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 6, 08:15 UTC
Fix attribute error in `isaac_patch_hf_runner` (#37685)
Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
Harry Mellor committed
Mar 20, 2026 at 14:49 UTC
9f6d9dd371c63154dddd2a8b85d7f337f3e10911
1 file changed
+9
-4
tests/models/multimodal/generation/vlm_utils/model_utils.py
+9
-4
@@ -24,6 +24,7 @@ from transformers import (
24
GenerationConfig,
25
GenerationMixin,
26
)
27
+from transformers.masking_utils import create_causal_mask
28
from transformers.video_utils import VideoMetadata
29
30
from vllm.logprobs import SampleLogprobs
@@ -680,10 +681,14 @@ def isaac_patch_hf_runner(hf_model: HfRunner) -> HfRunner:
681
sin = sin.to(inputs_embeds.dtype)
682
683
# Prepare attention mask
683
- if attention_mask is not None:
684
- attention_mask = self._update_causal_mask(
685
- attention_mask, inputs_embeds, cache_position, past_key_values, False
686
- )
684
+ attention_mask = create_causal_mask(
685
+ config=self.config,
686
+ input_embeds=inputs_embeds,
687
+ attention_mask=attention_mask,
688
+ past_key_values=past_key_values,
689
+ position_ids=position_ids,
690
+ cache_position=cache_position,
691
+ )
692
693
# Initialize and collect hidden states
694
hidden_states = inputs_embeds