Revert "Revert "Logfire Integration""

.circleci/config.yml

@@ -41,6 +41,7 @@ jobs:
             pip install langchain
             pip install lunary==0.2.5
             pip install "langfuse==2.27.1"
+            pip install "logfire==0.29.0"
             pip install numpydoc
             pip install traceloop-sdk==0.18.2
             pip install openai
@@ -88,15 +89,14 @@ jobs:
               exit 1
             fi
             cd ..
-
 
       # Run pytest and generate JUnit XML report
       - run:
           name: Run tests
           command: |
             pwd
             ls
-            python -m pytest -vv litellm/tests/ -x --junitxml=test-results/junit.xml --durations=5 
+            python -m pytest -vv litellm/tests/ -x --junitxml=test-results/junit.xml --durations=5
           no_output_timeout: 120m
 
       # Store test results
@@ -172,6 +172,7 @@ jobs:
             pip install "aioboto3==12.3.0"
             pip install langchain
             pip install "langfuse>=2.0.0"
+            pip install "logfire==0.29.0"
             pip install numpydoc
             pip install prisma            
             pip install fastapi            
@@ -224,15 +225,15 @@ jobs:
           name: Start outputting logs
           command: docker logs -f my-app
           background: true
-      - run: 
+      - run:
           name: Wait for app to be ready
           command: dockerize -wait http://localhost:4000 -timeout 5m
       - run:
           name: Run tests
           command: |
             pwd
             ls
-            python -m pytest -vv tests/ -x --junitxml=test-results/junit.xml --durations=5 
+            python -m pytest -vv tests/ -x --junitxml=test-results/junit.xml --durations=5
           no_output_timeout: 120m
 
       # Store test results
@@ -254,7 +255,7 @@ jobs:
           name: Copy model_prices_and_context_window File to model_prices_and_context_window_backup
           command: |
             cp model_prices_and_context_window.json litellm/model_prices_and_context_window_backup.json
-      
+
       - run:
           name: Check if litellm dir was updated or if pyproject.toml was modified
           command: |
@@ -339,4 +340,4 @@ workflows:
           filters:
             branches:
               only:
-                - main
+                - main

docs/my-website/docs/observability/logfire_integration.md

@@ -0,0 +1,60 @@
+import Image from '@theme/IdealImage';
+
+# Logfire - Logging LLM Input/Output
+
+Logfire is open Source Observability & Analytics for LLM Apps
+Detailed production traces and a granular view on quality, cost and latency
+
+<Image img={require('../../img/logfire.png')} />
+
+:::info
+We want to learn how we can make the callbacks better! Meet the LiteLLM [founders](https://calendly.com/d/4mp-gd3-k5k/berriai-1-1-onboarding-litellm-hosted-version) or
+join our [discord](https://discord.gg/wuPM9dRgDw)
+:::
+
+## Pre-Requisites
+
+Ensure you have run `pip install logfire` for this integration
+
+```shell
+pip install logfire litellm
+```
+
+## Quick Start
+
+Get your Logfire token from [Logfire](https://logfire.pydantic.dev/)
+
+```python
+litellm.success_callback = ["logfire"]
+litellm.failure_callback = ["logfire"] # logs errors to logfire
+```
+
+```python
+# pip install logfire
+import litellm
+import os
+
+# from https://logfire.pydantic.dev/
+os.environ["LOGFIRE_TOKEN"] = ""
+
+# LLM API Keys
+os.environ['OPENAI_API_KEY']=""
+
+# set logfire as a callback, litellm will send the data to logfire
+litellm.success_callback = ["logfire"]
+
+# openai call
+response = litellm.completion(
+  model="gpt-3.5-turbo",
+  messages=[
+    {"role": "user", "content": "Hi 👋 - i'm openai"}
+  ]
+)
+```
+
+## Support & Talk to Founders
+
+- [Schedule Demo 👋](https://calendly.com/d/4mp-gd3-k5k/berriai-1-1-onboarding-litellm-hosted-version)
+- [Community Discord 💭](https://discord.gg/wuPM9dRgDw)
+- Our numbers 📞 +1 (770) 8783-106 / ‭+1 (412) 618-6238‬
+- Our emails ✉️ ishaan@berri.ai / krrish@berri.ai

