4.5 View Results Locally¶
In the previous section, we noticed that the console provided a truncated summary of the results of the evaluation run. But what if you wanted to see the details or observe the traces for these runs? You have two options - local or Azure AI Foundry portal.
1. View Traces Locally¶
The console log from the execution run will show you a trace viewer link URL:
1 |
|
Navigate to that link in your browser, and you should get a trace viewer landing page that looks like this, with Traces and Collections.
Click on Traces - you should now be able to see records for the trace runs for each invocation of the chat application, with timestamps for each run.
Click on any row in the table to get the detailed trace view for that run - where you can drive into the call stack to understand the time taken for each call, the tokens consumer, and the inputs and outputs from each processing step.
COMPARE IT: Contrast the local trace view to the Azure Foundry Portal view seen earlier. How are they different?
2. View Results Locally¶
The evaluation run provides us a summary of results in the console, but it is truncated and hard to analyze. However, observe that the results are also stored in a local JSON file that you can open up in the Visual Studio Code IDE for exploration.
Check out the src.sample/api/myevalresults.json
for a sample file showing the outputs from a previous run. The snippet below reproduces one of the rows from those results - let's see what that provides:
- The
inputs.query
andinputs.truth
are from the test input file - The
outputs.context
reflects retrieved product documents used (RAG) - The
outputs.response
gives the target models returned response - The
outputs.groundedness.groundedness
has a rating of 1 (not grounded) - The
outputs.groundedness.groundedness_reaason
explains why (context unused)
Note: While the response is not grounded in the context, it does in fact reflect our prompt template guidance to ask for clarifications on questions where the intent is not clear.
Click to expand a view a sample row from myevalresults.json
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 |
|
But local results are not persistent in a way that helps us compare or process data over long periods of time. This is where having the Evaluations output stored in Azure AI Foundry portal can help.