A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 15, 19:45 UTC
[Bugfix] Missing thumbnail from NVLM-D processor (#14633)
Signed-off-by: ameyanjarlekar <aanjarlekar@nvidia.com>
ameyanjarlekar committed
Mar 12, 2025 at 08:50 UTC
4a754fcf15993c53309148ababa230df870aa47b
1 file changed
+1
-1
vllm/model_executor/models/nvlm_d.py
+1
-1
@@ -45,7 +45,7 @@ class NVLMProcessor(BaseInternVLProcessor):
45
raise NotImplementedError("Embedding inputs are not supported")
46
47
tile_pos_identifiers = [f"<tile_{i}>" for i in range(1, num_patches)]
48
- if self.use_thumbnail and num_patches != 1:
48
+ if self.use_thumbnail:
49
tile_pos_identifiers += ["<tile_global_thumbnail>"]
50
51
context_size = feature_size // num_patches