Skip to content

Ovis - enable merge_multimodal_embeddings_static #2003

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Draft
wants to merge 1 commit into
base: yeonsily/integ_internVL
Choose a base branch
from
+20 −3
Draft

Ovis - enable merge_multimodal_embeddings_static #2003

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
16 changes: 16 additions & 0 deletions vllm/model_executor/models/ovis2_5.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -33,6 +33,7 @@
from vllm.transformers_utils.processors.ovis2_5 import Ovis2_5Processor
from vllm.worker.hpu_model_runner import VisionBuckets
from .interfaces import MultiModalEmbeddings, SupportsMultiModal
from .utils import merge_multimodal_embeddings_static

logger = init_logger(__name__)

Expand Down Expand Up @@ -598,6 +599,21 @@ def get_multimodal_embeddings(

return tuple(embeddings) if embeddings else None

def get_input_embeddings_hpu(
self,
input_ids: torch.Tensor,
image_index_tensor: torch.Tensor,
multimodal_embeddings: Optional[MultiModalEmbeddings] = None,
) -> torch.Tensor:
inputs_embeds = self.llm.get_input_embeddings(input_ids)
if multimodal_embeddings is not None:
inputs_embeds = merge_multimodal_embeddings_static(
image_index_tensor,
inputs_embeds,
multimodal_embeddings,
)
return inputs_embeds

def get_input_embeddings(
self,
input_ids: torch.Tensor,
Expand Down
7 changes: 4 additions & 3 deletions vllm/worker/hpu_model_runner.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1802,8 +1802,9 @@ def _prepare_prompt(
dtype=torch.long,
flat=self.use_merged_prefill)

if seq_group_metadata.multi_modal_data and self.is_mm_optimized and \
'InternVLChatModel' in str(type(self.model.model)):
if (seq_group_metadata.multi_modal_data and self.is_mm_optimized
and ('InternVLChatModel' in str(type(self.model.model))
or 'Ovis2_5' in str(type(self.model.model)))):
is_image_flatten = (
input_tokens_tensor == self.image_token_id).flatten()
image_index_tensor = is_image_flatten.nonzero().squeeze(-1)
Expand Down Expand Up @@ -2769,7 +2770,7 @@ def create_dummy_multi_modal_seq_group_metadata(self, group_id, img_args,
elif "Ovis2_5" in str(type(self.model.model)):
vit_cfg = self.model.model.config.vit_config
self.image_token_id = getattr(self.model.model.config,
"image_token_id", -200)
"image_token_id", 151667)
image_h = 128
image_w = int(img_args / image_h)
num_image_tokens = int(image_h * image_w //
Expand Down
Loading