Tracing Qwen (DashScope)

MLflow Tracing provides automatic tracing capability for Qwen (DashScope) models through the OpenAI SDK integration. Since Qwen (DashScope) offers an OpenAI-compatible API format, you can use mlflow.openai.autolog() to trace interactions with Qwen models.

MLflow trace automatically captures the following information about Qwen calls:

• Prompts and completion responses
• Latencies
• Token usage
• Model name
• Additional metadata such as temperature, max_completion_tokens, if specified.
• Function calling if returned in the response
• Built-in tools such as web search, file search, computer use, etc.
• Any exception if raised

Getting Started

1

Install dependencies

Python

bash

pip install mlflow openai

JS / TS

bash

npm install mlflow-openai openai

2

Start MLflow server

Local (pip)

If you have a local Python environment >= 3.10, you can start the MLflow server locally using the mlflow CLI command.

bash

mlflow server

Local (docker)

MLflow also provides a Docker Compose file to start a local MLflow server with a postgres database and a minio server.

bash

git clone --depth 1 --filter=blob:none --sparse https://github.com/mlflow/mlflow.git
cd mlflow
git sparse-checkout set docker-compose
cd docker-compose
cp .env.dev.example .env
docker compose up -d

Refer to the instruction for more details, e.g., overriding the default environment variables.

3

Enable tracing and call Qwen

Python

python

import openai
import mlflow

# Enable auto-tracing for OpenAI (works with Qwen)
mlflow.openai.autolog()

# Optional: Set a tracking URI and an experiment
mlflow.set_tracking_uri("http://localhost:5000")
mlflow.set_experiment("Qwen")

# Initialize the OpenAI client with Qwen API endpoint
client = openai.OpenAI(
    base_url="https://dashscope-intl.aliyuncs.com/compatible-mode/v1",
    api_key="<DASHSCOPE_API_KEY>",
)

response = client.chat.completions.create(
    model="qwen-plus",
    messages=[
        {"role": "system", "content": "You are a helpful assistant."},
        {"role": "user", "content": "What is the capital of France?"},
    ],
)

JS / TS

typescript

import { OpenAI } from "openai";
import { tracedOpenAI } from "mlflow-openai";

// Wrap the OpenAI client and point to Qwen endpoint
const client = tracedOpenAI(
  new OpenAI({
    baseURL: "https://dashscope-intl.aliyuncs.com/compatible-mode/v1",
    apiKey: "<DASHSCOPE_API_KEY>",
  })
);

const response = await client.chat.completions.create({
  model: "qwen-plus",
  messages: [
    { role: "system", content: "You are a helpful assistant." },
    { role: "user", content: "What is the capital of France?" },
  ],
  temperature: 0.1,
  max_tokens: 100,
});

4

View traces in MLflow UI

Browse to your MLflow UI (for example, http://localhost:5000) and open the Qwen experiment to see traces for the calls above.

-> View Next Steps for learning about more MLflow features like user feedback tracking, prompt management, and evaluation.

Streaming and Async Support

MLflow supports tracing for streaming and async Qwen APIs. Visit the OpenAI Tracing documentation for example code snippets for tracing streaming and async calls through OpenAI SDK.

Combine with frameworks or manual tracing

The automatic tracing capability in MLflow is designed to work seamlessly with the Manual Tracing SDK or multi-framework integrations. The examples below show Python (manual span) and JS/TS (manual span) at the same level of complexity.

Python

python

import json
from openai import OpenAI
import mlflow
from mlflow.entities import SpanType

# Initialize the OpenAI client with Qwen API endpoint
client = OpenAI(
    base_url="https://dashscope-intl.aliyuncs.com/compatible-mode/v1",
    api_key="<DASHSCOPE_API_KEY>",
)


# Create a parent span for the Qwen call
@mlflow.trace(span_type=SpanType.CHAIN)
def answer_question(question: str):
    messages = [{"role": "user", "content": question}]
    response = client.chat.completions.create(
        model="qwen-plus",
        messages=messages,
    )

    # Attach session/user metadata to the trace
    mlflow.update_current_trace(
        metadata={
            "mlflow.trace.session": "session-12345",
            "mlflow.trace.user": "user-a",
        }
    )
    return response.choices[0].message.content


answer = answer_question("What is the capital of France?")

JS / TS

typescript

import * as mlflow from "mlflow-tracing";
import { OpenAI } from "openai";
import { tracedOpenAI } from "mlflow-openai";

mlflow.init({
  trackingUri: "http://localhost:5000",
  experimentId: "<your_experiment_id>",
});

// Wrap the OpenAI client and point to Qwen endpoint
const client = tracedOpenAI(
  new OpenAI({
    baseURL: "https://dashscope-intl.aliyuncs.com/compatible-mode/v1",
    apiKey: "<DASHSCOPE_API_KEY>",
  })
);

// Create a traced function that wraps the Qwen call
const answerQuestion = mlflow.trace(
  async (question: string) => {
    const resp = await client.chat.completions.create({
      model: "qwen-plus",
      messages: [{ role: "user", content: question }],
    });
    return resp.choices[0].message?.content;
  },
  { name: "answer-question" }
);

await answerQuestion("What is the capital of France?");

Running either example will produce a trace that includes the Qwen LLM span; the traced function creates the parent span automatically.

Next steps

Track User Feedback

Record user feedback on traces for tracking user satisfaction.

Learn about feedback ->

Manage Prompts

Learn how to manage prompts with MLflow's prompt registry.

Manage prompts ->

Evaluate Traces

Evaluate traces with LLM judges to understand and improve your AI application's behavior.

Evaluate traces ->