OpenTelemetry for AI — Animate | AI Wisdom

← Back to topic
🎞 ANIMATE · Mode 3 of 5 · +100 XP— OpenTelemetry for AI
📋SDK Setup
⚙️Span
🔄Attributes
✅Export
🎯 Next: Test →
Step 1 / 4
Speed
Configure OTel SDK
TracerProvider + OTLPSpanExporter sends traces to collector on :4317💡 Analogy
Like setting up a logging pipeline — configure where logs go before you log anything.
Set up OpenTelemetry with an OTLP exporter that sends traces to a collector (Jaeger, Grafana Tempo, etc.). BatchSpanProcessor buffers and sends spans efficiently.
Console Output
> TracerProvider created
> BatchSpanProcessor: buffer=512, delay=5000ms
> OTLP exporter: http://collector:4317
> Tracer "my-llm-app" ready
●otel_ai.py
Configure OTel SDKL8–14
1from opentelemetry import trace
2from opentelemetry.sdk.trace import TracerProvider
3from opentelemetry.sdk.trace.export import BatchSpanProcessor
4from opentelemetry.exporter.otlp.proto.grpc.trace_exporter import OTLPSpanExporter
5from opentelemetry.semconv.ai import SpanAttributes
6from openai import OpenAI
7 
8# Configure OTel SDK
9provider = TracerProvider()
10exporter = OTLPSpanExporter(endpoint="http://collector:4317")
11provider.add_span_processor(BatchSpanProcessor(exporter))
12trace.set_tracer_provider(provider)
13 
14tracer = trace.get_tracer("my-llm-app", "1.0.0")
15 
16def traced_llm_call(prompt: str, model: str = "gpt-4o-mini") -> str:
17    client = OpenAI()
18    with tracer.start_as_current_span("llm.chat") as span:
19        # Set semantic convention attributes for LLM spans
20        span.set_attribute(SpanAttributes.LLM_REQUEST_MODEL, model)
21        span.set_attribute(SpanAttributes.LLM_REQUEST_MAX_TOKENS, 512)
22        span.set_attribute(SpanAttributes.LLM_PROMPTS, [prompt])
23 
24        response = client.chat.completions.create(
25            model=model,
26            messages=[{"role": "user", "content": prompt}],
27        )
28 
29        usage = response.usage
30        span.set_attribute(SpanAttributes.LLM_USAGE_PROMPT_TOKENS, usage.prompt_tokens)
31        span.set_attribute(SpanAttributes.LLM_USAGE_COMPLETION_TOKENS, usage.completion_tokens)
32        span.set_attribute(SpanAttributes.LLM_COMPLETIONS, [response.choices[0].message.content])
33 
34        return response.choices[0].message.content
35 
36result = traced_llm_call("Explain observability in one sentence.")
36 lines
Timeline — 4 Steps
Configure OTel SDK
TracerProvider + OTLPSpanExporter sends traces to collector on :4317💡 Analogy
Like setting up a logging pipeline — configure where logs go before you log anything.
Set up OpenTelemetry with an OTLP exporter that sends traces to a collector (Jaeger, Grafana Tempo, etc.). BatchSpanProcessor buffers and sends spans efficiently.
Console Output
> TracerProvider created
> BatchSpanProcessor: buffer=512, delay=5000ms
> OTLP exporter: http://collector:4317
> Tracer "my-llm-app" ready

Was this page helpful?

Discussion

TracerProvider + OTLPSpanExporter sends traces to collector on :4317

💡 Analogy

Like setting up a logging pipeline — configure where logs go before you log anything.

Set up OpenTelemetry with an OTLP exporter that sends traces to a collector (Jaeger, Grafana Tempo, etc.). BatchSpanProcessor buffers and sends spans efficiently.

Console Output

> TracerProvider created
> BatchSpanProcessor: buffer=512, delay=5000ms
> OTLP exporter: http://collector:4317
> Tracer "my-llm-app" ready

TracerProvider + OTLPSpanExporter sends traces to collector on :4317

💡 Analogy

Like setting up a logging pipeline — configure where logs go before you log anything.

Set up OpenTelemetry with an OTLP exporter that sends traces to a collector (Jaeger, Grafana Tempo, etc.). BatchSpanProcessor buffers and sends spans efficiently.

Console Output

> TracerProvider created
> BatchSpanProcessor: buffer=512, delay=5000ms
> OTLP exporter: http://collector:4317
> Tracer "my-llm-app" ready