Understanding vLLM for Increasing LLM Throughput

October 17, 2023


In this article, a detailed examination of the intricate differences, features, and strengths of both vLLM and OpenLLM was presented. This comprehensive analysis showcased vLLM as a formidable open-source library developed at UC Berkeley, optimized for high throughput serving of LLMs, while highlighting OpenLLM as a versatile platform emphasizing an open approach for LLM deployment.

Transitioning to this article, it becomes imperative to recognize that the AI landscape isn't solely about the sheer power of LLMs; it also hinges on efficiency, speed, and the ability to cater to real-time demands. In the field of artificial intelligence, particularly with LLMs, throughput is more than just a technical term. It is a determinant of how quickly a system can process data, interpret it, and respond. Whether it's a chatbot or an AI-driven content generation tool crafting coherent text, the rate at which these tasks are executed plays a crucial role in determining user experience and operational efficiency.


Thus, throughput in LLMs, especially in real-world applications, becomes a pivotal consideration. It not only defines the efficiency of the model but also its viability in scenarios where promptness is paramount. As the discussion delves deeper into the world of vLLM, it seeks to explore its capabilities in maximizing LLM throughput, ensuring these language models are both intelligent and remarkably swift.

The Essence of vLLM

vLLM is quite distinct from other LLMs. While many platforms cater to the optimization of LLMs, vLLM's foundation is constructed on the principles of speed, efficiency, and versatility. Originating from UC Berkeley, its design is a testament to the synthesis of cutting-edge research with practical applicability.

At the heart of vLLM lies the PagedAttention mechanism, a groundbreaking approach to memory management. Drawing inspiration from traditional operating system concepts like paging and virtual memory, PagedAttention efficiently handles the attention key and value tensors, thereby significantly mitigating memory bottlenecks commonly associated with LLMs. This unique feature ensures vLLM's prowess in optimizing memory usage, contributing directly to its outstanding throughput capabilities.

Moreover, vLLM's adaptability with popular Hugging Face models amplifies its utility. This seamless integration not only simplifies the deployment process but also ensures that users are not confined to a limited set of architectures. They can choose, switch, and experiment with various models, each bringing its strengths to the fore. Adding to this, vLLM's dynamic memory allocation, achieved by its control over continuous batching, showcases its commitment to optimizing GPU memory usage. By making smart decisions on memory allocation based on real-time requirements, it minimizes wastage, ensuring the most efficient utilization of available resources. vLLM isn't just another tool in the vast LLM ecosystem. It embodies the future of fast and efficient language model inference, all while ensuring the flexibility and integration capabilities that modern AI applications demand.

DALL·E 2023-10-09 11.45.35 - Photograph of a high-tech computer workstation, where code and data analytics related to the vLLM model are presented. The words 'vLLM' and 'E2E Cloud

PagedAttention: The Cornerstone of vLLM

PagedAttention stands out as vLLM's core strength in LLM optimization. Unlike traditional LLMs that use contiguous memory allocation, PagedAttention, inspired by operating system concepts, allows non-contiguous memory storage. This shift to memory ‘pages’ enables efficient management of attention key and value tensors, or KV cache.

In standard memory systems, continuous blocks can lead to allocation challenges and inefficiencies. PagedAttention overcomes this by allowing dynamic, on-the-fly allocations, reducing memory wastage. The algorithm's design to handle block-aligned inputs means efficient attention computations over varied memory ranges.

In essence, PagedAttention is more than just an algorithm for vLLM—it is a transformative approach to LLM memory management, positioning vLLM as a leader in efficient language model serving.

Continuous Batching and Iteration-Level Scheduling

Continuous batching and iteration-level scheduling are pivotal to vLLM's optimized LLM serving. Unlike static batching, where the batch size remains constant, continuous batching adjusts dynamically. This dynamic approach, often termed dynamic or iteration-level scheduling, allows for immediate request injections, boosting throughput.

The distinction is clear: static batching maintains a constant batch size throughout the inference process, leading to potential inefficiencies. In contrast, dynamic batching adapts based on real-time requirements, maximizing compute resource utilization.

In practical terms, dynamic batching ensures faster response times and enhanced scalability for LLMs. It translates to more efficient, cost-effective LLM serving, especially in scenarios demanding high-throughput and low-latency.

Integration Strengths

vLLM's hallmark is its effortless integration with renowned Hugging Face models. This integration ensures users can tap into established LLM architectures without complications. The advantage is two-fold:

  • Streamlined Deployment Process
  • Optimized Model Performance. 

