Issue with RETRIEVER Trace Logging for Input Value

Issue with RETRIEVER Trace Logging for Input Value | Arize AI Community

4 comments

· Sorted by Oldest

RunLLM
APP
·
The issue might be related to how the input.value attribute is being set or logged. Ensure that the input.value is correctly set before any other operations that might affect the span. Additionally, verify that the span is active and not closed when setting the attribute. If the issue persists, consider checking for any errors or exceptions in the logging process that might prevent the input.value from being recorded.Would you like more detail on how to troubleshoot this issue?Sources: - https://docs.arize.com/phoenix/tracing/how-to-interact-with-traces/extract-data-from-spans#connect-to-phoenix - https://arize-ai.slack.com/archives/C04R3GXC8HK/p1719277912.189609
💡 Hint: Mention RunLLM in the thread for followups.
Sagar D.
·
Sorry I forgot to take off the self in the sample code above (corrected now). There really is no difference is how the span is called in both the instances. I am puzzled because the input.value is showing in another place where I have set up the trace the exact same way. I will keep digging.

A bit more of what I found: if I remove the document tracing step (like in the code-snippet below), then both the span kind and the input values are getting logged correctly. Is there a way that the document tracing steps are overwriting the span kind and input val? (screenhsots with and without the doc tracing also attached)

span.set_attribute("openinference.span.kind", "RETRIEVER")
span.set_attribute("input.value", "This is the query")

# documents = []
# for node in nodes:
#     documents.append(
#         {"document.id": node.node_id, "document.content": node.text, "document.score": node.score}
#     )

# for i, obj in enumerate(documents):
#     for key, value in obj.items():
#         span.set_attribute(f"retrieval.documents.{i}.{key}", value)

Sagar D.

A bit more on this: When I reordered the kind and input to follow after the document tracing step. the traces started flowing as they should. So the issue is fixed, but you may want to check what is causing the kind and inout traces to be overwritten if the ordering is reversed.

documents = []
for node in nodes:
    documents.append(
        {"document.id": node.node_id, "document.content": node.text, "document.score": node.score}
    )

for i, obj in enumerate(documents):
    for key, value in obj.items():
        span.set_attribute(f"retrieval.documents.{i}.{key}", value)

span.set_attribute("openinference.span.kind", "RETRIEVER")
span.set_attribute("input.value", "This is the query")

RunLLM
APP
·
The issue might be related to how the input.value attribute is being set or logged. Ensure that the input.value is correctly set before any other operations that might affect the span. Additionally, verify that the span is active and not closed when setting the attribute. If the issue persists, consider checking for any errors or exceptions in the logging process that might prevent the input.value from being recorded.Would you like more detail on how to troubleshoot this issue?Sources: - https://docs.arize.com/phoenix/tracing/how-to-interact-with-traces/extract-data-from-spans#connect-to-phoenix - https://arize-ai.slack.com/archives/C04R3GXC8HK/p1719277912.189609
💡 Hint: Mention RunLLM in the thread for followups.
Sagar D.
·
Sorry I forgot to take off the self in the sample code above (corrected now). There really is no difference is how the span is called in both the instances. I am puzzled because the input.value is showing in another place where I have set up the trace the exact same way. I will keep digging.

Sagar D.

span.set_attribute("openinference.span.kind", "RETRIEVER")
span.set_attribute("input.value", "This is the query")

# documents = []
# for node in nodes:
#     documents.append(
#         {"document.id": node.node_id, "document.content": node.text, "document.score": node.score}
#     )

# for i, obj in enumerate(documents):
#     for key, value in obj.items():
#         span.set_attribute(f"retrieval.documents.{i}.{key}", value)

Sagar D.

documents = []
for node in nodes:
    documents.append(
        {"document.id": node.node_id, "document.content": node.text, "document.score": node.score}
    )

for i, obj in enumerate(documents):
    for key, value in obj.items():
        span.set_attribute(f"retrieval.documents.{i}.{key}", value)

span.set_attribute("openinference.span.kind", "RETRIEVER")
span.set_attribute("input.value", "This is the query")