Deploy Llama with Continuous Batching Using Native vLLM Framework with QAIC using OpenAI Inference


This tutorial and the assets can be downloaded as part of the Wallaroo Tutorials repository.

Deploy Llama with Continuous Batching Using Native vLLM Framework and QAIC AI Acceleration using OpenAI Compatibility

The following tutorial demonstrates deploying the Llama LLM with the following enhancements:

  • The Wallaroo Native vLLM Framework: Provide performance optimizations with framework configuration options.
  • Continuous Batching: Configurable batch sizes balance latency and throughput use.
  • QAIC AI Acceleration: x86 compatible architecture at low power with AI acceleration.
  • OpenAI API compatibility: The LLM accepts inference requests using the OpenAI completion and chat/completion endpoints, compatible with OpenAI API clients.

For access to these sample models and for a demonstration of how to use a LLM deployment with QAIC acceleration, OpenAI API compatibility, continuous batching, and other features:

Tutorial Goals

This tutorial demonstrates the following procedure:

  • Upload a Llama LLM with:
    • The Wallaroo Native vLLM runtime
    • QAIC AI Acceleration enabled
    • Framework configuration options to enhance performance
  • After upload, set the LLM configuration options:
    • Configure continuous batching and settings.
    • Enable OpenAI API compatibility and set inference options.
  • Set a deployment configuration to allocate hardware resources and deploy the LLM.
  • Perform sample inferences via OpenAI API inference methods with and without token streaming.

Prerequisites

Tutorial Steps

Import libraries

The first step is to import the Python libraries required, mainly the Wallaroo SDK.

import base64

import wallaroo
from wallaroo.deployment_config import DeploymentConfigBuilder
from wallaroo.framework import Framework
from wallaroo.engine_config import Acceleration
from wallaroo.object import EntityNotFoundError
from wallaroo.engine_config import QaicConfig
from wallaroo.framework import VLLMConfig
import pyarrow as pa
import pandas as pd
from wallaroo.openai_config import OpenaiConfig
from wallaroo.continuous_batching_config import ContinuousBatchingConfig

Connect to the Wallaroo Instance

Next connect to Wallaroo through the Wallaroo client. The Python library is included in the Wallaroo install and available through the Jupyter Hub interface provided with your Wallaroo environment.

This is accomplished using the wallaroo.Client() command, which provides a URL to grant the SDK permission to your specific Wallaroo environment. When displayed, enter the URL into a browser and confirm permissions. Store the connection into a variable that can be referenced later.

If logging into the Wallaroo instance through the internal JupyterHub service, use wl = wallaroo.Client(). For more information on Wallaroo Client settings, see the Client Connection guide.

wl = wallaroo.Client()

LLM Upload

Uploading the LLM takes the following steps:

  • Define Schemas: The input and output schemas are defined in Apache PyArrow format. For this tutorial, they are converted to base64 strings used for uploading through the Wallaroo MLOps API.
  • Upload the model via either the Wallaroo SDK or the Wallaroo MLOps API.

Upload LLM

LLM uploads to Wallaroo are either via the Wallaroo SDK or the Wallaroo MLOps API.

The following demonstrates uploading the LLM via the SDK. In this example the QAIC acceleration configuration is defined. This is an optional step that fine tunes the QAIC AI Acceleration hardware performance to best fit the LLM.

qaic_config = QaicConfig(
    num_devices=4, 
    full_batch_size=16, 
    ctx_len=256, 
    prefill_seq_len=128, 
    mxfp6_matmul=True, 
    mxint8_kv_cache=True
)

LLMs are uploaded with the Wallaroo SDK method wallaroo.client.Client.upload_model. This this step, the following options are configured:

  • The model name and file path.
  • The framework, in this case the native vLLM runtime.
  • The optional framework configuration, which sets specific options for the LLM’s performance.
  • The input and output schemas. For OpenAI compatibility, these are ignored so are set as empty sets.
  • The hardware acceleration set to wallaroo.engine_config.Acceleration.QAIC.with_config. The addition with_config accepts the hardware configuration options.
llm = wl.upload_model(
    "llama-qaic-openai", 
    "llama-31-8b.zip", 
    framework=Framework.VLLM,
    framework_config=VLLMConfig(
        max_num_seqs=16,
        max_model_len=256,
        max_seq_len_to_capture=128, 
        quantization="mxfp6",
        kv_cache_dtype="mxint8", 
        gpu_memory_utilization=1
    ),
    input_schema=pa.schema([]),
    output_schema=pa.schema([]), 
    accel=Acceleration.QAIC.with_config(qaic_config)
)
Waiting for model loading - this will take up to 10min.

