Shlok0011/prompt-task-complexity-dataset

🚀 Prompt Intent & Complexity Benchmark Dataset

A comprehensive, curated multi-task benchmark of 55,154 user prompts designed for LLM router engineering, prompt complexity estimation, and prompt intent classification across 23 distinct technical and conversational categories.

---

📌 Why This Dataset?

Modern LLM agent architectures and multi-model router systems (e.g., routing prompts between fast lightweight models like Claude 3.5 Haiku / GPT-4o-mini and frontier reasoning models like o1 / Claude 3.5 Sonnet) require two fundamental capabilities:

1. Intent Understanding: Knowing whether a prompt is requesting Code Generation, Mathematical Reasoning, Safety Moderation, or Text Summarization.
2. Complexity Assessment: Estimating the cognitive complexity / difficulty of the prompt on a continuous $[0.0, 1.0]$ scale to optimize latency, inference cost, and model allocation.

This dataset provides ground-truth annotations across both dimensions in a standardized format.

---

📊 Dataset Splits & Statistics

Split	Rows	Purpose
train	44,123	Stratified training split
test	11,031	Unseen hold-out evaluation benchmark
Total	55,154	Full benchmark corpus

Data Fields

Each row in train.parquet and test.parquet contains:

• id (int64): Unique sample identifier.
• head (string): Annotation target type (classification or regression).
• prompt (string): The full text of the user prompt or query.
• task_type (string, nullable): One of 23 intent classes (populated for classification samples).
• complexity (float64, nullable): Continuous prompt complexity rating normalized in $[0.0, 1.0]$ (populated for regression samples).

---

🏷️ The 23 Intent Classes

The classification target spans 23 fine-grained prompt domains:

ID	Class Label	ID	Class Label
0	Code Generation	12	Reasoning
1	Code Understanding	13	Safety & Content Moderation
2	Dialogue & Conversation	14	Semantic & Relation Analysis
3	Explanation Generation	15	Summarization
4	Information Extraction	16	Text Classification
5	Linguistic Analysis	17	Text Completion
6	Mathematical Reasoning	18	Text Generation
7	Miscellaneous	19	Text Matching
8	Program Execution	20	Text Transformation
9	Question Analysis	21	Translation
10	Question Answering	22	Verification & Fact Checking
11	Question Generation

---

⚡ Quickstart Usage

Using Hugging Face datasets

from datasets import load_dataset

# Load dataset directly from Hugging Face
dataset = load_dataset("Shlok0011/prompt-task-complexity-dataset")

print(dataset)
# Output:
# DatasetDict({
#     train: Dataset({features: ['id', 'head', 'prompt', 'complexity', 'task_type'], num_rows: 44123}),
#     test: Dataset({features: ['id', 'head', 'prompt', 'complexity', 'task_type'], num_rows: 11031})
# })

# Inspect a sample
print(dataset["train"][0])

Using Pandas / PyArrow

import pandas as pd

# Direct parquet loading
df_train = pd.read_parquet("https://huggingface.co/datasets/Shlok0011/prompt-task-complexity-dataset/resolve/main/train.parquet")
print(df_train.head())

---

🤖 Pre-trained Model

A dual-headed fine-tuned model trained on this exact benchmark is available: 👉 Shlok0011/deberta-v3-large-prompt-analyzer

• 91.41% Accuracy on the 23-class test split.
• 0.134 MAE on complexity prediction.
• Supports both PyTorch and ONNX Runtime.

---

📜 License & Attribution

This dataset is released under the Creative Commons Attribution 4.0 International (CC BY 4.0) license.

You are free to:

• Share and redistribute the material in any medium or format.
• Adapt, remix, transform, and build upon the material for any purpose, even commercially.

Under the following terms:

• Attribution: You must give appropriate credit to Shlok Patel, provide a link to this repository, and indicate if changes were made.

Citation

@dataset{patel2026promptbenchmark,
  author       = {Shlok Patel},
  title        = {Prompt Intent & Complexity Benchmark Dataset},
  year         = {2026},
  publisher    = {Hugging Face},
  howpublished = {\url{https://huggingface.co/datasets/Shlok0011/prompt-task-complexity-dataset}}
}