|
|
--- |
|
|
base_model: TinyLlama/TinyLlama-1.1B-Chat-v1.0 |
|
|
library_name: transformers |
|
|
license: mit |
|
|
language: |
|
|
- en |
|
|
- pt |
|
|
metrics: |
|
|
- accuracy |
|
|
pipeline_tag: text-generation |
|
|
tags: |
|
|
- education |
|
|
- logic |
|
|
- math |
|
|
- low-resource |
|
|
- transformers |
|
|
- open-source |
|
|
- causal-lm |
|
|
- lambdaindie |
|
|
--- |
|
|
|
|
|
# lambdAI — Lightweight Math & Logic Reasoning Model |
|
|
|
|
|
**lambdAI** is a compact, fine-tuned language model built on top of `TinyLlama-1.1B-Chat-v1.0`, designed for educational reasoning tasks in both Portuguese and English. It focuses on logic, number theory, and mathematics, delivering fast performance with minimal computational requirements. |
|
|
|
|
|
## Model Architecture |
|
|
|
|
|
- **Base Model**: TinyLlama-1.1B-Chat |
|
|
- **Fine-Tuning Strategy**: LoRA (applied to `q_proj` and `v_proj`) |
|
|
- **Quantization**: 8-bit (NF4 via `bnb_config`) |
|
|
- **Dataset**: [`HuggingFaceH4/MATH`](https://huggingface.co/datasets/HuggingFaceH4/MATH) — subset: `number_theory` |
|
|
- **Max Tokens per Sample**: 512 |
|
|
- **Batch Size**: 20 per device |
|
|
- **Epochs**: 3 |
|
|
|
|
|
## Example Usage (Python) |
|
|
|
|
|
```python |
|
|
from transformers import AutoTokenizer, AutoModelForCausalLM |
|
|
|
|
|
model = AutoModelForCausalLM.from_pretrained("lambdaindie/lambdai") |
|
|
tokenizer = AutoTokenizer.from_pretrained("lambdaindie/lambdai") |
|
|
|
|
|
input_text = "Problema: Prove que 17 é um número primo." |
|
|
inputs = tokenizer(input_text, return_tensors="pt") |
|
|
|
|
|
output = model.generate(**inputs, max_new_tokens=100) |
|
|
print(tokenizer.decode(output[0], skip_special_tokens=True))``` |
|
|
|
|
|
About Lambda |
|
|
|
|
|
Lambda is an indie tech startup founded by Marius Jabami in Angola, focused on AI-driven educational tools, automation, and lightweight software solutions. The lambdAI model is the first release in a planned series of educational LLMs optimized for reasoning, logic, and low-resource deployment. |
|
|
|
|
|
Stay updated on the project at lambdaindie.github.io and huggingface.co/lambdaindie. |
|
|
|
|
|
|
|
|
--- |
|
|
|
|
|
Developed with care by Marius Jabami — Powered by ambition and open source. |
|
|
|
|
|
--- |
|
|
|
|
|
|