Model is pending loading to a container runtime..
Model is attempting loading to a container runtime...................................................................................................................................................................................................................................
Successful
Ready

The other upload option is the Wallaroo MLOps API endpoint v1/api/models/upload_and_convert. For this option, the base64 converted input and output schemas are used, and the framework_config and accel options are specified in dict format. Otherwise, the same parameters are set:

  • The model name and file path.
  • The conversion parameter which defines:
    • The framework as native vLLM
    • The optional framework configuration, which sets specific options for the LLM’s performance.
  • The input and output schemas set as base64 strings.
  • The accel parameter which specifies the AI accelerator as qaic with the additional hardware configuration options.
curl --progress-bar -X POST \
  -H "Content-Type: multipart/form-data" \
  -H "Authorization: Bearer <your-token-here>" \
  -F 'metadata={"name": "vllm-llama-31-8b-qaic-new-v1", "visibility": "private", "workspace_id": 6, "conversion": {"framework": "vllm", "framework_config": {"framework": "vllm", "config":{"max_num_seqs": 16, "max_model_len": 256, "max_seq_len_to_capture": 128, "quantization": "mxfp6", "kv_cache_dtype": "mxint8", "gpu_memory_utilization": 1}}, "accel": {"qaic":{"num_devices":4,"full_batch_size": 16, "ctx_len": 256, "prefill_seq_len": 128, "mxfp6_matmul":true,"mxint8_kv_cache":true}}, "python_version": "3.8", "requirements": []}, "input_schema": "/////7AAAAAQAAAAAAAKAAwABgAFAAgACgAAAAABBAAMAAAACAAIAAAABAAIAAAABAAAAAIAAABUAAAABAAAAMT///8AAAECEAAAACQAAAAEAAAAAAAAAAoAAABtYXhfdG9rZW5zAAAIAAwACAAHAAgAAAAAAAABQAAAABAAFAAIAAYABwAMAAAAEAAQAAAAAAABBRAAAAAcAAAABAAAAAAAAAAGAAAAcHJvbXB0AAAEAAQABAAAAA==", "output_schema": "/////8AAAAAQAAAAAAAKAAwABgAFAAgACgAAAAABBAAMAAAACAAIAAAABAAIAAAABAAAAAIAAABcAAAABAAAALz///8AAAECEAAAACwAAAAEAAAAAAAAABEAAABudW1fb3V0cHV0X3Rva2VucwAAAAgADAAIAAcACAAAAAAAAAFAAAAAEAAUAAgABgAHAAwAAAAQABAAAAAAAAEFEAAAACQAAAAEAAAAAAAAAA4AAABnZW5lcmF0ZWRfdGV4dAAABAAEAAQAAAA="};type=application/json' \
  -F "file=@llama-31-8b.zip;type=application/octet-stream" \
  https://qaic-poc.example.wallaroo.ai/v1/api/models/upload_and_convert | cat

When the llm is uploaded, we retrieve it via the wallaroo.client.Client.get_model for use in later steps.

llm = wl.get_model("llama-qaic-openai")
llm
Namellama-qaic-openai
Versiona9fb8483-f600-4537-8278-a4316f518c2d
File Namellama-31-8b.zip
SHA62c338e77c031d7c071fe25e1d202fcd1ded052377a007ebd18cb63eadddf838
Statusready
Image Pathproxy.replicated.com/proxy/wallaroo/ghcr.io/wallaroolabs/mac-deploy-qaic-vllm:v2025.1.0-6203
Architecturex86
Acceleration{'qaic': {'ctx_len': 256, 'num_cores': 16, 'num_devices': 4, 'mxfp6_matmul': True, 'full_batch_size': 16, 'mxint8_kv_cache': True, 'prefill_seq_len': 128, 'aic_enable_depth_first': False}}
Updated At2025-16-Jun 20:52:44
Workspace id9
Workspace nameyounes@wallaroo.ai - Default Workspace

