[Misc] Simplify max tokens in multimodal registry (#27500)

Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
This commit is contained in:
Cyrus Leung
2025-10-24 23:56:01 -07:00
committed by GitHub
parent b853540388
commit 4c5f632165
4 changed files with 17 additions and 38 deletions
+5 -2
View File
@@ -355,7 +355,11 @@ class MultiModalProfiler(Generic[_I]):
mm_counts=mm_counts,
)
if max_tokens_per_item is not None:
return max_tokens_per_item
return {
modality: max_tokens
for modality, max_tokens in max_tokens_per_item.items()
if mm_counts.get(modality, 0) > 0
}
mm_inputs = self._get_dummy_mm_inputs(seq_len, mm_counts)
return self._get_mm_num_tokens(mm_inputs, mm_embeddings_only=mm_embeddings_only)
@@ -375,5 +379,4 @@ class MultiModalProfiler(Generic[_I]):
This is important to take into account when profiling and
initializing the encoder cache size.
"""
return self._get_mm_max_tokens(seq_len, mm_counts, mm_embeddings_only=False)