Deepresearch

Set up the Tongyi DeepResearch inference environment and evaluation data

Prepare a Python environment, install dependencies, configure required services, and create supported evaluation input data.

Before you start

• Python 3.10.0 is recommended.
• An isolated Conda or virtualenv environment is strongly advised.
• Access to the required service keys and endpoints listed in the configuration.
• A model-weights path and an evaluation dataset name.

Steps

1. Create and activate an isolated Python 3.10.0 environment, for example with Conda.

2. From the repository, install dependencies with the command below.

   Documented commandCopy command

   pip install -r requirements.txt

3. Copy .env.example to .env, then provide the documented search, page-reading, summarization, file-parsing, sandbox, model, dataset, and output settings.

4. Prepare evaluation data as JSONL records containing question and answer, or as a JSON array of objects with those keys.

Guide limitations

• The documentation recommends Python 3.10.0 because other versions may cause dependency issues.
• The answer field is a ground-truth or reference answer used for evaluation.