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

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
+2 -3
View File
@@ -46,9 +46,8 @@ def _init_attn_metadata_from_tensor_dict(
# Extract the fields used to create AttentionMetadata.
valid_attn_kwargs = {}
for field in dataclasses.fields(attn_backend.get_metadata_cls()):
val = tensor_dict.pop(field.name, None)
if val is not None:
valid_attn_kwargs[field.name] = val
if field.name in tensor_dict:
valid_attn_kwargs[field.name] = tensor_dict.pop(field.name)
attn_metadata = attn_backend.make_metadata(**valid_attn_kwargs)
tensor_dict["attn_metadata"] = attn_metadata