hey ho let's go

HabanaAI · Feb 24, 2025 · 7ad7226 · 7ad7226
1 parent 83b568f
commit 7ad7226
Showing 1 changed file with 1 addition and 0 deletions.
diff --git a/vllm/v1/worker/hpu_model_runner.py b/vllm/v1/worker/hpu_model_runner.py
@@ -654,6 +654,7 @@ def __init__(
         self.max_prefill_batch_size = 16  # TODO(kzawora): add knob for that
         self.padding_aware_scheduling = True  # TODO(kzawora): add knob for that
         self.padding_ratio_threshold = 0.9  # TODO(kzawora): add knob for that
+        os.environ['VLLM_CONTIGUOUS_PA'] = 'false' # NOTE(kzawora): this is a workaround
         self.use_contiguous_pa = os.environ.get('VLLM_CONTIGUOUS_PA',
                                                 'true').lower() == 'true'
         self.seen_configs: set = set()