Large Machine Learning (ML) models – such as large language models, generative AI, and vision models – are dramatically increasing the number of trainable parameters and are achieving state-of-the-art results. Increasing the number of parameters results in the model being too large to fit on a single VM instance thus demands distributed compute to spread the model across multiple nodes. Google Kubernetes Engine (GKE) has built-in support for NCCL Fast Socket, to help improve the time to train large ML models with distributed, multi-node clusters.

Enterprises are looking for faster and cheaper performance to train their ML models. With distributed training, communicating gradients across nodes is a performance bottleneck. Optimizing inter-node latency is critical to reduce training time and costs. Distributed training uses collective communication as a transport layer over the network between the multiple hosts. Collective communication primitives such as all-gather, all-reduce, broadcast, reduce, reduce-scatter, and point-to-point send and receive are used in distributed training in Machine Learning. 

The NVIDIA Collective Communication Library (NCCL) is commonly used by popular ML frameworks such as TensorFlow and PyTorch. It is a highly optimized implementation for high bandwidth and low latency between NVIDIA GPUs. Google developed a proprietary version of NCCL called NCCL Fast Socket to optimize performance for deep learning on Google Cloud.

NCCL Fast Socket uses a number of techniques to achieve better and more consistent NCCL performance.

• Use of multiple network flows to attain maximum throughput. NCCL Fast Socket introduces additional optimizations over NCCL’s built-in multi-stream support, including better overlapping of multiple communication requests.
• Dynamic load balancing of multiple network flows. NCCL can adapt to changing network and host conditions. With this optimization, straggler network flows will not significantly slow down the entire NCCL collective operation.
• Integration with Google Cloud’s Andromeda virtual network stack.This increases overall network throughput by avoiding contentions in virtual machines (VMs).

We tested (NVIDIA NCCL tests) the performance of NCCL Fast Socket vs NCCL on various machine shapes with 2 node GKE clusters.

The following chart shows the results. For each machine shape, the NCCL performance without Fast Socket is normalized to 1. In each case, using NCCL Fast Socket demonstrated increased performance in a range of 1.3 to 2.6 times faster internetwork communication speed.

As a built-in feature, GKE users can take advantage of NCCL Fast Socket without changing or recompiling their applications, ML frameworks (such as TensorFlow or PyTorch), or even the NCCL library itself. To start using NCCL Fast Socket, create a node pool that uses the plugin with the --enable-fast-socket and --enable-gvnic flags. You can also update an existing node pool using gcloud container node-pools update.

gcloud container node-pools create NODEPOOL_NAME \
    --accelerator type=ACCELERATOR_TYPE, count=ACCELERATOR_COUNT \
    --machine-type=MACHINE_TYPE \
    --cluster=CLUSTER_NAME \
    --enable-fast-socket \
    --enable-gvnic

To achieve better network throughput with NCCL, Google Virtual NICs (gVNICs) must be enabled when creating VM instances. For detailed instructions on how to use gVNICs, please refer to the gVNIC guide. 

To verify that NCCL Fast Socket has been enabled, view the kube-system pods:

kubectl get pods -n kube-system

And the output should b similar to:

NAME                         READY   STATUS    RESTARTS   AGE
fast-socket-installer-qvfdw  2/2     Running   0          10m
fast-socket-installer-rtjs4  2/2     Running   0          10m
fast-socket-installer-tm294  2/2     Running   0          10m

To learn more visit GKE NCCL Fast Socket documentation. We look forward to hearing how NCCL Fast Socket improves your ML Training experience on GKE.

As one of the most innate and ubiquitous forms of expression, speech is a fundamental pillar of human interaction. It comes as no surprise, then, that Google Cloud’s Speech API  has become a crucial tool for enterprise customers, launched to general availability (GA) over six years ago and, now, processing over 1 billion voice minutes each month.

With the Speech API, we’ve been pleased to serve thousands of customers and provide industry-leading speech recognition quality and cost-effective products across a range of industries. We want to constantly evolve our offerings and bring new benefits to organizations, which is why today, we’re excited to announce the GA release of our new Speech-to-Text v2 API. 

Speech-to-Text v2 modernizes our API interface and introduces several new features. It also migrates all of our existing functionality, so you can use the same models and features that you were using in STT v1 or v1p1beta1 APIs. This new version of our API also allows us to take advantage of significant cost savings in our serving path, and as such we are reducing our base price, as well as adding pricing incentives for large workloads and those willing to accept longer turnaround times.

This new infrastructure also allows us to serve a wide variety of new types of models, including Chirp , our latest 2B-parameter large speech model. All of these are Generally Available to Google Cloud Platform customers and users starting today.

