Troubleshooting Span Types in Phoenix for LlamaIndex Retrieval Evals

·Sep 18, 2024 10:52 PM

Hi I am using the hosted version of Phoenix to collect traces from LlamaIndex. So far, I have chunked, embedded, ingested and retrieved documents using LlamaIndex. I am trying to view spans like so:

spans_df = px_client.get_spans_dataframe()

Per this tutorial, I should see different span kinds such as chain, embedding, retriever etc. but I am only seeing spans of type chain. I'm trying to do retrieval evals so was hoping to see a attributes.retrieval.documents column in the spans_df but am not seeing it. Any idea why?

3 comments

· Sorted by Oldest

RunLLM
APP
·
The issue might be related to how the spans are being generated and managed within LlamaIndex. Specifically, the retriever spans might not be correctly propagated or captured. According to this GitHub issue, there can be fragmentation in the parent-child graph of spans due to different callback managers, which might result in missing retriever spans.Would you like more detail on how to ensure retriever spans are correctly captured and propagated in your setup?Sources: - this GitHub issue
💡 Hint: Mention RunLLM in the thread for followups.
Apoorva J.
·
Seeing embedding as well in the UI
Apoorva J.
·
Hah ok I see them when I query for them directly. Thanks!
👍1
🙌1

RunLLM
APP
·
The issue might be related to how the spans are being generated and managed within LlamaIndex. Specifically, the retriever spans might not be correctly propagated or captured. According to this GitHub issue, there can be fragmentation in the parent-child graph of spans due to different callback managers, which might result in missing retriever spans.Would you like more detail on how to ensure retriever spans are correctly captured and propagated in your setup?Sources: - this GitHub issue
💡 Hint: Mention RunLLM in the thread for followups.
Apoorva J.
·
Seeing embedding as well in the UI
Apoorva J.
·
Hah ok I see them when I query for them directly. Thanks!
👍1
🙌1