litellm/integrations/logfire_logger.py

@@ -0,0 +1,178 @@
+#### What this does ####
+#    On success + failure, log events to Logfire
+
+import dotenv, os
+
+dotenv.load_dotenv()  # Loading env variables using dotenv
+import traceback
+import uuid
+from litellm._logging import print_verbose, verbose_logger
+
+from enum import Enum
+from typing import Any, Dict, NamedTuple
+from typing_extensions import LiteralString
+
+
+class SpanConfig(NamedTuple):
+    message_template: LiteralString
+    span_data: Dict[str, Any]
+
+
+class LogfireLevel(str, Enum):
+    INFO = "info"
+    ERROR = "error"
+
+
+class LogfireLogger:
+    # Class variables or attributes
+    def __init__(self):
+        try:
+            verbose_logger.debug(f"in init logfire logger")
+            import logfire
+
+            # only setting up logfire if we are sending to logfire
+            # in testing, we don't want to send to logfire
+            if logfire.DEFAULT_LOGFIRE_INSTANCE.config.send_to_logfire:
+                logfire.configure(token=os.getenv("LOGFIRE_TOKEN"))
+        except Exception as e:
+            print_verbose(f"Got exception on init logfire client {str(e)}")
+            raise e
+
+    def _get_span_config(self, payload) -> SpanConfig:
+        if (
+            payload["call_type"] == "completion"
+            or payload["call_type"] == "acompletion"
+        ):
+            return SpanConfig(
+                message_template="Chat Completion with {request_data[model]!r}",
+                span_data={"request_data": payload},
+            )
+        elif (
+            payload["call_type"] == "embedding" or payload["call_type"] == "aembedding"
+        ):
+            return SpanConfig(
+                message_template="Embedding Creation with {request_data[model]!r}",
+                span_data={"request_data": payload},
+            )
+        elif (
+            payload["call_type"] == "image_generation"
+            or payload["call_type"] == "aimage_generation"
+        ):
+            return SpanConfig(
+                message_template="Image Generation with {request_data[model]!r}",
+                span_data={"request_data": payload},
+            )
+        else:
+            return SpanConfig(
+                message_template="Litellm Call with {request_data[model]!r}",
+                span_data={"request_data": payload},
+            )
+
+    async def _async_log_event(
+        self,
+        kwargs,
+        response_obj,
+        start_time,
+        end_time,
+        print_verbose,
+        level: LogfireLevel,
+    ):
+        self.log_event(
+            kwargs=kwargs,
+            response_obj=response_obj,
+            start_time=start_time,
+            end_time=end_time,
+            print_verbose=print_verbose,
+            level=level,
+        )
+
+    def log_event(
+        self,
+        kwargs,
+        start_time,
+        end_time,
+        print_verbose,
+        level: LogfireLevel,
+        response_obj,
+    ):
+        try:
+            import logfire
+
+            verbose_logger.debug(
+                f"logfire Logging - Enters logging function for model {kwargs}"
+            )
+
+            if not response_obj:
+                response_obj = {}
+            litellm_params = kwargs.get("litellm_params", {})
+            metadata = (
+                litellm_params.get("metadata", {}) or {}
+            )  # if litellm_params['metadata'] == None
+            messages = kwargs.get("messages")
+            optional_params = kwargs.get("optional_params", {})
+            call_type = kwargs.get("call_type", "completion")
+            cache_hit = kwargs.get("cache_hit", False)
+            usage = response_obj.get("usage", {})
+            id = response_obj.get("id", str(uuid.uuid4()))
+            try:
+                response_time = (end_time - start_time).total_seconds()
+            except:
+                response_time = None
+
+            # Clean Metadata before logging - never log raw metadata
+            # the raw metadata can contain circular references which leads to infinite recursion
+            # we clean out all extra litellm metadata params before logging
+            clean_metadata = {}
+            if isinstance(metadata, dict):
+                for key, value in metadata.items():
+                    # clean litellm metadata before logging
+                    if key in [
+                        "endpoint",
+                        "caching_groups",
+                        "previous_models",
+                    ]:
+                        continue
+                    else:
+                        clean_metadata[key] = value
+
+            # Build the initial payload
+            payload = {
+                "id": id,
+                "call_type": call_type,
+                "cache_hit": cache_hit,
+                "startTime": start_time,
+                "endTime": end_time,
+                "responseTime (seconds)": response_time,
+                "model": kwargs.get("model", ""),
+                "user": kwargs.get("user", ""),
+                "modelParameters": optional_params,
+                "spend": kwargs.get("response_cost", 0),
+                "messages": messages,
+                "response": response_obj,
+                "usage": usage,
+                "metadata": clean_metadata,
+            }
+            logfire_openai = logfire.with_settings(custom_scope_suffix="openai")
+            message_template, span_data = self._get_span_config(payload)
+            if level == LogfireLevel.INFO:
+                logfire_openai.info(
+                    message_template,
+                    **span_data,
+                )
+            elif level == LogfireLevel.ERROR:
+                logfire_openai.error(
+                    message_template,
+                    **span_data,
+                    _exc_info=True,
+                )
+            print_verbose(f"\ndd Logger - Logging payload = {payload}")
+
+            print_verbose(
+                f"Logfire Layer Logging - final response object: {response_obj}"
+            )
+        except Exception as e:
+            traceback.print_exc()
+            verbose_logger.debug(
+                f"Logfire Layer Error - {str(e)}\n{traceback.format_exc()}"
+            )
+            pass