Let’s have a more thorough look though at the enhanced features of Speech-to-Text API V2 and illustrate how your business can benefit from our new capabilities:

Expanding Speech-to-Text features with V2 API

Since the official launch of Speech-to-Text API back in 2017, we’ve utilized Google’s global infrastructure to host and monitor our production-facing transcription models. This robust, well-connected network has been the backbone of our offering for all of our customers. 

However, a unified view of our Speech-to-Text service has been a crucial request for our enterprise customers  who need to satisfy data residency and compliance requirements, especially in regulated industries like banking and public sector. We listened carefully to this feedback, and starting today, our Speech-to-Text v2 API supports full regionalization, allowing our customers to invoke identical copies of all our transcription models in the Google Cloud Platform region of their choice. 

In addition to giving users the flexibility to deploy in any region, we are adding a number of new features to help developers build on the API: 

• Recognizers: A user-defined named configuration that combines a model identifier, the language-locale of the audio to be transcribed, and the cloud region for the transcription model to run. Once-created, the recognizer can be referenced to every subsequent transcription request, eliminating the need for users to repeatedly define the same configuration parameters. This resourceful implementation of recognizers allows for greater flexibility in authentication and authorization, as users are not longer required to set up dedicated service accounts.
• Cloud Logging: Requests performed using a recognizer object automatically support cloud logging by default. Since the recognizers are defined as named entities, customers can partition traffic based on the recognizer of interest or collectively. 
• Audio Format Auto-Detection: Instead of having our users analyze and manually define the audio configuration settings to pass in a transcription request, the new Speech-to-Text V2 API detects settings like encoding, sampling rate, and channel count, then automatically populates the request configuration parameters. 

Increasing accuracy at Enterprise Scale with Chirp

As part of our continuous investment in foundational speech models, in March 2023 we released research results for our Universal Speech Mode (USM), a family of state-of-the-art speech models with 2B parameters and support for transcriptions of 300+ languages. In May 2023, at Google I/O, we announced Chirp in Private Preview, the latest version of the USM family, fine-tuned for our Cloud-specific use-cases. 

Chirp is now GA through the Speech-to-Text v2 API. Following extensive testing and feedback from our customers, we are making the power of pre-trained large models accessible through a simple enterprise-grade API surface. Our early adopters have seen major strides in customer engagement, thanks to the market-leading accuracy and language coverage of the new model, and we cannot wait to see what opportunities our enterprise customers will unlock.

Introducing new pricing, tiers, and options

We have heard from customers that price can be just as important as quality for many workloads. That’s why the Speech-to-Text API v2 features totally new pricing. We have lowered the cost of real-time and batch transcription from $0.024 per minute to $0.016 per minute. Additionally, we know that pricing can be a concern for those that have very large transcription workloads. For that reason, we are also introducing standard volume tiers, allowing costs as low as $0.004 per minute. As always, additional discounts are available to those with even larger workloads.

Staying true to our commitment on flexibility and choice in pricing, we noticed that even though the majority of our customers are interested in our models for real-time transcription scenarios, many are also interested in non-real time transcriptions for data at rest. With our new API v2 infrastructure, we can take greater advantage of capacity that goes unused at certain times. With our new Dynamic Batch pricing, we are passing this savings on to customers that are less latency sensitive. Aimed at users who can wait up to 24 hours for transcription results, Dynamic Batch is a new discounted pricing tier that offers 75% lower price per minute for transcription audio relative to our Standard tier.

For more information on Dynamic Batch and all our new STT v2 API pricing, check out our pricing page.

Learn more and start you Speech-to-Text journey

For more information to help get you started on your migration journey from V1 to V2, head over to the detailed documentation and try our walkthroughs to make the most of regionalization and recognizers.

If you are curious to learn more about how to use Chirp, you can get  started by following our tutorial. You can also read more about how we built it in our research blog post.

We are very excited to bring all these improvements to our developer ecosystem and provide you with the tools to leverage the power of voice for your businesses, programs, and applications.

At Google, we believe anyone should be able to quickly and easily turn their artificial intelligence (AI) idea into reality. Open source software (OSS) has become increasingly important to this goal, heavily influencing the pace of innovation in AI and machine learning (ML) ecosystems. Over the last two decades, ML has transformed Google services including Search, YouTube, Assistant, and Maps, and the basis for this transformation has always been our “open first” approach through investments in projects and ecosystems like TensorFlow, Jax, and PyTorch.

