Viewing Prompt Images in Traces for 4o-mini via Llama Index

Rusty

·Mar 26, 2025 10:11 PM

Is there a way to view prompt images in traces? I'm using 4o-mini via llama index and they're not showing.

5 comments

· Sorted by Oldest

RunLLM
APP
·
Yes, Phoenix supports displaying images included in LLM traces. To view images in Phoenix, you need to ensure that your application is properly instrumented and that images are included as either base64 UTF-8 encoded images or image URLs in the calls made to your LLM. Make sure you are using the correct version of the Phoenix library and have set up the tracing environment correctly. For example, you can use the `OpenAIMultiModal` class to handle multimodal inputs, including images, in your LLM application (<https://github.com/Arize-ai/phoenix/blob/main/docs/tracing/how-to-tracing/multimodal-tracing.md|source>). If images are not showing, ensure that the image URLs or base64 data are correctly formatted and included in the trace. Additionally, check if there are any issues with the image size or format that might prevent them from being displayed (<https://github.com/Arize-ai/phoenix/issues/5564|source>). Would you like more detail on setting up the tracing environment or troubleshooting image display issues? Sources: - <https://github.com/Arize-ai/phoenix/blob/main/docs/tracing/how-to-tracing/multimodal-tracing.md|https://github.com/Arize-ai/phoenix/blob/main/docs/tracing/how-to-tracing/multimodal-tracing.md> - <https://github.com/Arize-ai/phoenix/issues/5564|https://github.com/Arize-ai/phoenix/issues/5564>
💡 Hint: Mention RunLLM in the thread for followups.
Rusty
·
The image base64 is showing on the Input tab, but the images themselves aren't displayed in Input Messages (or anywhere else I've seen).
Rusty
·
There are 20 images in this one LLM call. Maybe that's exceeding a limit for display? and/or am I using a different part of Llama Index compared to what's currently supported?
Xander S.
·
Hey Rusty, the auto-instrumentor might need to be enhanced to record the image information under the expected attributes. Can you file us an enhancement request under the OpenInference repo with a code snippet to reproduce the issue? https://github.com/Arize-ai/openinference
👍🏻1
Rusty
·
Thanks, issue created
🙏1

RunLLM
APP
·
Yes, Phoenix supports displaying images included in LLM traces. To view images in Phoenix, you need to ensure that your application is properly instrumented and that images are included as either base64 UTF-8 encoded images or image URLs in the calls made to your LLM. Make sure you are using the correct version of the Phoenix library and have set up the tracing environment correctly. For example, you can use the `OpenAIMultiModal` class to handle multimodal inputs, including images, in your LLM application (<https://github.com/Arize-ai/phoenix/blob/main/docs/tracing/how-to-tracing/multimodal-tracing.md|source>). If images are not showing, ensure that the image URLs or base64 data are correctly formatted and included in the trace. Additionally, check if there are any issues with the image size or format that might prevent them from being displayed (<https://github.com/Arize-ai/phoenix/issues/5564|source>). Would you like more detail on setting up the tracing environment or troubleshooting image display issues? Sources: - <https://github.com/Arize-ai/phoenix/blob/main/docs/tracing/how-to-tracing/multimodal-tracing.md|https://github.com/Arize-ai/phoenix/blob/main/docs/tracing/how-to-tracing/multimodal-tracing.md> - <https://github.com/Arize-ai/phoenix/issues/5564|https://github.com/Arize-ai/phoenix/issues/5564>
💡 Hint: Mention RunLLM in the thread for followups.
Rusty
·
The image base64 is showing on the Input tab, but the images themselves aren't displayed in Input Messages (or anywhere else I've seen).
Rusty
·
There are 20 images in this one LLM call. Maybe that's exceeding a limit for display? and/or am I using a different part of Llama Index compared to what's currently supported?
Xander S.
·
Hey Rusty, the auto-instrumentor might need to be enhanced to record the image information under the expected attributes. Can you file us an enhancement request under the OpenInference repo with a code snippet to reproduce the issue? https://github.com/Arize-ai/openinference
👍🏻1
Rusty
·
Thanks, issue created
🙏1