Adding telemetry for datagen

assets/training/distillation/components/data_generation/src/common/utils.py

@@ -15,6 +15,7 @@ from azureml.acft.common_components import get_logger_app
 from azureml.core import Run, Workspace
 from azureml.core.run import _OfflineRun
 from typing import List, Tuple, Union
+from timeit import default_timer as timer
 
 from common.constants import (
     REQUESTS_RETRY_DELAY,
@@ -348,3 +349,45 @@ def validate_student_model_details(model_asset_id: str) -> Tuple[str, str, str]:
         Tuple[str, str, str]: Tuple containing registry name, model name and model version
     """
     return _get_model_details(model_asset_id, SUPPORTED_STUDENT_MODEL_MAP)
+
+
+def format_time(sec):
+    if sec < 1e-6:
+        return '%8.2f ns' % (sec * 1e9)
+    elif sec < 1e-3:
+        return '%8.2f mks' % (sec * 1e6)
+    elif sec < 1:
+        return '%8.2f ms' % (sec * 1e3)
+    else:
+        return '%8.2f s' % sec
+
+
+time_formatters = {
+    'auto': format_time,
+    'ns': lambda sec: '%8.2f ns' % (sec * 1e9),
+    'mks': lambda sec: '%8.2f mks' % (sec * 1e6),
+    'ms': lambda sec: '%8.2f ms' % (sec * 1e3),
+    's': lambda sec: '%8.2f s' % sec,
+}
+
+
+class log_durations(object):
+    """Times each function call or block execution."""
+
+    def __init__(self, print_func, label=None, unit='auto', threshold=-1, repr_len=25):
+        self.print_func = print_func
+        self.label = label
+        if unit not in time_formatters:
+            raise ValueError('Unknown time unit: %s. It should be ns, mks, ms, s or auto.' % unit)
+        self.format_time = time_formatters[unit]
+        self.threshold = threshold
+
+    def __enter__(self):
+        self.start = timer()
+        return self
+
+    def __exit__(self, *exc):
+        duration = timer() - self.start
+        if duration >= self.threshold:
+            duration_str = self.format_time(duration)
+            self.print_func("%s : %s" % (self.label, duration_str) if self.label else duration_str)

assets/training/distillation/components/data_generation/src/generate_data.py

@@ -51,6 +51,7 @@ from common.utils import (
     get_endpoint_details,
     validate_teacher_model_details,
     retry,
+    log_durations
 )
 
 
@@ -197,7 +198,7 @@ def get_parser():
 
 
 @retry(3)
-def _invoke_endpoint(url: str, key: str, data: dict) -> Response:
+def _invoke_endpoint(url: str, key: str, data: dict, log_metadata: dict = None) -> Response:
     """Invoke endpoint with payload data.
 
     Args:
@@ -212,8 +213,10 @@ def _invoke_endpoint(url: str, key: str, data: dict) -> Response:
         "Content-Type": "application/json",
         "Authorization": f"Bearer {key}"
     }
-    response = requests.post(url, headers=request_headers, data=json.dumps(data))
-    return response
+    log_metadata = log_metadata or {}
+    with log_durations(logger, f"POST request to teacher model endpoint: {log_metadata}"):
+        response = requests.post(url, headers=request_headers, data=json.dumps(data))
+        return response
 
 
 def _validate_file_paths_with_supported_formats(file_paths: List[Optional[str]]):
@@ -337,7 +340,7 @@ def generate_synthetic_data(
                             }
             messages = normalize_messages(messages)
             synthetic_responses = []
-            for message in messages:
+            for turn_id, message in enumerate(messages):
                 role = message['role']
                 if role == 'system':
                     synthetic_responses.append(process_system_prompt(message))
@@ -349,7 +352,8 @@ def generate_synthetic_data(
                         data_with_inference_parameters[key] = value
                     # replace the assistant content from the model
                     response: Response = _invoke_endpoint(url=url, key=endpoint_key,
-                                                          data=data_with_inference_parameters)
+                                                          data=data_with_inference_parameters,
+                                                          log_metadata={"idx": idx, 'turn': turn_id})
                     if response.status_code != 200:
                         break
                     response_data = response.json()