Geyanth08/node_red_nl_prompts

---
language:
- en
task_categories:
- text-generation
pretty_name: Node-RED NL2Flow Prompt Pairs v1
size_categories:
- 1K<n<10K
---

# Geyanth08/node_red_nl_prompts

This dataset contains curated NL-to-Node-RED prompt and flow pairs.

Primary fine-tuning splits are under data:
- data/train.jsonl
- data/validation.jsonl

Each record includes:
- flow_id
- prompt_id
- prompt_style
- prompt_text
- generation_source
- source_title
- source_summary_text
- source_url
- target_flow_json
- created_at_utc

## Files

- data/train.jsonl: Curated train split
- data/validation.jsonl: Curated validation split
- data/kept_all.jsonl: All curated kept prompt-flow pairs
- data/raw_prompt_pairs_v1.jsonl: Unfiltered generated prompt-flow pairs
- metadata/nl_prompt_pairs_v1_filter_stats.json: Filtering statistics
- metadata/nl_prompt_pairs_v1_filter_rejects.csv: Filtering rejection log
- metadata/nl_prompt_pairs_v1_errors.csv: Prompt generation errors
- metadata/flow_summaries.jsonl: Source flow summaries
- metadata/dataset_v0.jsonl: Source extracted flow dataset
- metadata/flow_urls.csv: Source URL list
- metadata/operation_timings.csv: Collection and extraction timings

## Snapshot

- Raw generated prompt pairs: 1586
- Curated kept prompt pairs: 1444
- Curated rejected prompt pairs: 142
- Kept rate: 91.05%
- Train pairs: 1305
- Validation pairs: 139
- Kept all pairs file rows: 1444
- Raw dataset rows: 668

## Load in Python

from datasets import load_dataset

ds_train = load_dataset("Geyanth08/node_red_nl_prompts", data_files="data/train.jsonl", split="train")
ds_val = load_dataset("Geyanth08/node_red_nl_prompts", data_files="data/validation.jsonl", split="train")
print(ds_train[0])