By aligning with Hugging Face, vLLM harnesses an extensive model repository, broadening its application spectrum and catering to diverse LLM deployment needs.

Comparative Throughput Analysis

For LLMs, throughput is a defining metric, determining how swiftly and effectively a system can manage and process a multitude of requests. For businesses and developers, it's not just about achieving the right answer but how quickly those answers are provided. In this section, the throughput capabilities of vLLM will be pitted against other prominent LLM serving platforms.

vLLM's Benchmark Brilliance

Recent benchmarks have firmly placed vLLM as a frontrunner in the LLM arena. Tests conducted under identical conditions and hardware configurations revealed that vLLM consistently outperformed its contemporaries. The system showcased its prowess by handling a considerable number of simultaneous requests without any discernible lag, a testament to its advanced architectural design and the efficiency of its PagedAttention algorithm.

Comparative Throughput Analysis

When evaluating the efficiency of vLLM, it becomes evident from benchmark results that this platform outperforms many of its counterparts in terms of throughput1. To provide a clearer perspective, let's delve into some specific comparative metrics.

Throughput Enhancement: vLLM's design inherently promotes high throughput. In comparative tests, vLLM displayed up to 23x improvement in LLM inference throughput 2. This notable enhancement is primarily attributed to the continuous batching and memory optimization techniques vLLM employs.

Performance Against Naive Continuous Batching: vLLM's performance metrics indicate that it more than doubles the throughput compared to naive continuous batching. This achievement underscores vLLM's refined approach to handling batch processing and its implications on overall serving speed.

Comparison with FasterTransformer: While FasterTransformer's 4x improvement is undeniably impressive, vLLM's continuous batching capabilities outstrip it by a significant margin 2. The introduction of advanced memory optimizations in vLLM, facilitated by its iteration-level scheduling, grants it an edge in this comparison.

Handling of Saturation: Observations highlight that vLLM maintains a consistent performance up to a QPS (Queries Per Second) of around 8, after which it approaches saturation at nearly 1900 token/s. This data point is essential as it signifies vLLM's capability to handle high loads efficiently before experiencing any degradation in performance.

The comparative data underscores vLLM's unparalleled efficiency in the LLM serving landscape. Not only does it stand out in raw performance, but the consistency of its throughput, even under stress tests, also signifies its reliability. For those seeking an LLM solution that promises both accuracy and speed, vLLM has positioned itself as a clear frontrunner. The benchmarks speak for themselves, highlighting vLLM's commitment to pushing the boundaries of what's possible in the world of Large Language Models.

In light of these metrics, it becomes apparent that vLLM is not only equipped to provide rapid responses but is also adept at managing high request loads without compromising on efficiency. The platform's design, supplemented by its strategic memory management and batching techniques, places it at the forefront of LLM serving solutions in terms of throughput.

Real-World Applications and Success Stories

The real testament to any technological innovation is its applicability and efficacy in real-world scenarios. vLLM, with its robust architecture, has already showcased impressive results in various applications:

Vicuna Platform: As a groundbreaking web-based platform, Vicuna enables users to engage with LLMs in a multitude of ways, be it text summarization, question answering, or general text generation. vLLM's efficient serving has ensured that Vicuna provides seamless LLM services, enabling users to get rapid and accurate results. 3.

Chatbot Arena: In the domain of conversational AI, response time is crucial. Chatbot Arena, a platform that lets users create and test chatbots, leverages vLLM for its LLM serving needs. The result? Efficient, cost-effective, and swift LLM services that have notably improved chatbot responsiveness.

Future Prospects and Development Roadmap

While vLLM has already made significant strides in LLM serving efficiency, the journey doesn't end here. The platform has a clear vision for the future, aiming to push the boundaries of what's possible even further.

Quantization Support: Recognizing the benefits of memory-efficient models, future releases of vLLM aim to support quantized models, which would further optimize serving speed and reduce memory requirements4.

Fine-Tuning Capabilities: To cater to the custom needs of diverse applications, vLLM is looking into offering fine-tuning support, allowing users to tailor LLMs for specific tasks, thereby improving model performance and relevance.

Expanding Integrations: Building on its seamless compatibility with Hugging Face models, vLLM plans to expand its integration roster, enabling even more diverse AI ecosystems to benefit from its efficient serving capabilities.

The roadmap for vLLM reflects its commitment to evolving with the needs of the AI community. By consistently working towards enhancing its capabilities and integrating user feedback, vLLM aims to remain at the forefront of LLM serving solutions, catering to the ever-growing and changing demands of the AI world.


