=multi_modal_data)
return token_inputs(prompt_token_ids=new_token_ids,
prompt=new_prompt,
multi_modal_data=multi_modal_data)
elif is_list_of(video_data, np.ndarray):
raise NotImplementedError(