opentelemetry-instrumentation-openai 0.18.0__py3-none-any.whl → 0.18.2__py3-none-any.whl

opentelemetry/instrumentation/openai/shared/__init__.py

@@ -112,7 +112,9 @@ def _set_request_attributes(span, kwargs):
     _set_span_attribute(
         span, SpanAttributes.LLM_REQUEST_MAX_TOKENS, kwargs.get("max_tokens")
     )
-    _set_span_attribute(span, SpanAttributes.LLM_REQUEST_TEMPERATURE, kwargs.get("temperature"))
+    _set_span_attribute(
+        span, SpanAttributes.LLM_REQUEST_TEMPERATURE, kwargs.get("temperature")
+    )
     _set_span_attribute(span, SpanAttributes.LLM_REQUEST_TOP_P, kwargs.get("top_p"))
     _set_span_attribute(
         span, SpanAttributes.LLM_FREQUENCY_PENALTY, kwargs.get("frequency_penalty")
@@ -139,6 +141,10 @@ def _set_response_attributes(span, response):
 
     _set_span_attribute(span, SpanAttributes.LLM_RESPONSE_MODEL, response.get("model"))
 
+    _set_span_attribute(
+        span, "gen_ai.openai.system_fingerprint", response.get("system_fingerprint")
+    )
+
     usage = response.get("usage")
     if not usage:
         return

opentelemetry/instrumentation/openai/shared/chat_wrappers.py

@@ -87,18 +87,32 @@ def chat_wrapper(
 
     if is_streaming_response(response):
         # span will be closed after the generator is done
-        return ChatStream(
-            span,
-            response,
-            instance,
-            token_counter,
-            choice_counter,
-            duration_histogram,
-            streaming_time_to_first_token,
-            streaming_time_to_generate,
-            start_time,
-            kwargs,
-        )
+        if is_openai_v1():
+            return ChatStream(
+                span,
+                response,
+                instance,
+                token_counter,
+                choice_counter,
+                duration_histogram,
+                streaming_time_to_first_token,
+                streaming_time_to_generate,
+                start_time,
+                kwargs,
+            )
+        else:
+            return _build_from_streaming_response(
+                span,
+                response,
+                instance,
+                token_counter,
+                choice_counter,
+                duration_histogram,
+                streaming_time_to_first_token,
+                streaming_time_to_generate,
+                start_time,
+                kwargs,
+            )
 
     duration = end_time - start_time
 
@@ -161,18 +175,32 @@ async def achat_wrapper(
 
     if is_streaming_response(response):
         # span will be closed after the generator is done
-        return ChatStream(
-            span,
-            response,
-            instance,
-            token_counter,
-            choice_counter,
-            duration_histogram,
-            streaming_time_to_first_token,
-            streaming_time_to_generate,
-            start_time,
-            kwargs,
-        )
+        if is_openai_v1():
+            return ChatStream(
+                span,
+                response,
+                instance,
+                token_counter,
+                choice_counter,
+                duration_histogram,
+                streaming_time_to_first_token,
+                streaming_time_to_generate,
+                start_time,
+                kwargs,
+            )
+        else:
+            return _abuild_from_streaming_response(
+                span,
+                response,
+                instance,
+                token_counter,
+                choice_counter,
+                duration_histogram,
+                streaming_time_to_first_token,
+                streaming_time_to_generate,
+                start_time,
+                kwargs,
+            )
 
     duration = end_time - start_time
 
@@ -242,6 +270,7 @@ def _set_chat_metrics(
     shared_attributes = {
         "gen_ai.response.model": response_dict.get("model") or None,
         "server.address": _get_openai_base_url(instance),
+        "stream": False,
     }
 
     # token metrics
@@ -356,7 +385,11 @@ def _set_streaming_token_metrics(
     # prompt_usage
     if request_kwargs and request_kwargs.get("messages"):
         prompt_content = ""
-        model_name = request_kwargs.get("model") or None
+        # setting the default model_name as gpt-4. As this uses the embedding "cl100k_base" that
+        # is used by most of the other model.
+        model_name = (
+            request_kwargs.get("model") or complete_response.get("model") or "gpt-4"
+        )
         for msg in request_kwargs.get("messages"):
             if msg.get("content"):
                 prompt_content += msg.get("content")
@@ -366,7 +399,9 @@ def _set_streaming_token_metrics(
     # completion_usage
     if complete_response.get("choices"):
         completion_content = ""
-        model_name = complete_response.get("model") or None
+        # setting the default model_name as gpt-4. As this uses the embedding "cl100k_base" that
+        # is used by most of the other model.
+        model_name = complete_response.get("model") or "gpt-4"
 
         for choice in complete_response.get("choices"):
             if choice.get("message") and choice.get("message").get("content"):
@@ -398,6 +433,16 @@ def _set_streaming_token_metrics(
 
 
 class ChatStream(ObjectProxy):
+    _span = None
+    _instance = None
+    _token_counter = None
+    _choice_counter = None
+    _duration_histogram = None
+    _streaming_time_to_first_token = None
+    _streaming_time_to_generate = None
+    _start_time = None
+    _request_kwargs = None
+
     def __init__(
         self,
         span,
@@ -534,3 +579,166 @@ class ChatStream(ObjectProxy):
 