In the rapidly evolving landscape of Large Language Models, throughput optimization has surfaced as a paramount concern for practical applications. vLLM has distinctly positioned itself as an innovative solution, addressing many of the challenges inherent to LLM serving. Its unique features, from PagedAttention to dynamic batching, not only highlight its technical prowess but also its applicability in real-world scenarios.

The success stories from applications like Vicuna and Chatbot Arena stand as testimony to vLLM's efficacy. These case studies illuminate the tangible benefits that can be derived from optimizing LLM throughput, making it imperative for those in the field to consider vLLM as a viable tool for their endeavors.

For those intrigued by the potential of vLLM and eager to use its capabilities, exploring the offerings of E2E cloud can provide the necessary infrastructure and support. Their cloud services are tailor-made to harness the full potential of platforms like vLLM, ensuring that users can achieve optimal LLM serving performance without the complexities of infrastructure management.

In conclusion, vLLM emerges not just as a technical marvel but also as an indispensable asset for anyone keen on harnessing the true potential of Large Language Models in real-world applications.


1. VLLM Team. Running on Clouds with SkyPilot. Read the Docs https://vllm.readthedocs.io/en/latest/serving/run_on_sky.html (2023).

2. Daniel, C., Shen, C., Liang, E. & Liaw, R. How Continuous Batching Enables 23x Throughput in LLM Inference While Reducing p50 Latency. Anyscale https://www.anyscale.com/blog/continuous-batching-llm-inference?trk=public_post_comment-text (2023).

3. Kwon, W. et al. vLLM: Easy, Fast, and Cheap LLM Serving with PagedAttention. vLLM https://vllm.ai/ (2023).

4. Kwon, W. Loading Quantized Models. GitHub https://github.com/vllm-project/vllm/issues/392#issuecomment-1627461967 (2023).

Latest Blogs
This is a decorative image for: A Complete Guide To Customer Acquisition For Startups
October 18, 2022

A Complete Guide To Customer Acquisition For Startups

Any business is enlivened by its customers. Therefore, a strategy to constantly bring in new clients is an ongoing requirement. In this regard, having a proper customer acquisition strategy can be of great importance.

So, if you are just starting your business, or planning to expand it, read on to learn more about this concept.

The problem with customer acquisition

As an organization, when working in a diverse and competitive market like India, you need to have a well-defined customer acquisition strategy to attain success. However, this is where most startups struggle. Now, you may have a great product or service, but if you are not in the right place targeting the right demographic, you are not likely to get the results you want.

To resolve this, typically, companies invest, but if that is not channelized properly, it will be futile.

So, the best way out of this dilemma is to have a clear customer acquisition strategy in place.

How can you create the ideal customer acquisition strategy for your business?

  • Define what your goals are

You need to define your goals so that you can meet the revenue expectations you have for the current fiscal year. You need to find a value for the metrics –

  • MRR – Monthly recurring revenue, which tells you all the income that can be generated from all your income channels.
  • CLV – Customer lifetime value tells you how much a customer is willing to spend on your business during your mutual relationship duration.  
  • CAC – Customer acquisition costs, which tells how much your organization needs to spend to acquire customers constantly.
  • Churn rate – It tells you the rate at which customers stop doing business.

All these metrics tell you how well you will be able to grow your business and revenue.

  • Identify your ideal customers

You need to understand who your current customers are and who your target customers are. Once you are aware of your customer base, you can focus your energies in that direction and get the maximum sale of your products or services. You can also understand what your customers require through various analytics and markers and address them to leverage your products/services towards them.

  • Choose your channels for customer acquisition

How will you acquire customers who will eventually tell at what scale and at what rate you need to expand your business? You could market and sell your products on social media channels like Instagram, Facebook and YouTube, or invest in paid marketing like Google Ads. You need to develop a unique strategy for each of these channels. 

  • Communicate with your customers

If you know exactly what your customers have in mind, then you will be able to develop your customer strategy with a clear perspective in mind. You can do it through surveys or customer opinion forms, email contact forms, blog posts and social media posts. After that, you just need to measure the analytics, clearly understand the insights, and improve your strategy accordingly.

Combining these strategies with your long-term business plan will bring results. However, there will be challenges on the way, where you need to adapt as per the requirements to make the most of it. At the same time, introducing new technologies like AI and ML can also solve such issues easily. To learn more about the use of AI and ML and how they are transforming businesses, keep referring to the blog section of E2E Networks.

