Repository navigation
Conversation
| OUTPUT_MIME_TYPE = "output.mime_type" | ||
| INPUT_VALUE = "input.value" | ||
| INPUT_MIME_TYPE = "input.mime_type" | ||
| INPUT_RAW = "input.raw" |
There was a problem hiding this comment.
Don't introduce INPUT_RAW — adding new span attributes has downstream implications. The fix should not require a new attribute.
There was a problem hiding this comment.
yea, you are right the downstream impli would help.. lemme fix it.
| if eval_input: | ||
| eval_input_str = " \n ".join(map(str, eval_input)) | ||
| span.set_attribute(SpanAttributes.INPUT_VALUE, eval_input_str) | ||
| if eval_input and len(eval_input) > 0: | ||
| span.set_attribute(SpanAttributes.INPUT_VALUE, eval_input[0]) |
There was a problem hiding this comment.
Just Deleting these lines should solve the issue of data being overwritten , no need to introduce any new span attributes
|
@JayaSurya-27 let me know if the 2d31b8f works. |
|
Thanks, Ritesh, for tracking this down and for turning the review around so quickly. The full-conversation fix is now on dev and main (it landed with #245 in our release train), and we've confirmed it with tests, so we're closing this overlapping PR. If you spot a case it still misses, please open an issue with an example and we'll take a look. |
Fixes #151
Summary
Fixes an issue where
SpanAttributes.INPUT_VALUEwas overwritten multiple times in_process_input_data, with the final assignment reducing it to only the first message.Changes
INPUT_VALUEonly once using full joined inputeval_input[0]Impact
Preserves full input context for multi-message traces, improving trace correctness and observability.
Testing
PYTHONPATH=frameworks/openai poetry run pytest tests/test_framework_openai.py -v