         self._span.set_status(Status(StatusCode.OK))
         self._span.end()
+
+
+# Backward compatibility with OpenAI v0
+
+
+@dont_throw
+def _build_from_streaming_response(
+    span,
+    response,
+    instance=None,
+    token_counter=None,
+    choice_counter=None,
+    duration_histogram=None,
+    streaming_time_to_first_token=None,
+    streaming_time_to_generate=None,
+    start_time=None,
+    request_kwargs=None,
+):
+    complete_response = {"choices": [], "model": ""}
+
+    first_token = True
+    time_of_first_token = start_time  # will be updated when first token is received
+
+    for item in response:
+        span.add_event(name="llm.content.completion.chunk")
+
+        item_to_yield = item
+
+        if first_token and streaming_time_to_first_token:
+            time_of_first_token = time.time()
+            streaming_time_to_first_token.record(time_of_first_token - start_time)
+            first_token = False
+
+        _accumulate_stream_items(item, complete_response)
+
+        yield item_to_yield
+
+    shared_attributes = {
+        "gen_ai.response.model": complete_response.get("model") or None,
+        "server.address": _get_openai_base_url(instance),
+        "stream": True,
+    }
+
+    if not is_azure_openai(instance):
+        _set_streaming_token_metrics(
+            request_kwargs, complete_response, span, token_counter, shared_attributes
+        )
+
+    # choice metrics
+    if choice_counter and complete_response.get("choices"):
+        _set_choice_counter_metrics(
+            choice_counter, complete_response.get("choices"), shared_attributes
+        )
+
+    # duration metrics
+    if start_time and isinstance(start_time, (float, int)):
+        duration = time.time() - start_time
+    else:
+        duration = None
+    if duration and isinstance(duration, (float, int)) and duration_histogram:
+        duration_histogram.record(duration, attributes=shared_attributes)
+    if streaming_time_to_generate and time_of_first_token:
+        streaming_time_to_generate.record(time.time() - time_of_first_token)
+
+    _set_response_attributes(span, complete_response)
+
+    if should_send_prompts():
+        _set_completions(span, complete_response.get("choices"))
+
+    span.set_status(Status(StatusCode.OK))
+    span.end()
+
+
+@dont_throw
+async def _abuild_from_streaming_response(
+    span,
+    response,
+    instance=None,
+    token_counter=None,
+    choice_counter=None,
+    duration_histogram=None,
+    streaming_time_to_first_token=None,
+    streaming_time_to_generate=None,
+    start_time=None,
+    request_kwargs=None,
+):
+    complete_response = {"choices": [], "model": ""}
+
+    first_token = True
+    time_of_first_token = start_time  # will be updated when first token is received
+
+    async for item in response:
+        span.add_event(name="llm.content.completion.chunk")
+
+        item_to_yield = item
+
+        if first_token and streaming_time_to_first_token:
+            time_of_first_token = time.time()
+            streaming_time_to_first_token.record(time_of_first_token - start_time)
+            first_token = False
+
+        _accumulate_stream_items(item, complete_response)
+
+        yield item_to_yield
+
+    shared_attributes = {
+        "gen_ai.response.model": complete_response.get("model") or None,
+        "server.address": _get_openai_base_url(instance),
+        "stream": True,
+    }
+
+    if not is_azure_openai(instance):
+        _set_streaming_token_metrics(
+            request_kwargs, complete_response, span, token_counter, shared_attributes
+        )
+
+    # choice metrics
+    if choice_counter and complete_response.get("choices"):
+        _set_choice_counter_metrics(
+            choice_counter, complete_response.get("choices"), shared_attributes
+        )
+
+    # duration metrics
+    if start_time and isinstance(start_time, (float, int)):
+        duration = time.time() - start_time
+    else:
+        duration = None
+    if duration and isinstance(duration, (float, int)) and duration_histogram:
+        duration_histogram.record(duration, attributes=shared_attributes)
+    if streaming_time_to_generate and time_of_first_token:
+        streaming_time_to_generate.record(time.time() - time_of_first_token)
+
+    _set_response_attributes(span, complete_response)
+
+    if should_send_prompts():
+        _set_completions(span, complete_response.get("choices"))
+
+    span.set_status(Status(StatusCode.OK))
+    span.end()
+
+
+def _accumulate_stream_items(item, complete_response):
+    if is_openai_v1():
+        item = model_as_dict(item)
+
+    complete_response["model"] = item.get("model")
+
+    for choice in item.get("choices"):
+        index = choice.get("index")
+        if len(complete_response.get("choices")) <= index:
+            complete_response["choices"].append(
+                {"index": index, "message": {"content": "", "role": ""}}
+            )
+        complete_choice = complete_response.get("choices")[index]
+        if choice.get("finish_reason"):
+            complete_choice["finish_reason"] = choice.get("finish_reason")
+
+        delta = choice.get("delta")
+
+        if delta and delta.get("content"):
+            complete_choice["message"]["content"] += delta.get("content")
+        if delta and delta.get("role"):
+            complete_choice["message"]["role"] = delta.get("role")

opentelemetry/instrumentation/openai/version.py

@@ -1 +1 @@
-__version__ = "0.18.0"
+__version__ = "0.18.2"