Reference Links




This is a decorative image for: Constructing 3D objects through Deep Learning
October 18, 2022

Image-based 3D Object Reconstruction State-of-the-Art and trends in the Deep Learning Era

3D reconstruction is one of the most complex issues of deep learning systems. There have been multiple types of research in this field, and almost everything has been tried on it — computer vision, computer graphics and machine learning, but to no avail. However, that has resulted in CNN or convolutional neural networks foraying into this field, which has yielded some success.

The Main Objective of the 3D Object Reconstruction

Developing this deep learning technology aims to infer the shape of 3D objects from 2D images. So, to conduct the experiment, you need the following:

  • Highly calibrated cameras that take a photograph of the image from various angles.
  • Large training datasets can predict the geometry of the object whose 3D image reconstruction needs to be done. These datasets can be collected from a database of images, or they can be collected and sampled from a video.

By using the apparatus and datasets, you will be able to proceed with the 3D reconstruction from 2D datasets.

State-of-the-art Technology Used by the Datasets for the Reconstruction of 3D Objects

The technology used for this purpose needs to stick to the following parameters:

  • Input

Training with the help of one or multiple RGB images, where the segmentation of the 3D ground truth needs to be done. It could be one image, multiple images or even a video stream.

The testing will also be done on the same parameters, which will also help to create a uniform, cluttered background, or both.

  • Output

The volumetric output will be done in both high and low resolution, and the surface output will be generated through parameterisation, template deformation and point cloud. Moreover, the direct and intermediate outputs will be calculated this way.

  • Network architecture used

The architecture used in training is 3D-VAE-GAN, which has an encoder and a decoder, with TL-Net and conditional GAN. At the same time, the testing architecture is 3D-VAE, which has an encoder and a decoder.

  • Training used

The degree of supervision used in 2D vs 3D supervision, weak supervision along with loss functions have to be included in this system. The training procedure is adversarial training with joint 2D and 3D embeddings. Also, the network architecture is extremely important for the speed and processing quality of the output images.

  • Practical applications and use cases

Volumetric representations and surface representations can do the reconstruction. Powerful computer systems need to be used for reconstruction.

Given below are some of the places where 3D Object Reconstruction Deep Learning Systems are used:

  • 3D reconstruction technology can be used in the Police Department for drawing the faces of criminals whose images have been procured from a crime site where their faces are not completely revealed.
  • It can be used for re-modelling ruins at ancient architectural sites. The rubble or the debris stubs of structures can be used to recreate the entire building structure and get an idea of how it looked in the past.
  • They can be used in plastic surgery where the organs, face, limbs or any other portion of the body has been damaged and needs to be rebuilt.
  • It can be used in airport security, where concealed shapes can be used for guessing whether a person is armed or is carrying explosives or not.
  • It can also help in completing DNA sequences.

So, if you are planning to implement this technology, then you can rent the required infrastructure from E2E Networks and avoid investing in it. And if you plan to learn more about such topics, then keep a tab on the blog section of the website

Reference Links



This is a decorative image for: Comprehensive Guide to Deep Q-Learning for Data Science Enthusiasts
October 18, 2022

A Comprehensive Guide To Deep Q-Learning For Data Science Enthusiasts

For all data science enthusiasts who would love to dig deep, we have composed a write-up about Q-Learning specifically for you all. Deep Q-Learning and Reinforcement learning (RL) are extremely popular these days. These two data science methodologies use Python libraries like TensorFlow 2 and openAI’s Gym environment.

So, read on to know more.

What is Deep Q-Learning?

Deep Q-Learning utilizes the principles of Q-learning, but instead of using the Q-table, it uses the neural network. The algorithm of deep Q-Learning uses the states as input and the optimal Q-value of every action possible as the output. The agent gathers and stores all the previous experiences in the memory of the trained tuple in the following order:

State> Next state> Action> Reward

The neural network training stability increases using a random batch of previous data by using the experience replay. Experience replay also means the previous experiences stocking, and the target network uses it for training and calculation of the Q-network and the predicted Q-Value. This neural network uses openAI Gym, which is provided by taxi-v3 environments.

Now, any understanding of Deep Q-Learning   is incomplete without talking about Reinforcement Learning.

What is Reinforcement Learning?

Reinforcement is a subsection of ML. This part of ML is related to the action in which an environmental agent participates in a reward-based system and uses Reinforcement Learning to maximize the rewards. Reinforcement Learning is a different technique from unsupervised learning or supervised learning because it does not require a supervised input/output pair. The number of corrections is also less, so it is a highly efficient technique.

