[python] offload setting block size to neuron (#2705)

deepjavalibrary · Feb 3, 2025 · 1c4ab1a · 1c4ab1a
1 parent 950dc7b
commit 1c4ab1a
Showing 1 changed file with 0 additions and 3 deletions.
diff --git a/engines/python/setup/djl_python/properties_manager/vllm_rb_properties.py b/engines/python/setup/djl_python/properties_manager/vllm_rb_properties.py
@@ -187,9 +187,6 @@ def generate_vllm_engine_arg_dict(self,
         if self.max_rolling_batch_prefill_tokens is not None:
             vllm_engine_args[
                 'max_num_batched_tokens'] = self.max_rolling_batch_prefill_tokens
-        if self.device == 'neuron':
-            vllm_engine_args['block_size'] = passthrough_vllm_engine_args.get(
-                "max_model_len")
         vllm_engine_args.update(passthrough_vllm_engine_args)
         return vllm_engine_args