{opentelemetry_instrumentation_openai-0.18.0.dist-info → opentelemetry_instrumentation_openai-0.18.2.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: opentelemetry-instrumentation-openai
-Version: 0.18.0
+Version: 0.18.2
 Summary: OpenTelemetry OpenAI instrumentation
 Home-page: https://github.com/traceloop/openllmetry/tree/main/packages/opentelemetry-instrumentation-openai
 License: Apache-2.0

{opentelemetry_instrumentation_openai-0.18.0.dist-info → opentelemetry_instrumentation_openai-0.18.2.dist-info}/RECORD

@@ -1,6 +1,6 @@
 opentelemetry/instrumentation/openai/__init__.py,sha256=xl3Kvqry9glVhu8VtdknfUE9FpXQ7KWAFqtVlpjE-40,1344
-opentelemetry/instrumentation/openai/shared/__init__.py,sha256=eBVYPw2Zh28p9wak6Ncba3jSt-bdIgJjzJrad6snCxo,7359
-opentelemetry/instrumentation/openai/shared/chat_wrappers.py,sha256=Gq65KbByZnKvcvqWE65_04eaq-XJ2LIHdfxIuq1AiSI,16608
+opentelemetry/instrumentation/openai/shared/__init__.py,sha256=5kEyVhz2YDHvuq2SDQOsDhtjbG7R7GCn793oLq2_J_k,7490
+opentelemetry/instrumentation/openai/shared/chat_wrappers.py,sha256=6gJUpjp1tq3eYviTVTRNq5RcF4yduhI6yAKJBSEGToI,23344
 opentelemetry/instrumentation/openai/shared/completion_wrappers.py,sha256=-JHfgyxic5I3Wr3Uc_L-U7ztDVFcyovtF37tNLtaW3s,6604
 opentelemetry/instrumentation/openai/shared/config.py,sha256=5uekQEnmYo1o6tsTD2IGc-cVmHUo5KmUC4pOVdrFFNk,102
 opentelemetry/instrumentation/openai/shared/embeddings_wrappers.py,sha256=6I6I98T1a5Np0ZjcGZQbId4ZyEmMI6o9wVm8qoRpO9o,6595
@@ -10,8 +10,8 @@ opentelemetry/instrumentation/openai/v0/__init__.py,sha256=ngmmYyfTwRQSjTZAvNpBI
 opentelemetry/instrumentation/openai/v1/__init__.py,sha256=6XHk11JhkpZixgMDsjb0b-efd8LlBTG0jjMCyf0fOSo,8652
 opentelemetry/instrumentation/openai/v1/assistant_wrappers.py,sha256=T6Vtdp1fAZdcYjGiTMZwkn4F4DgsltD4p4xLEFW-GhI,5874
 opentelemetry/instrumentation/openai/v1/event_handler_wrapper.py,sha256=SAzYoun2yyOloofyOWtxpm8E2M9TL3Nm8TgKdNyXHuY,2779
-opentelemetry/instrumentation/openai/version.py,sha256=0EHw4xygmgkGSyfwNfEoMlQyN0uHxjHtlSFF79s6120,23
-opentelemetry_instrumentation_openai-0.18.0.dist-info/METADATA,sha256=WkE3L2VnQwU5Bu68MPgqgP98uDrwu2LOR6vKlD4RpPg,2255
-opentelemetry_instrumentation_openai-0.18.0.dist-info/WHEEL,sha256=sP946D7jFCHeNz5Iq4fL4Lu-PrWrFsgfLXbbkciIZwg,88
-opentelemetry_instrumentation_openai-0.18.0.dist-info/entry_points.txt,sha256=vTBfiX5yXji5YHikuJHEOoBZ1TFdPQ1EI4ctd2pZSeE,93
-opentelemetry_instrumentation_openai-0.18.0.dist-info/RECORD,,
+opentelemetry/instrumentation/openai/version.py,sha256=GYySpgpz2Cs3F3nz_H9h8KIG60jkP6f1a--08qCTJCQ,23
+opentelemetry_instrumentation_openai-0.18.2.dist-info/METADATA,sha256=8_mfPUYiS7Wxf5kBberCnJ4KEm4VjYtk7ESz_tXiSZ0,2255
+opentelemetry_instrumentation_openai-0.18.2.dist-info/WHEEL,sha256=sP946D7jFCHeNz5Iq4fL4Lu-PrWrFsgfLXbbkciIZwg,88
+opentelemetry_instrumentation_openai-0.18.2.dist-info/entry_points.txt,sha256=vTBfiX5yXji5YHikuJHEOoBZ1TFdPQ1EI4ctd2pZSeE,93
+opentelemetry_instrumentation_openai-0.18.2.dist-info/RECORD,,