When you run an evaluation in AI GO!, the platform generates evidence, i.e. a task result log — a structured record of every inference call and computed score.

Download a task result log to inspect evidence locally, feed results into analysis pipelines, integrate with third-party tools, or transfer results to another AI app.
Upload a task result log to reproduce results in another AI app, reuse a cached run without rerunning inference, or save time and cost.

Task Result Log

A task result log is a JSON file produced after a task specification finishes running. It contains the raw inputs, model responses, and computed scores for every sample in the dataset. Because it captures the complete output of a task run, it can be re-uploaded to a different AI app or deployment to reproduce the same results without any additional inference.

Download Task Result Log

Downloading a task result log exports the inference outputs and scores from a completed evaluation run to disk, along with snapshots of all entities used in the evaluation.

Step 1: Setup (Optional)

Run the evaluation that you want to download the evidence for.

lf run -f run.yaml

Step 2: Export Results

lf export eval serializes an evaluation into a portable YAML file.

--output sets the destination directory.
--link-logs downloads the task result log for each task specification and saves it alongside the YAML — the exported file then includes task_result_log_path entries pointing to those files.

lf export eval --id <eval ID> --output results --link-logs

The resulting directory looks like this:

results/
├── foobarization-evaluation.yaml
└── task_results/
    └── foobarization-spec/
        └── task_result_log.json

The exported foobarization-evaluation.yaml contains a task_result_log_path for each task specification:

evaluation:
  key: foobarization-evaluation
  task_specifications:
    - key: foobarization-spec
      task_result_log_path: task_results/foobarization-spec/task_result_log.json

Upload Task Result Log

Uploading a task result log injects pre-computed inference outputs and scores into a new evaluation run.

Step 1: Setup (Optional)

A pre-condition to uploading pre-computed task result logs is that all entities used in the evaluation that produced the log are integrated into the new AI app. You can use the lf add to integrate the entities, but not run any evaluation.

lf add -f run.yaml

Step 2: Run with Uploaded Log

Run the evaluation using the exported results. Because the exported evaluation YAML contains task_result_log_path entries, AI GO! uploads the logs instead of executing inference. The new evaluation will show the same results as the source evaluation.

lf run -f results/foobarization-evaluation.yaml

📘
Find a full code example for this guide in AI GO! integrations.