feat(smolagents): updates to latest and makes examples use openteleme…

…try-instrument (#1277)

Signed-off-by: Adrian Cole <[email protected]>

python/.gitignore

@@ -1,2 +1,5 @@
 # vendored virtual environments
 .venv
+
+# gradio work directory
+.gradio

python/instrumentation/openinference-instrumentation-smolagents/README.md

@@ -14,42 +14,58 @@ pip install openinference-instrumentation-smolagents
 
 ## Quickstart
 
-This quickstart shows you how to instrument your guardrailed LLM application 
+This quickstart shows you how to instrument your LLM agent application.
 
-Install required packages.
+You've already installed openinference-instrumentation-smolagents. Next is to install packages for smolagents,
+Phoenix and `opentelemetry-instrument`, which exports traces to it.
 
 ```shell
-pip install smolagents arize-phoenix opentelemetry-sdk opentelemetry-exporter-otlp
+pip install smolagents arize-phoenix opentelemetry-sdk opentelemetry-exporter-otlp-proto-grpc opentelemetry-distro
 ```
 
-Start Phoenix in the background as a collector. By default, it listens on `http://localhost:6006`. You can visit the app via a browser at the same address. (Phoenix does not send data over the internet. It only operates locally on your machine.)
+Start Phoenix in the background as a collector, which listens on `http://localhost:6006` and default gRPC port 4317.
+Note that Phoenix does not send data over the internet. It only operates locally on your machine.
 
 ```shell
 python -m phoenix.server.main serve
 ```
 
-Set up `SmolagentsInstrumentor` to trace your crew and send the traces to Phoenix at the endpoint defined below.
+Create an example like this:
+
+```python
+from smolagents import CodeAgent, DuckDuckGoSearchTool, HfApiModel
+
+agent = CodeAgent(tools=[DuckDuckGoSearchTool()], model=HfApiModel())
+
+agent.run("How many seconds would it take for a leopard at full speed to run through Pont des Arts?")
+```
+
+Then, run it like this:
+
+```shell
+opentelemetry-instrument python example.py
+```
+
+Finally, browse for your trace in Phoenix at `http://localhost:6006`!
+
+## Manual instrumentation
+
+`opentelemetry-instrument` is the [Zero-code instrumentation](https://opentelemetry.io/docs/zero-code/python) approach
+for Python. It avoids explicitly importing and configuring OpenTelemetry code in your main source. Alternatively, you
+can copy-paste the following into your main source and run it without `opentelemetry-instrument`.
 
 ```python
-from opentelemetry import trace
 from opentelemetry.sdk.trace import TracerProvider
-from opentelemetry.sdk.trace.export import BatchSpanProcessor
 
 from openinference.instrumentation.smolagents import SmolagentsInstrumentor
-from opentelemetry.exporter.otlp.proto.http.trace_exporter import OTLPSpanExporter
-from opentelemetry.sdk.trace.export import ConsoleSpanExporter, SimpleSpanProcessor
+from opentelemetry.exporter.otlp.proto.grpc.trace_exporter import OTLPSpanExporter
+from opentelemetry.sdk.trace.export import SimpleSpanProcessor
 
-endpoint = "http://0.0.0.0:6006/v1/traces"
+otlp_exporter = OTLPSpanExporter(endpoint="http://localhost:4317", insecure=True)
 trace_provider = TracerProvider()
-trace_provider.add_span_processor(SimpleSpanProcessor(OTLPSpanExporter(endpoint)))
+trace_provider.add_span_processor(SimpleSpanProcessor(otlp_exporter))
 
 SmolagentsInstrumentor().instrument(tracer_provider=trace_provider)
-
-from smolagents import CodeAgent, DuckDuckGoSearchTool, HfApiModel
-
-agent = CodeAgent(tools=[DuckDuckGoSearchTool()], model=HfApiModel())
-
-agent.run("How many seconds would it take for a leopard at full speed to run through Pont des Arts?")
 ```
 
 ## More Info

python/instrumentation/openinference-instrumentation-smolagents/examples/README.md