Configure Continuous Batching

Continuous batching options are applied for the model configuration with the model.Model.configure parameter. This method required both the input and output schemas, and the wallaroo.continuous_batching_config.ContinuousBatchingConfig settings.

from wallaroo.continuous_batching_config import ContinuousBatchingConfig
cbc = ContinuousBatchingConfig(max_concurrent_batch_size = 100)

Configure OpenAI Compatibility

OpenAI compatibility options are set through the wallaroo.openai_config.OpenaiConfig object, with the most important being:

  • enabled: Enables OpenAI compatibility
  • completion_config: Sets the OpenAI completion endpoint options except stream; the stream option is only provided at inference.
  • chat_completion_config: Sets the OpenAI chat/completion endpoint options except stream; the stream option is only provided at inference.
openai_config = OpenaiConfig(
    enabled=True,
    completion_config={
        "temperature": .3,
        "max_tokens": 200
    },
    chat_completion_config={
        "temperature": .3,
        "max_tokens": 200,
        "chat_template": """
        {% for message in messages %}
            {% if message['role'] == 'user' %}
                {{ '<|user|>\n' + message['content'] + eos_token }}
            {% elif message['role'] == 'system' %}
                {{ '<|system|>\n' + message['content'] + eos_token }}
            {% elif message['role'] == 'assistant' %}
                {{ '<|assistant|>\n'  + message['content'] + eos_token }}
            {% endif %}
            
            {% if loop.last and add_generation_prompt %}
                {{ '<|assistant|>' }}
            {% endif %}
        {% endfor %}"""
    })

Set LLM Configuration

Both the continuous deployment and OpenAI API compatibility are set through the LLM’s configure method.

llm = llm.configure(openai_config=openai_config, continuous_batching_config = cbc)
llm
Namellama-qaic-openai
Versiona9fb8483-f600-4537-8278-a4316f518c2d
File Namellama-31-8b.zip
SHA62c338e77c031d7c071fe25e1d202fcd1ded052377a007ebd18cb63eadddf838
Statusready
Image Pathproxy.replicated.com/proxy/wallaroo/ghcr.io/wallaroolabs/mac-deploy-qaic-vllm:v2025.1.0-6203
Architecturex86
Acceleration{'qaic': {'ctx_len': 256, 'num_cores': 16, 'num_devices': 4, 'mxfp6_matmul': True, 'full_batch_size': 16, 'mxint8_kv_cache': True, 'prefill_seq_len': 128, 'aic_enable_depth_first': False}}
Updated At2025-16-Jun 20:52:44
Workspace id9
Workspace nameyounes@wallaroo.ai - Default Workspace

Deploy the LLM

Deploying the LLM takes the following steps:

  • Set the deployment configuration.
  • Deploy the LLM with the deployment configuration.

Set the Deployment Configuration

The deployment configuration determines what hardware resources allocated for the LLMs exclusive use. The LLM options are set via the sidekick options.

For this example, the deployment hardware includes a Qualcomm AI 100 and allocates the following resources:

  • Replicas: 1 minimum, maximum 2. This provides scalability with additional replicas scaled up or down automatically based on resource usage.
  • Cpus: 4
  • RAM: 12 Gi
  • gpus: 4
    • For Wallaroo deployment configurations for QAIC, the gpu parameter specifies the number of System-on-Chips (SoCs) allocated.
  • Deployment label: Specifies the node with the gpus.
# sidekick_gpus is the number Qualcomm AI 100 SOCs 
deployment_config = DeploymentConfigBuilder() \
    .cpus(1).memory('1Gi') \
    .sidekick_cpus(llm, 4) \
    .sidekick_memory(llm, '12Gi') \
    .sidekick_gpus(llm, 4) \
    .deployment_label("kubernetes.io/os:linux") \
    .build()

The LLm is applied to a Wallaroo pipeline as a pipeline step. Once set, the pipeline is deployed with the deployment configuration. When the deployment is complete, the LLM is ready for inference requests.

