|
|
--- |
|
|
tags: |
|
|
- text-generation |
|
|
- transformer |
|
|
- chatbot |
|
|
license: mit |
|
|
library_name: transformers |
|
|
language: en |
|
|
datasets: |
|
|
- custom |
|
|
inference: true |
|
|
--- |
|
|
|
|
|
# Smily-ultra-1 |
|
|
|
|
|
Smily-ultra-1 is a custom fine-tuned language model designed for reasoning and chatbot-like interactions. |
|
|
|
|
|
# NOTE!!!: As this model has reasoning capacities, it is considerably slower than SAM-flash-mini-v1 |
|
|
However it is also more powerful and smarter than the SAM-flash-mini-v1 model at the expense of speed and size. |
|
|
|
|
|
|
|
|
|
|
|
# Smily-ultra-1 |
|
|
|
|
|
**Smily-ultra-1** is a fine-tuned language model optimized for chatbot-style conversations and basic logical reasoning. It was created by **Smilyai-labs** using a small dataset of synthetic examples and trained in Google Colab. The model is small and lightweight, making it suitable for experimentation, education, and simple chatbot tasks. |
|
|
|
|
|
|
|
|
## try it yourself! |
|
|
|
|
|
Try it with this space: [Try it here!] (https://huggingface.co/spaces/Smilyai-labs/smily-ultra-chatbot) |
|
|
## Model Details |
|
|
|
|
|
- **Base model**: GPT-Neo 125M |
|
|
- **Fine-tuned by**: Smilyai-labs |
|
|
- **Parameter count**: ~125 million |
|
|
- **Training examples**: ~1000 inline synthetic reasoning and dialogue samples |
|
|
- **Framework**: Hugging Face Transformers |
|
|
- **Trained in**: Google Colab |
|
|
- **Stored in**: Google Drive |
|
|
- **Uploaded to**: Hugging Face Hub |
|
|
|
|
|
## Intended Uses |
|
|
|
|
|
This model can be used for: |
|
|
- Learning how transformers work |
|
|
- Building experimental chatbots |
|
|
- Simple reasoning demos |
|
|
- Generating creative or silly responses |
|
|
|
|
|
## Example Usage |
|
|
|
|
|
```python |
|
|
from transformers import AutoTokenizer, AutoModelForCausalLM |
|
|
|
|
|
tokenizer = AutoTokenizer.from_pretrained("Smilyai-labs/Smily-ultra-1") |
|
|
model = AutoModelForCausalLM.from_pretrained("Smilyai-labs/Smily-ultra-1") |
|
|
|
|
|
prompt = "What is 2 + 2?" |
|
|
inputs = tokenizer(prompt, return_tensors="pt") |
|
|
outputs = model.generate(**inputs, max_new_tokens=30) |
|
|
print(tokenizer.decode(outputs[0])) |
|
|
``` |
|
|
|
|
|
## Limitations |
|
|
|
|
|
- Not accurate for factual tasks |
|
|
- Reasoning is simple and inconsistent |
|
|
- Can repeat or produce nonsensical outputs |
|
|
- Not safe for critical systems or real-world advice |
|
|
- Small training data limits its knowledge |
|
|
|
|
|
## Training |
|
|
|
|
|
- Trained for 3 epochs on ~1000 examples |
|
|
- Used Hugging Face `Trainer` API |
|
|
- Mixed reasoning and chatbot-style prompts |
|
|
- Stored in Google Drive and uploaded via `HfApi` |
|
|
|
|
|
## License |
|
|
|
|
|
MIT License or similar open-source license |
|
|
|
|
|
## Citation |
|
|
|
|
|
``` |
|
|
@misc{smilyultra1, |
|
|
author = {Smilyai-labs}, |
|
|
title = {Smily-ultra-1: Chatbot + Reasoning Toy Model}, |
|
|
year = 2025, |
|
|
url = {https://huggingface.co/Smilyai-labs/Smily-ultra-1} |
|
|
} |
|
|
``` |
|
|
|