@@ -0,0 +1,49 @@
+# OpenInference smolagents Examples
+
+This directory contains numerous examples that show how to use OpenInference to instrument smolagents applications.
+Specifically, this uses the [openinference-instrumentation-smolagents](..) package from source in the parent directory.
+
+## Installation
+
+```shell
+pip install -r requirements.txt
+```
+
+Start Phoenix in the background as a collector, which listens on `http://localhost:6006` and default gRPC port 4317.
+Note that Phoenix does not send data over the internet. It only operates locally on your machine.
+
+```shell
+python -m phoenix.server.main serve
+```
+
+## Running
+
+Copy [env.example](env.example) to `.env` and update variables your example uses, such as `OPENAI_API_KEY`.
+
+Then, run an example like this:
+
+```shell
+dotenv run -- opentelemetry-instrument python managed_agent.py
+```
+
+Finally, browse for your trace in Phoenix at `http://localhost:6006`!
+
+## Manual instrumentation
+
+`opentelemetry-instrument` is the [Zero-code instrumentation](https://opentelemetry.io/docs/zero-code/python) approach
+for Python. It avoids explicitly importing and configuring OpenTelemetry code in your main source. Alternatively, you
+can copy-paste the following into your main source and run it without `opentelemetry-instrument`.
+
+```python
+from opentelemetry.sdk.trace import TracerProvider
+
+from openinference.instrumentation.smolagents import SmolagentsInstrumentor
+from opentelemetry.exporter.otlp.proto.grpc.trace_exporter import OTLPSpanExporter
+from opentelemetry.sdk.trace.export import SimpleSpanProcessor
+
+otlp_exporter = OTLPSpanExporter(endpoint="http://localhost:4317", insecure=True)
+trace_provider = TracerProvider()
+trace_provider.add_span_processor(SimpleSpanProcessor(otlp_exporter))
+
+SmolagentsInstrumentor().instrument(tracer_provider=trace_provider)
+```

python/instrumentation/openinference-instrumentation-smolagents/examples/e2b_example.py

@@ -1,7 +1,3 @@
-from io import BytesIO
-
-import requests
-from PIL import Image
 from smolagents import CodeAgent, GradioUI, HfApiModel, Tool
 from smolagents.default_tools import VisitWebpageTool
 
@@ -17,6 +13,11 @@ def __init__(self):
         self.url = "https://em-content.zobj.net/source/twitter/53/robot-face_1f916.png"
 
     def forward(self):
+        from io import BytesIO
+
+        import requests
+        from PIL import Image
+
         response = requests.get(self.url)
 
         return Image.open(BytesIO(response.content))

python/instrumentation/openinference-instrumentation-smolagents/examples/env.example

@@ -0,0 +1,12 @@
+# API Key from https://platform.openai.com/api-keys
+OPENAI_API_KEY=sk-YOUR_API_KEY
+# API Key from https://e2b.dev/docs/legacy/getting-started/api-key
+E2B_API_KEY=e2b_YOUR_API_KEY
+
+# Phoenix listens on the default gRPC port 4317, so you don't need to change
+# exporter settings. If you prefer to export via HTTP, uncomment this:
+# OTEL_EXPORTER_OTLP_ENDPOINT=http://0.0.0.0:6006
+# OTEL_EXPORTER_OTLP_PROTOCOL=http/protobuf
+
+# Export traces every second instead of every 5 seconds
+OTEL_BSP_SCHEDULE_DELAY=1000

python/instrumentation/openinference-instrumentation-smolagents/examples/managed_agent.py

@@ -1,38 +1,16 @@
-import os
-
-from opentelemetry.exporter.otlp.proto.http.trace_exporter import OTLPSpanExporter
-from opentelemetry.sdk.trace import TracerProvider
-from opentelemetry.sdk.trace.export import SimpleSpanProcessor
 from smolagents import (
     CodeAgent,
     DuckDuckGoSearchTool,
-    ManagedAgent,
     OpenAIServerModel,
     ToolCallingAgent,
 )
 