pipeline = wl.build_pipeline("llamaqaic-openai")
pipeline.undeploy()
pipeline.add_model_step(llm)
pipeline.deploy(deployment_config=deployment_config)
pipeline.status()
{'status': 'Running',
 'details': [],
 'engines': [{'ip': '10.244.69.140',
   'name': 'engine-5686f7fb48-n5kg7',
   'status': 'Running',
   'reason': None,
   'details': [],
   'pipeline_statuses': {'pipelines': [{'id': 'llamaqaic-openai',
      'status': 'Running',
      'version': '06e205c9-0769-4681-bf0b-929de6b58613'}]},
   'model_statuses': {'models': [{'model_version_id': 83,
      'name': 'llama-qaic-openai',
      'sha': '62c338e77c031d7c071fe25e1d202fcd1ded052377a007ebd18cb63eadddf838',
      'status': 'Running',
      'version': 'a9fb8483-f600-4537-8278-a4316f518c2d'}]}}],
 'engine_lbs': [{'ip': '10.244.69.160',
   'name': 'engine-lb-864866fb86-99w6k',
   'status': 'Running',
   'reason': None,
   'details': []}],
 'sidekicks': [{'ip': '10.244.69.165',
   'name': 'engine-sidekick-llama-qaic-openai-83-775844444c-4qkgl',
   'status': 'Running',
   'reason': None,
   'details': [],
   'statuses': '\n'}]}

Inference Examples

LLMs deployed in Wallaroo accept pandas DataFrames as inference inputs. These examples use the OpenAI API inference via the Wallaroo SDK and API clients.

Inference requests via OpenAI compatible methods override the OpenAI configurations applied at the LLM level. This provides additional optimizations and flexibility as needed.

OpenAI Inference via the Wallaroo SDK

Inference requests with OpenAI compatible enabled models in Wallaroo via the Wallaroo SDK use the following methods:

  • wallaroo.pipeline.Pipeline.openai_chat_completion: Submits an inference request using the OpenAI API chat/completion endpoint parameters.
  • wallaroo.pipeline.Pipeline.openai_completion: Submits an inference request using the OpenAI API completion endpoint parameters.

Each example demonstrates using these methods with and without token streaming.

The following demonstrates performing an inference with openai_chat_completion. Note that the same parameters passed match the ones used for the OpenAI chat/completion endpoint.

# Performing completions

pipeline.openai_chat_completion(messages=[{"role": "user", "content": "good morning"}]).choices[0].message.content
'Good morning. Is there something I can help you with or would you like to chat?'

This example uses the openai_completion method.

pipeline.openai_completion(prompt="tell me about wallaroo.AI", max_tokens=200).choices[0].text
"\nWallaroo is a cloud-based AI platform that uses machine learning to help businesses automate and optimize their operations. It provides a suite of tools and features that enable companies to build, deploy, and manage AI models, as well as integrate them with existing systems and workflows.\nHere are some key features and benefits of Wallaroo:\n1. **Automated AI model deployment**: Wallaroo allows businesses to deploy AI models in a matter of minutes, without requiring extensive technical expertise.\n2. **Real-time data integration**: Wallaroo can integrate with various data sources, including databases, APIs, and IoT devices, to provide real-time insights and automate decision-making.\n3. **Predictive analytics**: Wallaroo's machine learning algorithms can analyze large datasets to identify patterns, predict outcomes, and provide actionable recommendations.\n4. **Customizable workflows**: Wallaroo enables businesses to create custom workflows that automate tasks, trigger actions, and integrate with existing systems.\n5. **Scalability and security**: Wallaroo is built"

The following examples show the same methods, with token streaming enabled.

# Now with streaming

for chunk in pipeline.openai_chat_completion(messages=[{"role": "user", "content": "this is a short story about love"}], max_tokens=100, stream=True):
    print(chunk.choices[0].delta.content, end="", flush=True)
I'd love to hear it. Please go ahead and share the short story about love.
# Now with streaming

for chunk in pipeline.openai_completion(prompt="tell me about wallaroo.AI", max_tokens=200, stream=True):
    print(chunk.choices[0].text, end="", flush=True)

Wallaroo is a cloud-based, AI-powered platform that enables developers to build, deploy, and manage AI and machine learning (ML) models in a scalable and secure manner. The platform provides a range of features and tools to simplify the development and deployment of AI and ML models, including:
Model development: Wallaroo provides a visual interface for building and training AI and ML models, making it easier for developers to create and deploy models without requiring extensive expertise in machine learning.
Model deployment: Wallaroo allows developers to deploy models to a variety of environments, including cloud, on-premises, and edge devices, making it easier to integrate AI and ML capabilities into existing applications.
Model management: Wallaroo provides a centralized platform for managing AI and ML models, including model versioning, model monitoring, and model security.
Scalability: Wallaroo is designed to scale with the needs of the organization, allowing developers to build and deploy models that can handle large volumes of data and traffic.
Security:

