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

[Core][VLM] Add precise multi-modal placeholder tracking (#8346)

Browse Source 
Signed-off-by: Peter Salas <peter@fixie.ai>
This commit is contained in:
Peter Salas
2024-11-01 16:21:10 -07:00
committed by GitHub
parent d151fde834
commit 6c0b7f548d
53 changed files with 913 additions and 281 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
vllm/model_executor/models/minicpmv.py
View File

@@ -36,8 +36,8 @@ from typing_extensions import NotRequired
from vllm.attention import AttentionMetadata
from vllm.config import CacheConfig, LoRAConfig, MultiModalConfig
from vllm.inputs import (INPUT_REGISTRY, DecoderOnlyInputs, InputContext,
token_inputs)
from vllm.inputs import (INPUT_REGISTRY, DecoderOnlyInputs, DummyData,
InputContext, token_inputs)
from vllm.model_executor.layers.logits_processor import LogitsProcessor
from vllm.model_executor.layers.quantization import QuantizationConfig
from vllm.model_executor.layers.resampler import (BaseResampler, Resampler2,
@@ -277,7 +277,7 @@ def dummy_data_for_minicpmv(ctx: InputContext, seq_len: int,
seq_data = dummy_seq_data_for_minicpmv(seq_len, num_images)
mm_data = dummy_image_for_minicpmv(ctx, hf_config, num_images)
return seq_data, mm_data
return DummyData(seq_data, mm_data)
def input_processor_for_minicpmv(ctx: InputContext, inputs: DecoderOnlyInputs):