do not send experiments i/o and metadata as strings so they can be searchable

Author: gary-huang

ddtrace/llmobs/_experiment.py

@@ -457,9 +457,11 @@ def _process_record(self, idx_record: Tuple[int, DatasetRecord], run: _Experimen
             except Exception:
                 span.set_exc_info(*sys.exc_info())
             self._llmobs_instance.annotate(span, input_data=input_data, output_data=output_data, tags=tags)
-            span._set_ctx_item(EXPERIMENT_EXPECTED_OUTPUT, safe_json(record["expected_output"]))
+
+            span._set_ctx_item(EXPERIMENT_EXPECTED_OUTPUT, record["expected_output"])
             if "metadata" in record:
                 span._set_ctx_item(EXPERIMENT_RECORD_METADATA, record["metadata"])
+
             return {
                 "idx": idx,
                 "span_id": span_id,

ddtrace/llmobs/_llmobs.py

@@ -1609,9 +1609,9 @@ def _tag_freeform_io(cls, span, input_value=None, output_value=None):
         arbitrary structured or non structured IO values in its spans
         """
         if input_value is not None:
-            span._set_ctx_item(EXPERIMENTS_INPUT, safe_json(input_value))
+            span._set_ctx_item(EXPERIMENTS_INPUT, input_value)
         if output_value is not None:
-            span._set_ctx_item(EXPERIMENTS_OUTPUT, safe_json(output_value))
+            span._set_ctx_item(EXPERIMENTS_OUTPUT, output_value)
 
     @staticmethod
     def _set_dict_attribute(span: Span, key, value: Dict[str, Any]) -> None:

setup.cfg

@@ -1,21 +1,10 @@
+[bdist_wheel]
+universal=1
+
 [codespell]
 skip = *.json,*.h,*.cpp,*.c,.riot,.tox,.mypy_cache,.git,*ddtrace/vendor,tests/contrib/openai/cassettes/*,tests/contrib/langchain/cassettes/*,ddtrace/appsec/_iast/_taint_tracking/_vendor/*
 exclude-file = .codespellignorelines
 ignore-words-list = asend,dne,fo,medias,ment,nin,ot,setttings,statics,ba,spawnve,doas
 
-# DEV: We use `conftest.py` as a local pytest plugin to configure hooks for collection
-[tool:pytest]
-# --cov-report is intentionally empty else pytest-cov will default to generating a report
-addopts =
-  --cov=ddtrace/
-  --cov=tests/
-  --cov-append
-  --cov-report=
-  --durations=10
-  --junitxml=test-results/junit.xml
-# DEV: The default is `test_*\.py` which will miss `test.py` files
-python_files = test*\.py
-asyncio_mode = auto
-
 [flake8]
 max-line-length = 120