OpenAI Inference via the OpenAI API Requests

Inference requests via the OpenAI API Client use the pipeline’s deployment inference endpoint with the OpenAI API endpoints extensions. For deployments with OpenAI compatibility enabled, the following additional endpoints are provided:

  • {Deployment inference endpoint}/openai/v1/completions: Compatible with the OpenAI API endpoint completion.
  • {Deployment inference endpoint}/openai/v1/chat/completions: Compatible with the OpenAI API endpoint chat/completion.

These requests require the following:

  • A Wallaroo pipeline deployed with Wallaroo native vLLM runtime or Wallaroo Custom Models with OpenAI compatibility enabled.
  • Authentication to the Wallaroo MLOps API. For more details, see the Wallaroo API Connection Guide.
  • Access to the deployed pipeline’s OpenAPI API endpoints.

The first example shows retrieving the authentication token to the Wallaroo instance.

token = wl.auth.auth_header()['Authorization'].split()[1]

This example performs an inference request with token streaming enabled on the completions endpoint.

# Streaming: Completion
!curl -X POST \
  -H "Authorization: Bearer abc123" \
  -H "Content-Type: application/json" \
  -d '{"model": "whatever", "prompt": "tell me a short story", "max_tokens": 100, "stream": true, "stream_options": {"include_usage": true}}' \
  https://qaic-poc.example.wallaroo.ai/v1/api/pipelines/infer/llamaqaic-openai-32/llamaqaic-openai/openai/v1/completions
data: {"id":"cmpl-5a1adc32e65849f2aee5edf2e37fdb7a","created":1750125678,"model":"llama-31-8b.zip","choices":[{"text":" about","index":0,"logprobs":null,"finish_reason":null,"stop_reason":null}],"usage":{"prompt_tokens":6,"completion_tokens":1,"total_tokens":7,"ttft":0.091563446,"tps":10.921388869527693}}

data: {"id":"cmpl-5a1adc32e65849f2aee5edf2e37fdb7a","created":1750125678,"model":"llama-31-8b.zip","choices":[{"text":" a","index":0,"logprobs":null,"finish_reason":null,"stop_reason":null}],"usage":{"prompt_tokens":6,"completion_tokens":2,"total_tokens":8,"ttft":0.091563446,"tps":16.38253480899562}}

data: {"id":"cmpl-5a1adc32e65849f2aee5edf2e37fdb7a","created":1750125678,"model":"llama-31-8b.zip","choices":[{"text":" character","index":0,"logprobs":null,"finish_reason":null,"stop_reason":null}],"usage":{"prompt_tokens":6,"completion_tokens":3,"total_tokens":9,"ttft":0.091563446,"tps":16.485729285058817}}

...

data: {"id":"cmpl-5a1adc32e65849f2aee5edf2e37fdb7a","created":1750125678,"model":"llama-31-8b.zip","choices":[{"text":" mere","index":0,"logprobs":null,"finish_reason":null,"stop_reason":null}],"usage":{"prompt_tokens":6,"completion_tokens":99,"total_tokens":105,"ttft":0.091563446,"tps":10.019649983937533}}

data: {"id":"cmpl-5a1adc32e65849f2aee5edf2e37fdb7a","created":1750125678,"model":"llama-31-8b.zip","choices":[{"text":" touch","index":0,"logprobs":null,"finish_reason":"length","stop_reason":null}],"usage":{"prompt_tokens":6,"completion_tokens":100,"total_tokens":106,"ttft":0.091563446,"tps":10.020860629999026}}

data: {"id":"cmpl-5a1adc32e65849f2aee5edf2e37fdb7a","created":1750125678,"model":"llama-31-8b.zip","choices":[],"usage":{"prompt_tokens":6,"completion_tokens":100,"total_tokens":106,"ttft":0.091563446,"tps":10.020269063389119}}

data: [DONE]

This example performs an inference request with token streaming enabled on the chat/completions endpoint.