These OSS efforts are important because many AI technologies rely on closed or exclusive approaches. This wall-garden approach creates high barriers to entry for developers; limits efforts to make AI explainable, ethical, equitable; and stunts innovation. We’re committed to open ecosystems, as we firmly believe no one company should own AI/ML innovation. In this blog post, we’ll explore some of Google’s most significant OSS AI and ML contributions from recent years, as well as how our commitment to open technologies can help organizations innovate faster and more flexibly.

Openness is the way to operate as an ecosystem, not a single project

Google’s OSS initiatives extend and enable AI initiatives according to three pillars:

• Access — OSS allows developers, researchers and organizations of all sizes to leverage the latest ML technology. It is a key part of democratizing innovation in ML, fostering software diversity and choice for customers, and lowering operating cost while accelerating scale for everyone.
• Transparency — Open source datasets, ML algorithms, training models, frameworks, and compilers ensure due diligence and validation by the larger community. This is paramount when it comes to ML as it bolsters reproducibility, interpretability, ensures equity, and boosts security.
• Innovation — With more access and transparency, more innovation comes naturally. Our customers and partners take advantage of open source ML toolsets and frameworks to create more innovation in the field by contributing their own OSS.

Google’s ongoing commitment to open source AI

Google’s commitment to open standards spans over two decades of OSS contributions like TensorFlow, JAX, TFX, MLIR, KubeFlow, and Kubernetes, as well as sponsorship for critical OSS data science initiatives like Project Jupyter and NumFOCUS. Initiatives like these have helped Google become the leading Cloud Native Computing Foundation (CNCF) contributor—and by building on these efforts, Google Cloud seeks to be the best platform for the OSS AI community and ecosystem.

The perils of closed technologies can emerge at many points across ML pipelines, which is why Google’s OSS strategy encompasses the entire “idea-to-production” lifecycle, from acquiring data, to training models, to managing infrastructure, to facilitating experimentation and model refinement:

Data acquisition: starting the journey from idea to production-ready ML model

The journey from an idea to a production ML model starts with data. TensorFlow Datasets not only help users acquire ready-to-use, customizable, and highly-optimized datasets (including image, audio, and text), but also provides a set of helpful APIs that make it easy for users to organize their own datasets, regardless of whether they build with TensorFlow, Jax, or other ML frameworks.

Model development and training: shortening the path from data to useful ML

OSS libraries help developers and researchers design, implement, train, test, and debug ML algorithms. Our contributors on this front include:

• The TensorFlow core framework, which offers APIs to help data scientists and developers build and train production-grade ML models on distributed and accelerated infrastructure powered by GPUs or TPUs;
• Google’s founding membership of the PyTorch Foundation, which positions us to increase adoption of ML by building an ecosystem of open-source projects with PyTorch;
• Keras, a simple and powerful ML framework, well integrated with TensorFLow, that makes it easy for developers to quickly build and train ML models, or to leverage pre-trained AI applications;
• Model Garden, which provides implementations of many state-of-the-art computer vision and natural language processing models, maintained by Google and accessible to all, alongside APIs to accelerate training and experiments;
• Jax, a lean, intuitive, and composable system that brings together automatic differentiation (Autograd) and the Accelerated Linear Algebra (XLA) optimizing compiler to offer high-performance ML for fast research and production;
• TensorFlow Hub, a repository of trained ML models ready for fine-tuning and deployment; and,
• MediaPipe open source cross-platform, which lets users leverage customizable ML solutions for live and streaming media, including text and video.

ML infrastructure management: scaling valuable models with powerful backends

Accessing and managing infrastructure for ML, especially at scale, can be a blocker for many organizations, which is why Google has invested in initiatives including:

• The TFX (or TensorFlow Extended) platform, which offers software frameworks and tooling for full MLOps deployments, helping developers with data automation, model tracking, performance monitoring, and model retraining;
• Kubeflow, which makes deployments of ML workflows on Kubernetes simple, portable and scalable; and,
• TRC (TPU Research Cloud), which gives access to a cluster of more than 1,000 Cloud TPU devices at no charge to selected researchers who publish peer-reviewed papers and/or open source code.

Experimentation and model optimization: encouraging discovery and iteration

Data, tools for model training, and infrastructure can achieve only so much without strong processes for experimentation and optimization—which is why we’ve contribution to projects like xManager, which enables anyone to run and keep track of ML experiments locally or on Vertex AI and Tensorboard, which simplifies tracking and visualizing of model performance metrics.

These areas of focus will help not only our customers but the open-source AI community as a whole, and we’re excited to share more OSS news in coming days and months. To start exploring why many organizations choose Google Cloud for their open-source AI needs, visit our “open cloud” page and be sure to register for Google Cloud Next ‘22 for all our latest news.

