IBM Granite 3.2 Brings New Vision Language Model, Chain of Thought Reasoning, Improved TimeSeries - Related to improves, cdk, granite, new, dashboard

IBM Granite 3.2 Brings New Vision Language Model, Chain of Thought Reasoning, Improved TimeSeries

IBM has introduced its new Granite [website] multi-modal and reasoning model. Granite [website] elements experimental chain-of-thought reasoning capabilities that significantly improve its predecessor's performance, a new vision language model (VLM) outperforming larger models on several benchmarks, and smaller models for more efficient deployments.

IBM says its Granite [website] 8B Instruct and Granite [website] 2B Instruct significantly outperform their [website] predecessors thanks to enhanced reasoning capabilities. Instead of providing specialized reasoning models as other companies currently do, IBM chose to include reasoning in their Instruct models as an option that can be toggled on and off depending on the particular task at hand.

One technique IBM is using in Granite [website] to build their reasoning capabilities is inference scaling, which is inspired by the idea of letting an LLM generate multiple answers and then pick the best based on some reward model — only, applied to the reasoning process.

In the context of reasoning tasks, this idea of scoring multiple answers to pick the best answer can be applied also to the “chain of thought” that often precedes answer generation. In fact, you don’t need to wait for the entire reasoning to be completed before deciding whether the reasoning was good or not.

IBM's approach advances the one popularized by DeepSeek, which uses one inference model to measure its own progress by also using a search model to explore the reasoning space. So, the process reward model helps the LLM detect and avoid wrong reasoning turns, while the search algorithm makes the process more flexible.

, their inference scaling approach is able to boost performance on the MATH500 and AIME2024 math-reasoning benchmarks and let Granite [website] outperform much larger models like GPT-4o-0513 and [website] for single-pass inference.

Granite [website] also includes a VLM particularly aimed at document understanding, named Granite Vision [website] 2B. , this lightweight model rivals larger models on enterprise benchmarks, such as DocVQA and ChartQA, but is not intended to be used as a replacement for text-only Granite models. It was trained using a specific dataset, DocFM, that IBM built on curated enterprise data, including general document images, charts, flowcharts, and diagrams.

Another component of the Granite family is Granite Guardian [website], a guardrail model able to detect risks in prompts and responses. Guardian [website] provides a similar performance to Guardian [website] at greater speed with lower inference costs and memory usage, IBM says. It introduces a new feature, verbalized confidence, to assess the potential risk in a more nuanced way by providing a confidence value.

Guardian [website] comes in two variants, Guardian [website] 5B (down from 8B in Granite [website] and Guardian [website] 3B-A800M, with the added optimization of activating only 800 million parameters out of the total three billion at inference time.

As a final note on Granite [website], it is worth mentioning that it brings new timeseries models (TTM) supporting weekly and daily forecasting in addition to the minutely to hourly resolutions already supported by its predecessor.

TTM-R2 models (including the new [website] variants) top all models for point forecasting accuracy as measured by mean absolute scaled error (MASE). TTM-R2 also ranks in the top 5 for probabilistic forecasting, as measured by continuous ranked probability score (CRPS).

In its announcement, IBM does not let go unnoticed that its TTM models are "tiny" in comparison to Google’s [website] (500M parameters) and Amazon’s Chronos-Bolt-Base (205M parameters), which ranks second and third by MASE.

While IBM's announcement appeared an impressive feat to some reddit individuals, others highlighted the fact that their reported performance may look like overfitting a few benchmarks while ignoring others. Still, although it would be naive to think that such small models (8B and 2B parameters) can be preferable to larger models performing much more effective overall or for complex tasks like coding, it is true that they may be a good fit for more specialized tasks.

Others speculate about the fact that IBM's offering specifically targets enterprises, where it is significant to have legal guarantees in case things go wrong or with potential IP issues with datasets used for training.

All Granite models are licensed under the Apache [website] license and available on HuggingFace, [website], Ollama, and LM Studio.

Tomasz Jakut reflects on the evolution of web design, recalling the days when tabl......

Ben and Ryan chat with Babak Behzad, senior engineering manager at Verkada, about running a pipeline that vectorizes 25,000 images per second into a c......

Writing Logs to Files in [website] with LogLayer.

Writing logs to files is a crucial aspect of application monitoring and debugging in production enviro......

.NET Aspire 9.1 Improves Dashboard Features

.NET Aspire team has released version [website] of the platform. This minor revision is focused on new dashboard attributes and small overall improvements across different platform components.

The dashboard functionality, built into the .NET Aspire app host, is meant to closely track various aspects of the app such as logs, traces, resources, and environment configurations. In this version, it has been enhanced with six new elements.

The first enhancement is that the "child" resources, that depend on other resources (like 1:N relationships), are now displayed nested under the "parent" resource instance. For example, if a database server resource has one or more databases in the .NET Aspire app, they are displayed underneath the server resource.