# Streaming: Chat completion
!curl -X POST \
  -H "Authorization: Bearer abc123" \
  -H "Content-Type: application/json" \
  -d '{"model": "whatever", "messages": [{"role": "user", "content": "tell me a story"}], "max_tokens": 100, "stream": true, "stream_options": {"include_usage": true}}' \
  https://qaic-poc.example.wallaroo.ai/v1/api/pipelines/infer/llamaqaic-openai-32/llamaqaic-openai/openai/v1/chat/completions
data: {"id":"chat-02fbfe3ae2b54133a28b2deffd3aaab6","object":"chat.completion.chunk","created":1750125657,"model":"llama-31-8b.zip","choices":[{"index":0,"finish_reason":null,"message":null,"delta":{"role":"assistant"}}],"usage":{"prompt_tokens":39,"completion_tokens":0,"total_tokens":39,"ttft":0.093523807,"tps":0.0}}

data: {"id":"chat-02fbfe3ae2b54133a28b2deffd3aaab6","object":"chat.completion.chunk","created":1750125657,"model":"llama-31-8b.zip","choices":[{"index":0,"finish_reason":null,"message":null,"delta":{"role":null,"content":"Once"}}],"usage":{"prompt_tokens":39,"completion_tokens":1,"total_tokens":40,"ttft":0.093523807,"tps":10.679028938385025}}

data: {"id":"chat-02fbfe3ae2b54133a28b2deffd3aaab6","object":"chat.completion.chunk","created":1750125657,"model":"llama-31-8b.zip","choices":[{"index":0,"finish_reason":null,"message":null,"delta":{"role":null,"content":" upon"}}],"usage":{"prompt_tokens":39,"completion_tokens":2,"total_tokens":41,"ttft":0.093523807,"tps":15.893348258670727}}

...

data: {"id":"chat-02fbfe3ae2b54133a28b2deffd3aaab6","object":"chat.completion.chunk","created":1750125657,"model":"llama-31-8b.zip","choices":[{"index":0,"finish_reason":null,"message":null,"delta":{"role":null,"content":","}}],"usage":{"prompt_tokens":39,"completion_tokens":99,"total_tokens":138,"ttft":0.093523807,"tps":10.516952064534435}}

data: {"id":"chat-02fbfe3ae2b54133a28b2deffd3aaab6","object":"chat.completion.chunk","created":1750125657,"model":"llama-31-8b.zip","choices":[{"index":0,"finish_reason":"length","message":null,"delta":{"role":null,"content":" pointing"}}],"usage":{"prompt_tokens":39,"completion_tokens":100,"total_tokens":139,"ttft":0.093523807,"tps":10.511513285045494}}

data: {"id":"chat-02fbfe3ae2b54133a28b2deffd3aaab6","object":"chat.completion.chunk","created":1750125657,"model":"llama-31-8b.zip","choices":[],"usage":{"prompt_tokens":39,"completion_tokens":100,"total_tokens":139,"ttft":0.093523807,"tps":10.511361686804662}}

data: [DONE]

The following uses the OpenAI Python library to perform the inferences, using the same OpenAI endpoints.

from openai import OpenAI
client = OpenAI(
    base_url='https://qaic-poc.example.wallaroo.ai/v1/api/pipelines/infer/llamaqaic-openai-32/llamaqaic-openai/openai/v1',
    api_key=token
)
for chunk in client.chat.completions.create(model="dummy", messages=[{"role": "user", "content": "this is a short story about love"}], max_tokens=100, stream=True):
    print(chunk.choices[0].delta.content, end="", flush=True)
I'd love to hear it. Please go ahead and share the short story about love. I'll be happy to listen and respond.
for chunk in client.completions.create(model="dummy", prompt="tell me about wallaroo.AI", max_tokens=100, stream=True):
    print(chunk.choices[0].text, end="", flush=True)

Introducing wallaroo.AI
Wallايي буду Towards a Optimization Approach
For AI-Driven Sports Optimization and Trading
Background: One-click strategy models, platform agnostic, similarity testing
Theory: Extreme Value Theory ( EVT ) , GARCH , Kalman Filter algorithm
Impact: speeding through complex sett FL startup focusing implementation wallaroo.Readingmy Business model
wallaroo.ai is an AI-driven platform that aims to revolutionize the way we approach sports optimization and trading. The platform leverages

For access to these sample models and for a demonstration of how to use a LLM deployment with QAIC acceleration, OpenAI API compatibility, continuous batching, and other features: