opentelemetry-instrumentation-openai 0.21.3__tar.gz → 0.21.4__tar.gz

{opentelemetry_instrumentation_openai-0.21.3 → opentelemetry_instrumentation_openai-0.21.4}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: opentelemetry-instrumentation-openai
-Version: 0.21.3
+Version: 0.21.4
 Summary: OpenTelemetry OpenAI instrumentation
 Home-page: https://github.com/traceloop/openllmetry/tree/main/packages/opentelemetry-instrumentation-openai
 License: Apache-2.0

{opentelemetry_instrumentation_openai-0.21.3 → opentelemetry_instrumentation_openai-0.21.4}/opentelemetry/instrumentation/openai/shared/__init__.py

@@ -245,3 +245,24 @@ def get_token_count_from_string(string: str, model_name: str):
 
     token_count = len(encoding.encode(string))
     return token_count
+
+
+def _token_type(token_type: str):
+    if token_type == "prompt_tokens":
+        return "input"
+    elif token_type == "completion_tokens":
+        return "output"
+
+    return None
+
+
+def _metric_shared_attributes(
+    response_model: str, operation: str, server_address: str, is_streaming: bool = False
+):
+    return {
+        "gen_ai.system": "openai",
+        "gen_ai.response.model": response_model,
+        "gen_ai.operation.name": operation,
+        "server.address": server_address,
+        "stream": is_streaming,
+    }

{opentelemetry_instrumentation_openai-0.21.3 → opentelemetry_instrumentation_openai-0.21.4}/opentelemetry/instrumentation/openai/shared/chat_wrappers.py

@@ -14,10 +14,12 @@ from opentelemetry.instrumentation.openai.utils import (
     dont_throw,
 )
 from opentelemetry.instrumentation.openai.shared import (
+    _metric_shared_attributes,
     _set_client_attributes,
     _set_request_attributes,
     _set_span_attribute,
     _set_functions_attributes,
+    _token_type,
     set_tools_attributes,
     _set_response_attributes,
     is_streaming_response,
@@ -267,11 +269,12 @@ def _handle_response(
 def _set_chat_metrics(
     instance, token_counter, choice_counter, duration_histogram, response_dict, duration
 ):
-    shared_attributes = {
-        "gen_ai.response.model": response_dict.get("model") or None,
-        "server.address": _get_openai_base_url(instance),
-        "stream": False,
-    }
+    shared_attributes = _metric_shared_attributes(
+        response_model=response_dict.get("model") or None,
+        operation="chat",
+        server_address=_get_openai_base_url(instance),
+        is_streaming=False,
+    )
 
     # token metrics
     usage = response_dict.get("usage")  # type: dict
@@ -303,7 +306,7 @@ def _set_token_counter_metrics(token_counter, usage, shared_attributes):
         if name in OPENAI_LLM_USAGE_TOKEN_TYPES:
             attributes_with_token_type = {
                 **shared_attributes,
-                "llm.usage.token_type": name.split("_")[0],
+                "gen_ai.token.type": _token_type(name),
             }
             token_counter.record(val, attributes=attributes_with_token_type)
 
@@ -420,14 +423,14 @@ def _set_streaming_token_metrics(
         if type(prompt_usage) is int and prompt_usage >= 0:
             attributes_with_token_type = {
                 **shared_attributes,
-                "llm.usage.token_type": "prompt",
+                "gen_ai.token.type": "input",
             }
             token_counter.record(prompt_usage, attributes=attributes_with_token_type)
 
         if type(completion_usage) is int and completion_usage >= 0:
             attributes_with_token_type = {
                 **shared_attributes,
-                "llm.usage.token_type": "completion",
+                "gen_ai.token.type": "output",
             }
             token_counter.record(
                 completion_usage, attributes=attributes_with_token_type
@@ -523,7 +526,8 @@ class ChatStream(ObjectProxy):
         if self._first_token and self._streaming_time_to_first_token:
             self._time_of_first_token = time.time()
             self._streaming_time_to_first_token.record(
-                self._time_of_first_token - self._start_time
+                self._time_of_first_token - self._start_time,
+                attributes=self._shared_attributes(),
             )
             self._first_token = False
 
@@ -549,20 +553,24 @@ class ChatStream(ObjectProxy):
             if delta and delta.get("role"):
                 complete_choice["message"]["role"] = delta.get("role")
 
-    def _close_span(self):
-        shared_attributes = {
-            "gen_ai.response.model": self._complete_response.get("model") or None,
-            "server.address": _get_openai_base_url(self._instance),
-            "stream": True,
-        }
+    def _shared_attributes(self):
+        return _metric_shared_attributes(
+            response_model=self._complete_response.get("model")
+            or self._request_kwargs.get("model")
+            or None,
+            operation="chat",
+            server_address=_get_openai_base_url(self._instance),
+            is_streaming=True,
+        )
 
+    def _close_span(self):
         if not is_azure_openai(self._instance):
             _set_streaming_token_metrics(
                 self._request_kwargs,
                 self._complete_response,
                 self._span,
                 self._token_counter,
-                shared_attributes,
+                self._shared_attributes(),
             )
 
         # choice metrics
@@ -570,7 +578,7 @@ class ChatStream(ObjectProxy):
             _set_choice_counter_metrics(
                 self._choice_counter,
                 self._complete_response.get("choices"),
-                shared_attributes,
+                self._shared_attributes(),
             )
 
         # duration metrics
@@ -579,10 +587,13 @@ class ChatStream(ObjectProxy):
         else:
             duration = None
         if duration and isinstance(duration, (float, int)) and self._duration_histogram:
-            self._duration_histogram.record(duration, attributes=shared_attributes)
+            self._duration_histogram.record(
+                duration, attributes=self._shared_attributes()
+            )
         if self._streaming_time_to_generate and self._time_of_first_token:
             self._streaming_time_to_generate.record(
-                time.time() - self._time_of_first_token
+                time.time() - self._time_of_first_token,
+                attributes=self._shared_attributes(),
             )
 
         _set_response_attributes(self._span, self._complete_response)

{opentelemetry_instrumentation_openai-0.21.3 → opentelemetry_instrumentation_openai-0.21.4}/opentelemetry/instrumentation/openai/shared/embeddings_wrappers.py

@@ -12,10 +12,12 @@ from opentelemetry.instrumentation.openai.utils import (
     _with_embeddings_telemetry_wrapper,
 )
 from opentelemetry.instrumentation.openai.shared import (
+    _metric_shared_attributes,
     _set_client_attributes,
     _set_request_attributes,
     _set_span_attribute,
     _set_response_attributes,
+    _token_type,
     should_send_prompts,
     model_as_dict,
     _get_openai_base_url,
@@ -188,10 +190,11 @@ def _set_embeddings_metrics(
     response_dict,
     duration,
 ):
-    shared_attributes = {
-        "gen_ai.response.model": response_dict.get("model") or None,
-        "server.address": _get_openai_base_url(instance),
-    }
+    shared_attributes = _metric_shared_attributes(
+        response_model=response_dict.get("model") or None,
+        operation="embeddings",
+        server_address=_get_openai_base_url(instance),
+    )
 
     # token count metrics
     usage = response_dict.get("usage")
@@ -200,7 +203,7 @@ def _set_embeddings_metrics(
             if name in OPENAI_LLM_USAGE_TOKEN_TYPES:
                 attributes_with_token_type = {
                     **shared_attributes,
-                    "llm.usage.token_type": name.split("_")[0],
+                    "gen_ai.token.type": _token_type(name),
                 }
                 token_counter.record(val, attributes=attributes_with_token_type)
 

{opentelemetry_instrumentation_openai-0.21.3 → opentelemetry_instrumentation_openai-0.21.4}/opentelemetry/instrumentation/openai/shared/image_gen_wrappers.py

@@ -8,6 +8,7 @@ from opentelemetry.metrics import Counter, Histogram
 from opentelemetry.instrumentation.openai import is_openai_v1
 from opentelemetry.instrumentation.openai.shared import (
     _get_openai_base_url,
+    _metric_shared_attributes,
     model_as_dict,
 )
 from opentelemetry.instrumentation.openai.utils import (
@@ -52,11 +53,12 @@ def image_gen_metrics_wrapper(
     else:
         response_dict = response
 
-    shared_attributes = {
-        # not provide response.model in ImagesResponse response, use model in request kwargs
-        "gen_ai.response.model": kwargs.get("model") or None,
-        "server.address": _get_openai_base_url(instance),
-    }
+    # not provide response.model in ImagesResponse response, use model in request kwargs
+    shared_attributes = _metric_shared_attributes(
+        response_model=kwargs.get("model") or None,
+        operation="image_gen",
+        server_address=_get_openai_base_url(instance),
+    )
 
     duration = end_time - start_time
     if duration_histogram:

{opentelemetry_instrumentation_openai-0.21.3 → opentelemetry_instrumentation_openai-0.21.4}/opentelemetry/instrumentation/openai/v0/__init__.py

@@ -38,7 +38,7 @@ class OpenAIV0Instrumentor(BaseInstrumentor):
             tokens_histogram = meter.create_histogram(
                 name="gen_ai.client.token.usage",
                 unit="token",
-                description="Number of tokens used in prompt and completions",
+                description="Measures number of input and output tokens used",
             )
 
             chat_choice_counter = meter.create_counter(
@@ -50,7 +50,7 @@ class OpenAIV0Instrumentor(BaseInstrumentor):
             chat_duration_histogram = meter.create_histogram(
                 name="gen_ai.client.operation.duration",
                 unit="s",
-                description="Duration of chat completion operation",
+                description="GenAI operation duration",
             )
 
             chat_exception_counter = meter.create_counter(

{opentelemetry_instrumentation_openai-0.21.3 → opentelemetry_instrumentation_openai-0.21.4}/opentelemetry/instrumentation/openai/v1/__init__.py

@@ -52,7 +52,7 @@ class OpenAIV1Instrumentor(BaseInstrumentor):
             tokens_histogram = meter.create_histogram(
                 name="gen_ai.client.token.usage",
                 unit="token",
-                description="Number of tokens used in prompt and completions",
+                description="Measures number of input and output tokens used",
             )
 
             chat_choice_counter = meter.create_counter(
@@ -64,7 +64,7 @@ class OpenAIV1Instrumentor(BaseInstrumentor):
             chat_duration_histogram = meter.create_histogram(
                 name="gen_ai.client.operation.duration",
                 unit="s",
-                description="Duration of chat completion operation",
+                description="GenAI operation duration",
             )
 
             chat_exception_counter = meter.create_counter(

opentelemetry_instrumentation_openai-0.21.4/opentelemetry/instrumentation/openai/version.py

@@ -0,0 +1 @@
+__version__ = "0.21.4"

{opentelemetry_instrumentation_openai-0.21.3 → opentelemetry_instrumentation_openai-0.21.4}/pyproject.toml

@@ -8,7 +8,7 @@ show_missing = true
 
 [tool.poetry]
 name = "opentelemetry-instrumentation-openai"
-version = "0.21.3"
+version = "0.21.4"
 description = "OpenTelemetry OpenAI instrumentation"
 authors = [
   "Gal Kleinman <gal@traceloop.com>",

opentelemetry_instrumentation_openai-0.21.3/opentelemetry/instrumentation/openai/version.py

@@ -1 +0,0 @@
-__version__ = "0.21.3"