Thanks to all the contributors to this blog post: Matt Vasey, George Elissaios, Warren Barkley, Manvinder Singh, James Rubin, Abhishek Ratna, Thea Lamkin, Amin Vahdat, Andrew Moore, Max Sapozhnikov, Gandhi, Vikram Kasivajhula

Gartner® named Google as a Leader in the 2022 Magic Quadrant™ for Cloud AI Developer Services report. This evaluation covered Google’s language, vision and structured data products including AutoML, all of which we deliver through Google Cloud. We believe this recognition is a reflection of the confidence and satisfaction that customers have in our language, vision, and AutoML products for developers. Google remains a Leader for the third year in a row, based upon the completeness of our vision and our ability to execute.

Developers benefit in many ways by using Cloud AI services and solutions. Customers recognize the advantages of Google’s AI and ML services for developers, such as Vertex AI, BigQuery ML, AutoML and AI APIs. In addition, customers benefit from the pace of progress in the field of Responsible AI and actionable ethics processes applied to all customer and partner solutions leveraging Google Cloud technology, as well as our core architecture including the Vertex AI platform, vision, conversational AI, language and structured data, and optimization services and key vertical industry solutions.

We believe that our ‘Leader’ placement validates this vision for AI developer tools. Let’s take a closer look at some of the report findings.

ML tools purpose-built for developers

Google’s machine learning tools have been built by developers, for developers, based on the groundbreaking research generated from Google Research and DeepMind. This developer empathy drives product development, which supports the developer community to achieve deep value from Google’s AI and ML services. An example of this is the unification of all of the tools needed for building, deploying and managing ML models into one ML platform, Vertex AI, resulting in accelerated time to production. They also cite BigQuery ML, AutoML for language, vision video and tabular data) and prebuilt ML APIs (such as speech and translation) as having high utility for developers at all levels of ML expertise to build custom AI and quickly infuse AI into their applications.

Leading organizations like OTOY, Allen Institute for AI and DeepMind (an Alphabet subsidiary) choose Google for ML, and enterprises like Twitter, Wayfair and The Home Depot shared more about their partnership with Google in their recent sessions at Google Next 2021.

Responsible AI principles and practices

Responsible AI is a critical component of successful AI. A 2020 study commissioned by Google Cloud and the Economic Intelligence Unit highlighted that ethical AI does not only prevent organizations from making egregious mistakes, but that the value of responsible AI practices for competitive edge, as well as talent acquisition and retention are notable. At Google, we not only apply our ethics review process to first party platforms and solutions, to ensure that our services design-in responsible AI from the outset, we also consult with customers and partners based on AI principles to deliver accountability and avoid unfair biases. In addition, our best-in-class tools provide developers with the functionality they need to evaluate fairness and biases in datasets and models. Our Explainable AI tools such as model cards provide model transparency in a structured, accessible way, and the What-If Tool is essential for developers and data scientists to evaluate, debug and improve their ML models.

Clear and understandable product architecture

Google Cloud’s investment in our ML product portfolio has led to a comprehensive, integrated and open offering that spans breadth (across vision, conversational AI, language and structured data, and optimization services) and depth (core AI services, with features such as Vertex AI Pipelines and Vertex Explainable AI built on top). Industry-specific solutions tailored by Google for retail, financial services, manufacturing, media and healthcare customers, such as Recommendations AI, Visual Inspection AI, Media Translation, Healthcare Data Engine, add another layer leveraging this foundational platform to help organizations and users adopt machine learning solutions more easily.

At Google Cloud, we refuse to make developers jump through hoops to derive value out of our technology; instead, we bring the value directly to them by ensuring that all of our AI and ML products and solutions work seamlessly together. To download the full report, click here. Get started on Vertex AI and talk with our sales team.

Disclaimer:

Gartner, Magic Quadrant for Cloud AI Developer Services, Van Baker, Arun Batchu, Erick Brethenoux, Svetlana Sicular, Mike Fang, May 23, 2022.

Gartner and Magic Quadrant are registered trademarks of Gartner, Inc. and/or its affiliates in the U.S. and internationally and is used herein with permission. All rights reserved.

Gartner does not endorse any vendor, product or service depicted in its research publications, and does not advise technology users to select only those vendors with the highest ratings or other designation. Gartner research publications consist of the opinions of Gartner’s research organization and should not be construed as statements of fact. Gartner disclaims all warranties, expressed or implied, with respect to this research, including any warranties of merchantability or fitness for a particular purpose.