The resources in the dashboard now feature a more complete details pane, adding the references, back references and volume information. In addition, there is a filter button on the resources list that allows for filtering by desired resource types, state or health state.

Another small enhancement is the ability to change the language of the dashboard visualisation, independently of the browser language. It only affects the UI language, not the number and date formatting, which is still tied to browser settings.

The dashboard application has an OpenTelemetry endpoint for receiving the telemetry from client-side apps, which is restricted to allow only HTTP POST method for sending the telemetry. The allowed CORS origins are set in the configuration file. In .NET Aspire [website] developers can override these CORS settings and add allowed origins for custom localhost domains. It is achieved by setting the DOTNET_DASHBOARD_CORS_ALLOWED_ORIGINS environment variable.

The last dashboard addition in this version is the possibility of downloading app logs for detailed analysis in an external tool.

Lastly, there are local development Azure integration enhancements in the new version. Specifically, Azure Service Bus, Azure Cosmos DB on Linux, and Azure SignalR are now added as emulated services, allowing developers to test them without having to use real cloud resources.

This release comes three months after the release of .NET Aspire [website] , .NET Aspire major versions are released roughly with major .NET versions, while minor versions are released frequently and out of band.

, the .NET developers appreciate the orchestration capabilities that .NET Aspire brings to local development experience.

The official release notes hold more information about this release for interested readers, together with breaking changes in version [website].

A data culture is the collective behaviors and beliefs of people who value, practice, and encourage the use of data and AI to propel organizational tr......

The traditional way of building Docker images using the docker build command is simple and straightforward, but when working with complex applications......

The Case For Minimal WordPress Setups: A Contrarian View On Theme Frameworks.

AWS CDK Introduces Garbage Collection to Remove Outdated Assets

Amazon in the recent past introduced the preview of garbage collection in the AWS CDK. The new feature automatically deletes old assets in bootstrapped S3 buckets and ECR repositories, reducing maintenance and deployment costs.

The recent cdk gc command performs garbage collection on unused assets stored in the resources of the bootstrap stack, allowing developers to view, manage, and delete assets that are no longer needed. Kaizen Conroy, software engineer at AWS, and Adam Keller, senior cloud architect at AWS, explain:

For CDK developers that leverage assets at scale, they may notice over time that the bootstrapped bucket or repository accumulated old or unused data. If people wanted to clean this data on their own, CDK didn’t provide a clear way of determining which data is safe to delete. (...) We expect CDK Garbage Collection to help AWS CDK end-people save on storage costs associated with using the product while not affecting how end-people use CDK.

The AWS Cloud Development Kit (CDK) is an open source framework that provides higher-level abstractions and enables developers to define cloud infrastructure using TypeScript, JavaScript, Python, Java, C#/.NET, and Go. Developers define reusable cloud components known as constructs that can be composed together into stacks and apps. The garbage collection feature has been a long-standing request by the community, with Janne Sinivirta, principal DevOps consultant at Polar Squad, highlighting the issue as far back as 2019:

Each cdk build creates a new assets folder under [website] If this includes node_modules, the total size of the [website] folder can add up pretty quickly (mine was over 10Gb)!

, the cdk gc command is still in development and preview mode, and while the current aspects of this command are considered production-ready and safe to use, the scope of the command and its aspects might be subject to change. Developers are required to explicitly opt-in by providing the -- unstable=gc option. For example, while the current version of garbage collection is scoped to an individual account and region, there is a feature request to scope it instead to each stack.

CDK Garbage Collection exposes some parameters to help developers customize the experience, determining how aggressive the garbage collection should be. This is achieved using the -- rollback-buffer-days and -- created-buffer-days parameters, specifying respectively the days an asset has to be marked as isolated before it is eligible for deletion and the days the asset must live before it is eligible for deletion. Conroy and Keller clarify:

Rollback Buffer Days should be considered when you are not using cdk deploy and instead use a deployment method that operates on templates only, like a pipeline. If your pipeline can rollback without any involvement of the CDK CLI, this parameter will help ensure that assets are not prematurely deleted.

Adam Keller, senior cloud architect at AWS, summarizes on LinkedIn:

This was a pain point for a lot of folks as they were required to clean up resources on their own without any sort of intervention from the CDK. With the new garbage collection feature in the CDK toolkit, old, unused assets can be cleaned up with ease.

The CDK Garbage Collection is available starting in AWS CDK version [website].

Microservice architecture has become the standard for modern IT projects, enabling the creation of autonomous services with independent lifecycles. In......

Kubernetes has emerged as the go-to orchestration tool for managing containerized applications. ’s 2024 Voice of Kubernetes Exper......

Amazon ne veut pas que les candidat(e)s aux offres d'emploi puissent utiliser l'IA / GenIA pour rédiger un CV, une lettre de motivation ou lire des ré......