Merge pull request #157 from Chainlit/clement/eng-2181-otel-exporter

Clement/eng 2181 otel exporter

Author: willydouhard

examples/langchain_toolcall.py

@@ -19,6 +19,7 @@
 tools = [search]
 
 lai_client = LiteralClient()
+lai_client.initialize()
 lai_prompt = lai_client.api.get_or_create_prompt(
     name="LC Agent",
     settings={
@@ -37,13 +38,13 @@
         {"role": "assistant", "content": "{{agent_scratchpad}}"},
     ],
 )
-prompt = lai_prompt.to_langchain_chat_prompt_template()
+prompt = lai_prompt.to_langchain_chat_prompt_template(
+    additional_messages=[("placeholder", "{agent_scratchpad}")],
+)
 
 agent: BaseSingleActionAgent = create_tool_calling_agent(model, tools, prompt)  # type: ignore
 agent_executor = AgentExecutor(agent=agent, tools=tools)
 
-cb = lai_client.langchain_callback()
-
 # Replace with ainvoke for asynchronous execution.
 agent_executor.invoke(
     {
@@ -56,5 +57,5 @@
         ],
         "input": "whats the weather in sf?",
     },
-    config=RunnableConfig(callbacks=[cb], run_name="Weather SF"),
+    config=RunnableConfig(run_name="Weather SF"),
 )

examples/langchain_variable.py

@@ -1,12 +1,13 @@
 from langchain.chat_models import init_chat_model
 from literalai import LiteralClient
-from langchain.schema.runnable.config import RunnableConfig
+
 
 from dotenv import load_dotenv
 
 load_dotenv()
 
 lai = LiteralClient()
+lai.initialize()
 
 prompt = lai.api.get_or_create_prompt(
     name="user intent",
@@ -29,13 +30,14 @@
 input_messages = messages.format_messages(
     user_message="The screen is cracked, there are scratches on the surface, and a component is missing."
 )
-cb = lai.langchain_callback()
 
 # Returns a langchain_openai.ChatOpenAI instance.
 gpt_4o = init_chat_model(  # type: ignore
     model_provider=prompt.provider,
     **prompt.settings,
 )
-print(gpt_4o.invoke(input_messages, config=RunnableConfig(callbacks=[cb])))
+
+lai.set_properties(prompt=prompt)
+print(gpt_4o.invoke(input_messages))
 
 lai.flush_and_stop()

examples/llamaindex_workflow.py

@@ -0,0 +1,55 @@
+import asyncio
+from llama_index.core.workflow import (
+    Event,
+    StartEvent,
+    StopEvent,
+    Workflow,
+    step,
+)
+from llama_index.llms.openai import OpenAI
+from literalai.client import LiteralClient
+
+lai_client = LiteralClient()
+lai_client.initialize()
+
+
+class JokeEvent(Event):
+    joke: str
+
+class RewriteJoke(Event): 
+    joke: str
+
+
+class JokeFlow(Workflow):
+    llm = OpenAI()
+
+    @step()
+    async def generate_joke(self, ev: StartEvent) -> JokeEvent:
+        topic = ev.topic
+
+        prompt = f"Write your best joke about {topic}."
+        response = await self.llm.acomplete(prompt)
+        return JokeEvent(joke=str(response))
+
+    @step()
+    async def return_joke(self, ev: JokeEvent) -> RewriteJoke:
+        return RewriteJoke(joke=ev.joke + "What is funny?")
+
+    @step()
+    async def critique_joke(self, ev: RewriteJoke) -> StopEvent:
+        joke = ev.joke
+
+        prompt = f"Give a thorough analysis and critique of the following joke: {joke}"
+        response = await self.llm.acomplete(prompt)
+        return StopEvent(result=str(response))
+
+
+@lai_client.thread(name="JokeFlow")
+async def main():
+    w = JokeFlow(timeout=60, verbose=False)
+    result = await w.run(topic="pirates")
+    print(str(result))
+
+
+if __name__ == "__main__":
+    asyncio.run(main())

