diff --git a/Dockerfile b/Dockerfile
index e354ac5..dcec772 100644
--- a/Dockerfile
+++ b/Dockerfile
@@ -30,6 +30,6 @@ RUN --mount=type=cache,target=/root/.cache/pip \
 ADD src .
 
 # Quick temporary updates
-RUN pip install git+https://github.com/runpod/runpod-python@multijob#egg=runpod                                                        --compile
+RUN pip install git+https://github.com/runpod/runpod-python@multijob2#egg=runpod    --compile
 
 CMD python -u /handler.py
diff --git a/src/handler.py b/src/handler.py
index cd4952c..921bc43 100644
--- a/src/handler.py
+++ b/src/handler.py
@@ -8,6 +8,7 @@ from typing import AsyncGenerator, Dict
 from vllm import AsyncLLMEngine, SamplingParams, AsyncEngineArgs
 from vllm.utils import random_uuid
 import runpod
+import asyncio
 
 # Prepare the model and tokenizer
 MODEL = 'facebook/opt-125m'
@@ -27,8 +28,6 @@ engine_args = AsyncEngineArgs(
 # Create the vLLM asynchronous engine
 llm = AsyncLLMEngine.from_engine_args(engine_args)
 
-# Run the engine for one step without inputs to ensure it's ready for handling.
-llm.engine.step()
 
 def handler_fully_utilized() -> bool:
     # Compute pending sequences
@@ -96,7 +95,7 @@ async def handler(job):
     prompt = job_input['prompt']
 
     # Streaming
-    streaming = job_input['streaming'] or False
+    streaming = job_input.get('streaming', False)
 
     # Validate the inputs
     sampling_params = job_input.get('sampling_params', None)