Get Rich Insights on LLM Applications and ML Models with Customizable Dashboards
Explore Fiddler's fully customizable dashboards for visualizing the health of your LLM applications and ML models.
In this product tour, see how to configure Fiddler’s dashboards for specific teams, business goals, and use cases. With interactive charts, drag-and-drop functionality, and root cause analysis built in, Fiddler’s dashboards empower your team to monitor LLM applications and ML models across your organization.
[00:00:00] Fiddler's dashboards are the cockpit for your AI projects and models under a single pane of glass across your entire organization. These dashboards are fully configurable and customizable for any kind of model types that you might be using. That might be traditional ML models, it could be large language models, computer vision, or really any model across your entire organization.
[00:00:23] These dashboards can also be configured for specific teams, business groups, users, or even goals that your organization can have. Let me walk you through a couple examples here.
[00:00:34] The first example I want to walk us through is a business intelligence dashboard. So this is a dashboard that was built for business users related to a predictive ML model that we are observing and monitoring called bank churn.
[00:00:48] When I navigate into this dashboard, you're going to see a number of charts that are specific to this user's goals of what they want to observe and monitor over time. In this case, they were interested in revenue impact by state of the usage of this model, which is a customizable metric that was configured within the Fiddler platform and tracked for these users and this use case.
[00:01:10] We also see other charts that were included in this dashboard, such as accuracy, feature drift, prediction drift, etc. All of these charts are very easy to configure and are even interactive right here from within the dashboard. So, for example, if I want to click into a specific metric, I can jump right into Root Cause Analysis and understand different aspects of this data and of what we're tracking and monitoring related to this chart that it might include feature drift in this case.
[00:01:45] Additionally, these different dashboards and charts are very easy to configure in a point click drag drop type of interface. So, for example, if I wanted to create a new dashboard or add charts to this dashboard, I can simply click this plus button and I can access all of the saved charts that my team members may have already pre-configured, or I can create my own chart right here from within the interface.
[00:02:13] Another use case example that we see with dashboards may be around large language models. While we're going to see very similar functionality and usefulness, the metrics are going to be a bit different. So in this case, this dashboard was generated for users who want to track the operation, efficiency, and performance of an LLM.
[00:02:34] So some of the metrics that this user is interested in are going to be a cost tracker, an embedding drift, and also safety metrics that Fiddler is providing as enrichments on top of the data coming in from this particular LLM.
[00:02:52] Additionally, we have interactive UMAP charts that can be included within these dashboards as well, which are 3D visualizations of embedding vectors and the enrichments that can be interacted right here from within the tool.
[00:03:08] These dashboards, as I mentioned, are all fully configurable. Users can create as many as they would like for any kind of use case that's going to be important for you.