Now, the understanding of reinforcement learning is incomplete without knowing about Markov Decision Process (MDP). MDP is involved with each state that has been presented in the results of the environment, derived from the state previously there. The information which composes both states is gathered and transferred to the decision process. The task of the chosen agent is to maximize the awards. The MDP optimizes the actions and helps construct the optimal policy.

For developing the MDP, you need to follow the Q-Learning Algorithm, which is an extremely important part of data science and machine learning.

What is Q-Learning Algorithm?

The process of Q-Learning is important for understanding the data from scratch. It involves defining the parameters, choosing the actions from the current state and also choosing the actions from the previous state and then developing a Q-table for maximizing the results or output rewards.

The 4 steps that are involved in Q-Learning:

  1. Initializing parameters – The RL (reinforcement learning) model learns the set of actions that the agent requires in the state, environment and time.
  2. Identifying current state – The model stores the prior records for optimal action definition for maximizing the results. For acting in the present state, the state needs to be identified and perform an action combination for it.
  3. Choosing the optimal action set and gaining the relevant experience – A Q-table is generated from the data with a set of specific states and actions, and the weight of this data is calculated for updating the Q-Table to the following step.
  4. Updating Q-table rewards and next state determination – After the relevant experience is gained and agents start getting environmental records. The reward amplitude helps to present the subsequent step.  

In case the Q-table size is huge, then the generation of the model is a time-consuming process. This situation requires Deep Q-learning.

Hopefully, this write-up has provided an outline of Deep Q-Learning and its related concepts. If you wish to learn more about such topics, then keep a tab on the blog section of the E2E Networks website.

Reference Links



This is a decorative image for: GAUDI: A Neural Architect for Immersive 3D Scene Generation
October 13, 2022

GAUDI: A Neural Architect for Immersive 3D Scene Generation

The evolution of artificial intelligence in the past decade has been staggering, and now the focus is shifting towards AI and ML systems to understand and generate 3D spaces. As a result, there has been extensive research on manipulating 3D generative models. In this regard, Apple’s AI and ML scientists have developed GAUDI, a method specifically for this job.

An introduction to GAUDI

The GAUDI 3D immersive technique founders named it after the famous architect Antoni Gaudi. This AI model takes the help of a camera pose decoder, which enables it to guess the possible camera angles of a scene. Hence, the decoder then makes it possible to predict the 3D canvas from almost every angle.

What does GAUDI do?

GAUDI can perform multiple functions –

  • The extensions of these generative models have a tremendous effect on ML and computer vision. Pragmatically, such models are highly useful. They are applied in model-based reinforcement learning and planning world models, SLAM is s, or 3D content creation.
  • Generative modelling for 3D objects has been used for generating scenes using graf, pigan, and gsn, which incorporate a GAN (Generative Adversarial Network). The generator codes radiance fields exclusively. Using the 3D space in the scene along with the camera pose generates the 3D image from that point. This point has a density scalar and RGB value for that specific point in 3D space. This can be done from a 2D camera view. It does this by imposing 3D datasets on those 2D shots. It isolates various objects and scenes and combines them to render a new scene altogether.
  • GAUDI also removes GANs pathologies like mode collapse and improved GAN.
  • GAUDI also uses this to train data on a canonical coordinate system. You can compare it by looking at the trajectory of the scenes.

How is GAUDI applied to the content?

The steps of application for GAUDI have been given below:

  • Each trajectory is created, which consists of a sequence of posed images (These images are from a 3D scene) encoded into a latent representation. This representation which has a radiance field or what we refer to as the 3D scene and the camera path is created in a disentangled way. The results are interpreted as free parameters. The problem is optimized by and formulation of a reconstruction objective.
  • This simple training process is then scaled to trajectories, thousands of them creating a large number of views. The model samples the radiance fields totally from the previous distribution that the model has learned.
  • The scenes are thus synthesized by interpolation within the hidden space.
  • The scaling of 3D scenes generates many scenes that contain thousands of images. During training, there is no issue related to canonical orientation or mode collapse.
  • A novel de-noising optimization technique is used to find hidden representations that collaborate in modelling the camera poses and the radiance field to create multiple datasets with state-of-the-art performance in generating 3D scenes by building a setup that uses images and text.

To conclude, GAUDI has more capabilities and can also be used for sampling various images and video datasets. Furthermore, this will make a foray into AR (augmented reality) and VR (virtual reality). With GAUDI in hand, the sky is only the limit in the field of media creation. So, if you enjoy reading about the latest development in the field of AI and ML, then keep a tab on the blog section of the E2E Networks website.

Reference Links




Build on the most powerful infrastructure cloud

A vector illustration of a tech city using latest cloud technologies & infrastructure