# Best practice for external LLM script to prodigy spancat import/launch

**URL:** https://support.prodi.gy/t/best-practice-for-external-llm-script-to-prodigy-spancat-import-launch/6904
**Category:** Uncategorized
**Tags:** usage, best-practices, spancat, ner
**Created:** [November 18, 2023, 12:42pm UTC](https://support.prodi.gy/t/best-practice-for-external-llm-script-to-prodigy-spancat-import-launch/6904 "2023-11-18T12:42:33Z")
**Posts on this page:** 1
**Page:** 1

<div class="post-metadata">

### Author: ![grahama](https://sea2.discourse-cdn.com/flex020/user_avatar/support.prodi.gy/grahama/32/4232_2.png) [@grahama](https://support.prodi.gy/u/grahama)
#### Post date: [November 18, 2023, 12:42pm UTC](https://support.prodi.gy/t/best-practice-for-external-llm-script-to-prodigy-spancat-import-launch/6904/1 "2023-11-18T12:42:33Z")

</div>

Hello. I’m very new to prodigy. I would like to import annotation data from my externally run lllm script directly into prodigy for correcting/re-labeling spans. So in my limited understanding, the pipeline would be LLM, to prodigy span-correct launch-import , to model training.

- is the below format what prodigy needs for to import it directly?
- other than using the span correct recipe, is that everything I need to for a best practice and Optimized workflow for prodigy
- I’m assuming I could launch the recipe script directly from my LLM script  
Any push in the right direction is appreciated. The below is only using one label for simplicity—i have a whole bunch of labels….and sentences

```python
[
  {
    "text": "When less than 6 milliseconds, the Kit Kat clock will reset.",
    "spans": [
      {
        "start": 5,
        "end": 28,
        "label": "FREQ",
        "text": "less than 6 milliseconds"
      },
      {
        "start": 30,
        "end": 45,
        "label": "FREQ",
        "text": "the Kit Kat clock"
      },
      {
        "start": 50,
        "end": 55,
        "label": "FREQ",
        "text": "reset"
      }
    ]
  }
]

```