examples/multimodal.py

@@ -0,0 +1,83 @@
+import base64
+import requests  # type: ignore
+import time
+
+from literalai import LiteralClient
+from openai import OpenAI
+
+from dotenv import load_dotenv
+
+
+load_dotenv()
+
+openai_client = OpenAI()
+
+literalai_client = LiteralClient()
+literalai_client.initialize()
+
+
+def encode_image(url):
+    return base64.b64encode(requests.get(url).content)
+
+
+@literalai_client.step(type="run")
+def generate_answer(user_query, image_url):
+    literalai_client.set_properties(
+        name="foobar",
+        metadata={"foo": "bar"},
+        tags=["foo", "bar"],
+    )
+    completion = openai_client.chat.completions.create(
+        model="gpt-4o-mini",
+        messages=[
+            {
+                "role": "user",
+                "content": [
+                    {"type": "text", "text": user_query},
+                    {
+                        "type": "image_url",
+                        "image_url": {"url": image_url},
+                    },
+                ],
+            },
+        ],
+        max_tokens=300,
+    )
+    return completion.choices[0].message.content
+
+
+def main():
+    with literalai_client.thread(name="Meal Analyzer") as thread:
+        welcome_message = (
+            "Welcome to the meal analyzer, please upload an image of your plate!"
+        )
+        literalai_client.message(
+            content=welcome_message, type="assistant_message", name="My Assistant"
+        )
+
+        user_query = "Is this a healthy meal?"
+        user_image = "https://www.eatthis.com/wp-content/uploads/sites/4/2021/05/healthy-plate.jpg"
+        user_step = literalai_client.message(
+            content=user_query, type="user_message", name="User"
+        )
+
+        time.sleep(1)  # to make sure the user step has arrived at Literal AI
+
+        literalai_client.api.create_attachment(
+            thread_id=thread.id,
+            step_id=user_step.id,
+            name="meal_image",
+            content=encode_image(user_image),
+        )
+
+        answer = generate_answer(user_query=user_query, image_url=user_image)
+        literalai_client.message(
+            content=answer, type="assistant_message", name="My Assistant"
+        )
+
+
+main()
+# Network requests by the SDK are performed asynchronously.
+# Invoke flush_and_stop() to guarantee the completion of all requests prior to the process termination.
+# WARNING: If you run a continuous server, you should not use this method.
+literalai_client.flush_and_stop()

examples/streaming.py

@@ -12,7 +12,7 @@
 
 
 sdk = LiteralClient(batch_size=2)
-sdk.instrument_openai()
+sdk.initialize()
 
 
 @sdk.thread

literalai/client.py

@@ -1,5 +1,10 @@
+import json
 import os
+from traceloop.sdk import Traceloop
 from typing import Any, Dict, List, Optional, Union
+from typing_extensions import deprecated
+import io
+from contextlib import redirect_stdout
 
 from literalai.api import AsyncLiteralAPI, LiteralAPI
 from literalai.callback.langchain_callback import get_langchain_callback
@@ -10,6 +15,7 @@
     experiment_item_run_decorator,
 )
 from literalai.event_processor import EventProcessor
+from literalai.exporter import LoggingSpanExporter
 from literalai.instrumentation.mistralai import instrument_mistralai
 from literalai.instrumentation.openai import instrument_openai
 from literalai.my_types import Environment
@@ -23,6 +29,7 @@
     step_decorator,
 )
 from literalai.observability.thread import ThreadContextManager, thread_decorator
+from literalai.prompt_engineering.prompt import Prompt
 from literalai.requirements import check_all_requirements
 
 
@@ -92,18 +99,21 @@ def to_sync(self) -> "LiteralClient":
         else:
             return self  # type: ignore
 
+    @deprecated("Use Literal.initialize instead")
     def instrument_openai(self):
         """
         Instruments the OpenAI SDK so that all LLM calls are logged to Literal AI.
         """
         instrument_openai(self.to_sync())
 
+    @deprecated("Use Literal.initialize instead")
     def instrument_mistralai(self):
         """
         Instruments the Mistral AI SDK so that all LLM calls are logged to Literal AI.
         """
         instrument_mistralai(self.to_sync())
 
+    @deprecated("Use Literal.initialize instead")
     def instrument_llamaindex(self):
         """
         Instruments the Llama Index framework so that all RAG & LLM calls are logged to Literal AI.
@@ -119,6 +129,13 @@ def instrument_llamaindex(self):
 
         instrument_llamaindex(self.to_sync())
 
+    def initialize(self):
+        with redirect_stdout(io.StringIO()):
+            Traceloop.init(
+                disable_batch=True,
+                exporter=LoggingSpanExporter(event_processor=self.event_processor),
+            )
+
     def langchain_callback(
         self,
         to_ignore: Optional[List[str]] = None,
@@ -352,6 +369,29 @@ def get_current_root_run(self):
         """
         return active_root_run_var.get()
 
+    def set_properties(
+        self,
+        name: Optional[str] = None,
+        tags: Optional[List[str]] = None,
+        metadata: Optional[Dict[str, Any]] = None,
+        prompt: Optional[Prompt] = None,
+    ):
+        thread = active_thread_var.get()
+        root_run = active_root_run_var.get()
+        parent = active_steps_var.get()[-1] if active_steps_var.get() else None
+
+        Traceloop.set_association_properties(
+            {
+                "literal.thread_id": str(thread.id) if thread else "None",
+                "literal.parent_id": str(parent.id) if parent else "None",
+                "literal.root_run_id": str(root_run.id) if root_run else "None",
+                "literal.name": str(name) if name else "None",
+                "literal.tags": json.dumps(tags) if tags else "None",
+                "literal.metadata": json.dumps(metadata) if metadata else "None",
+                "literal.prompt": json.dumps(prompt.to_dict()) if prompt else "None",
+            }
+        )
+
     def reset_context(self):
         """
         Resets the context, forgetting active steps & setting current thread to None.