-from openinference.instrumentation.smolagents import SmolagentsInstrumentor
-
-endpoint = "http://0.0.0.0:6006/v1/traces"
-trace_provider = TracerProvider()
-trace_provider.add_span_processor(SimpleSpanProcessor(OTLPSpanExporter(endpoint)))
-
-SmolagentsInstrumentor().instrument(tracer_provider=trace_provider, skip_dep_check=True)
-
-
-model = OpenAIServerModel(
-    model_id="gpt-4o",
-    api_base="https://api.openai.com/v1",
-    api_key=os.environ["OPENAI_API_KEY"],
-)
+model = OpenAIServerModel(model_id="gpt-4o")
 agent = ToolCallingAgent(
     tools=[DuckDuckGoSearchTool()],
     model=model,
     max_steps=3,
-)
-managed_agent = ManagedAgent(
-    agent=agent,
-    name="managed_agent",
+    name="search",
     description=(
         "This is an agent that can do web search. "
         "When solving a task, ask him directly first, he gives good answers. "
@@ -42,7 +20,7 @@
 manager_agent = CodeAgent(
     tools=[DuckDuckGoSearchTool()],
     model=model,
-    managed_agents=[managed_agent],
+    managed_agents=[agent],
 )
 manager_agent.run(
     "How many seconds would it take for a leopard at full speed to run through Pont des Arts? "

python/instrumentation/openinference-instrumentation-smolagents/examples/openai_model.py

@@ -1,22 +1,5 @@
-import os
-
-from opentelemetry.exporter.otlp.proto.http.trace_exporter import OTLPSpanExporter
-from opentelemetry.sdk.trace import TracerProvider
-from opentelemetry.sdk.trace.export import (
-    SimpleSpanProcessor,
-)
 from smolagents import OpenAIServerModel
 
-from openinference.instrumentation.smolagents import SmolagentsInstrumentor
-
-endpoint = "http://0.0.0.0:6006/v1/traces"
-trace_provider = TracerProvider()
-trace_provider.add_span_processor(SimpleSpanProcessor(OTLPSpanExporter(endpoint)))
-
-SmolagentsInstrumentor().instrument(tracer_provider=trace_provider, skip_dep_check=True)
-
-model = OpenAIServerModel(
-    model_id="gpt-4o", api_key=os.environ["OPENAI_API_KEY"], api_base="https://api.openai.com/v1"
-)
+model = OpenAIServerModel(model_id="gpt-4o")
 output = model(messages=[{"role": "user", "content": "hello world"}])
-print(output)
+print(output.content)

python/instrumentation/openinference-instrumentation-smolagents/examples/openai_model_tool_call.py

@@ -1,21 +1,6 @@
-import os
-
-from opentelemetry.exporter.otlp.proto.http.trace_exporter import OTLPSpanExporter
-from opentelemetry.sdk.trace import TracerProvider
-from opentelemetry.sdk.trace.export import (
-    SimpleSpanProcessor,
-)
 from smolagents import OpenAIServerModel
 from smolagents.tools import Tool
 
-from openinference.instrumentation.smolagents import SmolagentsInstrumentor
-
-endpoint = "http://0.0.0.0:6006/v1/traces"
-trace_provider = TracerProvider()
-trace_provider.add_span_processor(SimpleSpanProcessor(OTLPSpanExporter(endpoint)))
-
-SmolagentsInstrumentor().instrument(tracer_provider=trace_provider, skip_dep_check=True)
-
 
 class GetWeatherTool(Tool):
     name = "get_weather"
@@ -27,10 +12,8 @@ def forward(self, location: str) -> str:
         return "sunny"
 
 
-model = OpenAIServerModel(
-    model_id="gpt-4o", api_key=os.environ["OPENAI_API_KEY"], api_base="https://api.openai.com/v1"
-)
-output_message = model(
+model = OpenAIServerModel(model_id="gpt-4o")
+output = model(
     messages=[
         {
             "role": "user",
@@ -39,4 +22,4 @@ def forward(self, location: str) -> str:
     ],
     tools_to_call_from=[GetWeatherTool()],
 )
-print(output_message)
+print(output.tool_calls[0].function)

python/instrumentation/openinference-instrumentation-smolagents/examples/rag.py

@@ -1,24 +1,9 @@
-import os
-
 import datasets
 from langchain.docstore.document import Document
 from langchain.text_splitter import RecursiveCharacterTextSplitter
 from langchain_community.retrievers import BM25Retriever
-from opentelemetry.exporter.otlp.proto.http.trace_exporter import OTLPSpanExporter
-from opentelemetry.sdk.trace import TracerProvider
-from opentelemetry.sdk.trace.export import (
-    SimpleSpanProcessor,
-)
 from smolagents import CodeAgent, OpenAIServerModel, Tool
 
