---
library_name: transformers
license: apache-2.0
base_model: asi/gpt-fr-cased-small
tags:
- generated_from_trainer
model-index:
- name: gpt2-quote-fr-v1.1
  results: []
pipeline_tag: text-generation
---

# GPT-2 French Quote Generator 🇫🇷

A fine-tuned GPT-2 model for generating philosophical 
and inspirational quotes in French.

## Model Description

This model is based on `asi/gpt-fr-cased-small` and has been 
fine-tuned on a dataset of 975 carefully collected French quotes 
covering various philosophical themes.

## Themes Covered

- ❤️ Love (Amour)
- 🌱 Life (Vie)
- 🌌 Existence
- 😊 Happiness (Bonheur)
- 🦉 Wisdom (Sagesse)
- ⏳ Time (Temps)
- 🕊️ Freedom (Liberté)
- 🙏 Spirituality (Spiritualité)

## Usage
```python
from transformers import pipeline

generator = pipeline(
    "text-generation",
    model="Mayko2995/gpt2-quote-fr-v1.1"
)

result = generator(
    "Le bonheur est",
    max_new_tokens=20,
    temperature=0.8,
    do_sample=True,
    repetition_penalty=2.0,
    no_repeat_ngram_size=3
)

print(result[0]["generated_text"])
```

## Generated Examples
```
"Le bonheur est une petite chose que l'on invente 
 avec la sagesse de chaque instant."

"Dans l'existence, il faut avoir une âme pour 
 comprendre que tout existe."

"La vie est un voyage vers son avenir qu'il nous 
 faudra parcourir avec sagesse et humilité."
```

## Dataset

- 975 French quotes
- Collected from Wikiquote and other sources
- Various authors including Victor Hugo, Albert Camus,
  Pablo Neruda, Einstein and others

## Limitations

- Relatively small dataset — 975 quotes
- May generate incomplete sentences
- Best results with max_new_tokens between 15 and 25
- Designed for French language only

## Intended Uses

✅ Generating inspirational French quotes
✅ Creative writing assistance
✅ Educational NLP projects
✅ French language generation research

## Out of Scope

❌ Factual information generation
❌ Translation tasks
❌ Other languages

## Author

**Jonasson Claubert Joachin** — Self-taught AI Engineer 🇭🇹
Based in Haiti — passionate about NLP and LLMs

*Building AI tools adapted to the Haitian context*

## Versions

| Version | Dataset | Val Loss |
|---------|---------|----------|
| v1.0 | 300 quotes | 1.12 |
| v1.1 | 975 quotes | 0.71 ✅ |

## Training procedure

### Training hyperparameters

The following hyperparameters were used during training:
- learning_rate: 2e-05
- train_batch_size: 2
- eval_batch_size: 8
- seed: 42
- gradient_accumulation_steps: 8
- total_train_batch_size: 16
- optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
- lr_scheduler_type: linear
- num_epochs: 10

### Training results

| Training Loss | Epoch | Step | Validation Loss |
|:-------------:|:-----:|:----:|:---------------:|
| 1.5437        | 1.0   | 49   | 1.2013          |
| 1.0311        | 2.0   | 98   | 0.9540          |
| 0.8106        | 3.0   | 147  | 0.8370          |
| 0.7146        | 4.0   | 196  | 0.7859          |
| 0.6868        | 5.0   | 245  | 0.7580          |
| 0.5707        | 6.0   | 294  | 0.7380          |
| 0.5532        | 7.0   | 343  | 0.7201          |
| 0.5705        | 8.0   | 392  | 0.7145          |
| 0.5311        | 9.0   | 441  | 0.7129          |
| 0.5057        | 10.0  | 490  | 0.7113          |


### Framework versions

- Transformers 5.0.0
- Pytorch 2.10.0+cu128
- Datasets 4.0.0
- Tokenizers 0.22.2