Humor Generation Model – SemEval 2026
Model Description
This model was developed for participation in SemEval 2026 – Humor Generation Task.
It is fine-tuned to generate creative, logically coherent, and instruction-compliant jokes following a structured output format.
The system emphasizes:
- High logical coherence
- Creative and clever humor
- Strict instruction adherence
- Consistent formatting compliance
All outputs strictly follow the required format:
JOKE: <generated joke text>
Intended Use
Primary Use
This model is intended for:
- Humor generation research
- Controlled joke generation
- Instruction-following text generation experiments
- Academic evaluation and benchmarking
Out-of-Scope Use
- Harmful or offensive content generation
- Misinformation or deceptive content
- Automated large-scale content spam
Users are responsible for ensuring ethical and appropriate usage.
Training Details
Training Setup
- Epochs: 4
- Final Training Loss: ~1.65
- Dataset Format: TSV
- Output Prefix Enforcement:
JOKE: - Instruction Compliance: 100%
The model was fine-tuned to balance creativity and coherence while maintaining formatting reliability.
Dataset
The training dataset consists of cleaned and structured humorous text samples.
Preprocessing steps included:
- Removal of malformed entries
- Structural normalization
- Prefix enforcement
- Formatting consistency validation
The dataset was prepared specifically for the SemEval 2026 Humor Generation task.
Evaluation
Performance Summary
| Metric | Result |
|---|---|
| Training Loss | ~1.65 |
| Joke Quality | Creative / Clever |
| Instruction Following | 100% (Perfect) |
| Logic / Coherence | High |
| Formatting Compliance | Correct JOKE: Prefix |
Evaluation focused on:
- Creativity
- Logical consistency
- Instruction adherence
- Structural compliance
Example Usage
from transformers import AutoTokenizer, AutoModelForCausalLM
model_name = "your-username/your-model-name"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(model_name)
prompt = "Generate a clever joke about programming."
inputs = tokenizer(prompt, return_tensors="pt")
outputs = model.generate(**inputs, max_length=100)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))
Ethical Considerations
Humor generation may occasionally produce unintended or sensitive content depending on input prompts.
Users should:
- Apply moderation filters if deploying publicly
- Monitor outputs in real-world applications
- Ensure compliance with ethical AI guidelines
Limitations
- Performance depends on prompt clarity
- May struggle with highly niche or domain-specific humor
- Creativity is bounded by training data diversity
Citation
If you use this model in academic work, please cite:
@misc{abbas2026humor,
title={Humor Generation Model for SemEval 2026},
author={Insa Abbas},
year={2026},
note={SemEval 2026 Submission},
url={https://huggingface.co/your-username/your-model-name}
}
Author
Insa Abbas
Email: insaabbas675@gmail.com
Acknowledgment
This model was developed as part of participation in SemEval 2026 – Humor Generation Task.
- Downloads last month
- 17