import pyarrow as pa
import wallaroo

from wallaroo.framework import Framework
from wallaroo.engine_config import Acceleration
from wallaroo.openai_config import OpenaiConfig
from wallaroo.continuous_batching_config import ContinuousBatchingConfig
from wallaroo.framework import SGLangConfig, VLLMConfig

wl = wallaroo.Client()

workspace = wl.get_workspace(name='amd-llama-test', create_if_not_exist=True)
wl.set_current_workspace(workspace)

{'name': 'amd-llama-test', 'id': 15, 'archived': False, 'created_by': 'jason.mccampbell@wallaroo.ai', 'created_at': '2026-01-26T21:10:10.214594+00:00', 'models': [{'name': 'llama-31-8b-sglang-config-v1', 'versions': 1, 'owner_id': '""', 'last_update_time': datetime.datetime(2026, 2, 19, 18, 39, 39, 859744, tzinfo=tzutc()), 'created_at': datetime.datetime(2026, 2, 19, 18, 39, 39, 859744, tzinfo=tzutc())}, 'pipelines': []}

model = wl.get_model("llama-31-8b-sglang-config-v1")
model

# Configuring as OpenAI

continuous_batch_config = ContinuousBatchingConfig(max_concurrent_batch_size=1024)
openai_config = OpenaiConfig(chat_completion_config={"temperature": .3, "max_tokens": 1024})
model = model.configure(openai_config=openai_config, continuous_batching_config=continuous_batch_config)

# Deploying

deployment_config = wallaroo.DeploymentConfigBuilder() \
    .cpus(.5) \
    .memory("1Gi") \
    .sidekick_cpus(model, 1) \
    .sidekick_memory(model, '12Gi') \
    .sidekick_gpus(model, 1) \
    .deployment_label('amd.com/gpu.product-name:AMD_Instinct_MI325X_VF') \
    .build()

pipeline = wl.build_pipeline('llama-31-8b-pipe-sglang-config-v2')
pipeline.clear()
pipeline.add_model_step(model)
pipeline.deploy(deployment_config = deployment_config)

Waiting for deployment - this will take up to 600s ....................................................................................... ok

pipeline.status()

{'status': 'Running',
 'details': [],
 'engines': [{'ip': '10.244.57.17',
   'name': 'engine-0',
   'status': 'Running',
   'reason': None,
   'details': [],
   'pipeline_statuses': {'pipelines': [{'id': 'llama-31-8b-pipe-sglang-config-v2',
      'status': 'Running',
      'version': '53d41331-2f6a-46dd-b950-315601ab225c'}]},
   'model_statuses': {'models': [{'model_version_id': 28,
      'name': 'llama-31-8b-sglang-config-v1',
      'sha': 'c737ece29898860ff157c548bfb727778c06a35963f06152bca4b96171691cf7',
      'status': 'Running',
      'version': '3dfe7ec3-a76f-4ef6-8aa4-f6a8d1806f73'}]}}],
 'engine_lbs': [{'ip': '10.244.57.18',
   'name': 'engine-lb-777c5f4844-r9jhr',
   'status': 'Running',
   'reason': None,
   'details': []}],
 'sidekicks': [{'ip': '10.244.57.1',
   'name': 'engine-sidekick-llama-31-8b-sglang-config-v1-28-0',
   'status': 'Running',
   'reason': None,
   'details': [],
   'statuses': ''}]}

pipeline.openai_chat_completion(messages=[{"role": "user", "content": "are you running in a test environment?", "max_tokens": 50}]).choices[0].message.content

"I'm running in a production environment. I'm a cloud-based AI model, and I don't have a traditional test environment like a software development project would. However, my training data is constantly being updated and improved by my developers to ensure I provide the most accurate and helpful responses possible."

pipeline.openai_chat_completion(messages=[{"role": "user", "content": "what is wallaroo.ai?", "max_tokens": 50}]).choices[0].message.content

'Wallaroo.ai is a cloud-based platform that enables developers to build, deploy, and manage real-time data pipelines and event-driven applications. It provides a scalable, fault-tolerant, and highly available infrastructure for processing large volumes of data in real-time.\n\nWallaroo.ai is designed to handle complex event processing, data streaming, and IoT workloads, making it suitable for a wide range of use cases, including:\n\n1. IoT data processing\n2. Real-time analytics\n3. Event-driven applications\n4. Streaming data processing\n5. Machine learning model serving\n\nThe platform offers several key features, including:\n\n1. **Stream Processing**: Wallaroo.ai provides a stream processing engine that can handle high-volume, high-velocity data streams.\n2. **Event-Driven Architecture**: The platform supports event-driven architecture, enabling developers to build applications that respond to real-time events.\n3. **Scalability**: Wallaroo.ai is designed to scale horizontally, allowing developers to add or remove nodes as needed to handle changing workloads.\n4. **Fault Tolerance**: The platform provides built-in fault tolerance, ensuring that applications remain available even in the event of node failures.\n5. **Real-Time Data Pipelines**: Wallaroo.ai enables developers to build real-time data pipelines that can handle large volumes of data.\n\nOverall, Wallaroo.ai is a powerful platform for building and deploying real-time data pipelines and event-driven applications, making it a popular choice for developers and organizations working with large volumes of data.'

pipeline.undeploy()

Please log into the following URL in a web browser:

	sample.wallaroo.ai/auth/realms/master/device?user_code=CDWW-YCSX

Login successful!
.................................... okke up to 600s