-from openinference.instrumentation.smolagents import SmolagentsInstrumentor
-
-endpoint = "http://0.0.0.0:6006/v1/traces"
-trace_provider = TracerProvider()
-trace_provider.add_span_processor(SimpleSpanProcessor(OTLPSpanExporter(endpoint)))
-
-SmolagentsInstrumentor().instrument(tracer_provider=trace_provider)
-
 knowledge_base = datasets.load_dataset("m-ric/huggingface_doc", split="train")
 knowledge_base = knowledge_base.filter(
     lambda row: row["source"].startswith("huggingface/transformers")
@@ -78,13 +63,9 @@ def forward(self, query: str) -> str:
 retriever_tool = RetrieverTool(docs_processed)
 agent = CodeAgent(
     tools=[retriever_tool],
-    model=OpenAIServerModel(
-        "gpt-4o",
-        api_base="https://api.openai.com/v1",
-        api_key=os.environ["OPENAI_API_KEY"],
-    ),
+    model=OpenAIServerModel("gpt-4o"),
     max_steps=4,
-    verbose=True,
+    verbosity_level=2,
 )
 
 agent_output = agent.run(

python/instrumentation/openinference-instrumentation-smolagents/examples/requirements.txt

@@ -1,9 +1,19 @@
+# Main dependencies of the examples in this directory:
+smolagents[e2b,gradio,litellm,openai]
 datasets
 langchain
-langchain-community
-opentelemetry-exporter-otlp
-opentelemetry-exporter-otlp-proto-http
-opentelemetry-sdk
+langchain_community
 rank_bm25
-requests
-sqlalchemy
+
+# Install `opentelemetry-instrument` for zero code instrumentation. This
+# defaults to export traces to localhost 4317, which Phoenix listens on.
+opentelemetry-sdk
+opentelemetry-exporter-otlp-proto-grpc
+opentelemetry-distro
+# Source of openinference-instrumentation-smolagents
+-e ../
+
+
+# Both `opentelemetry-instrument` and main need .env variables. Run like this:
+#   dotenv run -- opentelemetry-instrument python e2b_example.py
+python-dotenv[cli]

python/instrumentation/openinference-instrumentation-smolagents/examples/text2sql.py

@@ -1,10 +1,3 @@
-import os
-
-from opentelemetry.exporter.otlp.proto.http.trace_exporter import OTLPSpanExporter
-from opentelemetry.sdk.trace import TracerProvider
-from opentelemetry.sdk.trace.export import (
-    SimpleSpanProcessor,
-)
 from smolagents import (
     CodeAgent,
     OpenAIServerModel,
@@ -23,14 +16,6 @@
     text,
 )
 
-from openinference.instrumentation.smolagents import SmolagentsInstrumentor
-
-endpoint = "http://0.0.0.0:6006/v1/traces"
-trace_provider = TracerProvider()
-trace_provider.add_span_processor(SimpleSpanProcessor(OTLPSpanExporter(endpoint)))
-
-SmolagentsInstrumentor().instrument(tracer_provider=trace_provider)
-
 engine = create_engine("sqlite:///:memory:")
 metadata_obj = MetaData()
 
@@ -90,10 +75,6 @@ def sql_engine(query: str) -> str:
 
 agent = CodeAgent(
     tools=[sql_engine],
-    model=OpenAIServerModel(
-        "gpt-4o-mini",
-        api_base="https://api.openai.com/v1",
-        api_key=os.environ["OPENAI_API_KEY"],
-    ),
+    model=OpenAIServerModel("gpt-4o-mini"),
 )
 agent.run("Can you give me the name of the client who got the most expensive receipt?")