Posts

Voxtral-Mini-4B-Realtime for real-time speech transcription is now available in Amazon SageMaker JumpStart - devamazonaws.blogspot.com

Today, AWS announced the availability of Voxtral-Mini-4B-Realtime-2602 in Amazon SageMaker JumpStart, expanding the portfolio of foundation models available to AWS customers. This model from Mistral AI is a multilingual, real-time speech-transcription model, enabling customers to build low-latency speech applications on AWS infrastructure. Voxtral-Mini-4B-Realtime excels at high-quality transcription of audio to text with a natively streaming architecture that enables real-time transcription. It supports multilingual transcription across 13 languages and offers configurable transcription delays, allowing users to balance latency and accuracy based on their needs. With SageMaker JumpStart, customers can deploy this model with just a few clicks to address their specific AI use cases. To get started with this model, navigate to the Models section of SageMaker Studio or use the SageMaker Python SDK to deploy the model to your AWS account. For more information about deploying and using ...
Post a Comment
Read more

Qwen3 embedding and reranking models for retrieval are now available in Amazon SageMaker JumpStart - devamazonaws.blogspot.com

Today, AWS announced the availability of Qwen3-VL-Embedding-2B and Qwen3-Reranker-4B in Amazon SageMaker JumpStart, expanding the portfolio of foundation models available to AWS customers. These models from Qwen are designed for information retrieval and cross-modal understanding, enabling customers to build comprehensive search pipelines on AWS infrastructure. The two models are typically used in tandem: the embedding model performs efficient initial recall, while the reranker refines results in a subsequent re-ranking stage. These models address different stages of the retrieval pipeline with specialized capabilities: Qwen3-VL-Embedding-2B accepts diverse inputs including text, images, screenshots, and videos, as well as inputs containing a mixture of these modalities, and generates semantically rich vectors that capture both visual and textual information in a shared space. It delivers performance across diverse multimodal tasks such as image-text retrieval, video-text matching...
Post a Comment
Read more

[MS] Why don't we just make the entire stack out of guard pages? - devamazonaws.blogspot.com

In my earlier overview of how compilers on different architectures perform stack probes , Cole Tobin asked , "Why not have a page fault handler that detects the faulting address being the stack and page in the other pages?" Csaba Varga replied , "My guess: you don't want an invalid pointer dereference to allocate a huge chunk of stack, just because the pointer happens to be pointing where the stack might grow, eventually. You want an invalid pointer dereference to segfault most of the time." I agree with Csaba on this. If the entire stack were made of guard pages, then it means that a single page fault far below the stack limit could take arbitrary long and allocate arbitrarily large quantities of memory. The program might have said that it wants stacks to default to 1GB, and now a single page fault on the stack could result in a long pause as the system allocates 1GB of memory. If you study the problem in the debugger, what you see is that a single memory...
Post a Comment
Read more

Amazon SageMaker HyperPod now supports AMI-based node lifecycle configuration for Slurm clusters using continuous provisioning - devamazonaws.blogspot.com

Amazon SageMaker HyperPod now supports AMI-based configuration for Slurm clusters that use continuous provisioning. Continuous provisioning adds nodes to the cluster as capacity becomes available, and this launch extends AMI-based configuration to clusters using this mode. With this support, clusters using continuous provisioning can be created without downloading, configuring, or uploading lifecycle configuration scripts to Amazon S3. AMI-based configuration provisions nodes with the software and configurations needed for a production-ready environment to run AI/ML training workloads, including required software such as Docker, Enroot, and Pyxis, and configurations such as Slurm accounting, SSH key generation, and log rotation. When using continuous provisioning, each node is configured from the AMI as it is added to the cluster, without the need to manage lifecycle configuration scripts, so nodes become available to schedule jobs sooner. To enable AMI-based configuration, omit the ...
Post a Comment
Read more

AWS Config now supports 191 additional managed rules - devamazonaws.blogspot.com

AWS Config now supports 191 additional managed rules across key services including Amazon Bedrock, Amazon SageMaker, Amazon ECS, Amazon EKS, Amazon RDS, Amazon Redshift, Amazon S3, and Amazon CloudTrail. This expansion increases built-in governance coverage across AI workloads and core cloud infrastructure. Examples of the new managed rules include evaluating resource configurations for encryption, logging, public access, network security, data protection, and other operational best practices across AWS services.  With this launch, you can deploy these new managed rules individually or as part of a conformance pack in the AWS Regions where the corresponding AWS services are available.    AWS Certificate Manager  ACM_CERTIFICATE_RSA_CHECK   Amazon API Gateway  API_GWV2_ACCESS_LOGS_ENABLED   AWS AppSync  APPSYNC_AUTHORIZATION_CHECK   APPSYNC_LOGGING_ENABLED   Amazon Athena...
Post a Comment
Read more

Amazon MSK Replicator now supports replication from external Apache Kafka clusters to MSK Standard brokers - devamazonaws.blogspot.com

Amazon MSK Replicator now supports data replication from external Apache Kafka clusters - including on-premises, self-managed on AWS, or other cloud providers to Amazon MSK Standard brokers. This capability extends replication support to MSK Standard brokers, in addition to the existing support for MSK Express brokers. With this launch, you can migrate workloads to MSK Standard brokers, support disaster recovery by using MSK clusters as a failover or backup target, and enable data distribution across hybrid and multi-cloud environments. MSK Replicator is a feature of Amazon MSK that automates data replication between Kafka clusters, eliminating the need to manage custom replication infrastructure or configure open-source tools. Previously, MSK Replicator supported replication from external Apache Kafka clusters to MSK Express brokers only. With this launch, you can now also replicate data from external Kafka clusters to MSK Standard brokers, using either SASL/SCRAM or mutual TLS (mTL...
Post a Comment
Read more

[MS] From Noisy Queries to Precise Frames: Query Decomposition for Media Asset Search - devamazonaws.blogspot.com

Image
Introduction: The Problem Media and entertainment teams work with massive asset catalogs for films and TV series. Each asset often includes structured metadata such as episode ID, scene ID, and shot ID, along with the visual content itself. Teams want a unified search experience that understands both metadata constraints and natural language intent. The problem appears when users provide both types of information in one query, for example: Two soldiers facing off across a battlefield in episode 20, scene 5, shot 2 If we embed this raw query directly for semantic retrieval, two issues appear: Metadata tokens dilute visual intent in the embedding representation. Retrieval can return visually similar assets from the wrong episode, scene, or shot. This gap is common in real workflows. Editors, archivists, and production teams often know part of the context, but they still describe the visual moment in natural language. Query decomposition solves this by splitting one noisy que...
Post a Comment
Read more