Make not accumulating the requests the default behavior for the OpenA…

…I monitor

examples/monitoring/quickstart/llms/openai_llm_monitor.ipynb

@@ -115,24 +115,6 @@
     "# )"
    ]
   },
-  {
-   "cell_type": "markdown",
-   "id": "f7c3dfbc",
-   "metadata": {},
-   "source": [
-    "You can also access all the data accumulated (and in this case, published to Openlayer) with the `data` attribute:"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "id": "27bb2bdc",
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "openai_monitor.data"
-   ]
-  },
   {
    "cell_type": "code",
    "execution_count": null,

openlayer/llm_monitors.py

@@ -23,6 +23,10 @@ class OpenAIMonitor:
         Whether to publish the data to Openlayer as soon as it is available. If True,
         the Openlayer credentials must be provided (either as keyword arguments or as
         environment variables).
+    accumulate_data : bool, False
+        Whether to accumulate the data in a dataframe. If False (default), only the
+        latest request is stored. If True, all the requests are stored in a dataframe,
+        accessed through the `data` attribute.
     client : openai.api_client.Client, optional
         The OpenAI client. It is required if you are using openai>=1.0.0.
     openlayer_api_key : str, optional
@@ -101,6 +105,7 @@ def __init__(
         self,
         publish: bool = False,
         client=None,
+        accumulate_data: bool = False,
         openlayer_api_key: Optional[str] = None,
         openlayer_project_name: Optional[str] = None,
         openlayer_inference_pipeline_name: Optional[str] = None,
@@ -134,6 +139,7 @@ def __init__(
 
         self.df = pd.DataFrame(columns=["input", "output", "tokens", "latency"])
         self.publish = publish
+        self.accumulate_data = accumulate_data
         self.monitoring_on = False
 
     def _initialize_openlayer(
@@ -311,7 +317,10 @@ def _append_row_to_df(
                 }
             ]
         )
-        self.df = pd.concat([self.df, row], ignore_index=True)
+        if self.accumulate_data:
+            self.df = pd.concat([self.df, row], ignore_index=True)
+        else:
+            self.df = row
         self.df = self.df.astype(
             {"input": object, "output": object, "tokens": int, "latency": float}
         )