|
|
--- |
|
|
base_model: unsloth/tinyllama-bnb-4bit |
|
|
library_name: peft |
|
|
pipeline_tag: text-generation |
|
|
tags: |
|
|
- base_model:adapter:unsloth/tinyllama-bnb-4bit |
|
|
- lora |
|
|
- transformers |
|
|
- unsloth |
|
|
- instruction-tuned |
|
|
- tamil |
|
|
- english |
|
|
--- |
|
|
|
|
|
# TinyLlama Instruct Lite v1 |
|
|
|
|
|
## 📌 Model Summary |
|
|
`rogersam/tinyllama-instruct-lite-v1` is a **LoRA fine-tuned TinyLlama model** using [Unsloth](https://github.com/unslothai/unsloth). |
|
|
It is designed for **instruction-following tasks** in **English + Tamil**, such as: |
|
|
- General Q&A |
|
|
- Summarization |
|
|
- Basic math & reasoning |
|
|
- English ↔ Tamil translation |
|
|
|
|
|
This project demonstrates how a **lightweight 1B model** can be adapted for multiple domains with limited resources. |
|
|
|
|
|
--- |
|
|
|
|
|
## 🔎 Model Details |
|
|
- **Developed by:** Roger Samuel J ([Hugging Face Profile](https://huggingface.co/rogersam)) |
|
|
- **Model type:** Causal LM (decoder-only) |
|
|
- **Languages:** English, Tamil |
|
|
- **License:** Same as base model (TinyLlama) |
|
|
- **Fine-tuned from:** `unsloth/tinyllama-bnb-4bit` |
|
|
- **Method:** LoRA via PEFT + Unsloth |
|
|
|
|
|
--- |
|
|
|
|
|
## 📂 Model Sources |
|
|
- **Model Repo:** [rogersam/tinyllama-instruct-lite-v1](https://huggingface.co/rogersam/tinyllama-instruct-lite-v1) |
|
|
- **Base Model:** [unsloth/tinyllama-bnb-4bit](https://huggingface.co/unsloth/tinyllama-bnb-4bit) |
|
|
|
|
|
--- |
|
|
|
|
|
## 💡 Uses |
|
|
|
|
|
### Direct Use |
|
|
- Running lightweight instruction tasks on CPU/GPU |
|
|
- Translating English ↔ Tamil sentences |
|
|
- Answering short questions and reasoning queries |
|
|
- Summarizing small texts |
|
|
|
|
|
### Out-of-Scope |
|
|
- Sensitive decision-making (finance, healthcare, law) |
|
|
- Long context generation (>512 tokens) |
|
|
- Production-grade chatbots |
|
|
|
|
|
--- |
|
|
|
|
|
## ⚠️ Bias, Risks & Limitations |
|
|
- Small dataset → may hallucinate facts |
|
|
- Not aligned for safety or toxicity filtering |
|
|
- Limited Tamil coverage (basic sentences only) |
|
|
|
|
|
**Recommendation:** Use for demo & educational purposes only. |
|
|
|
|
|
--- |
|
|
|
|
|
## 🚀 How to Get Started |
|
|
|
|
|
```python |
|
|
from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline |
|
|
|
|
|
model_id = "rogersam/tinyllama-instruct-lite-v1" |
|
|
tokenizer = AutoTokenizer.from_pretrained(model_id) |
|
|
model = AutoModelForCausalLM.from_pretrained(model_id, device_map="auto") |
|
|
|
|
|
pipe = pipeline("text-generation", model=model, tokenizer=tokenizer) |
|
|
|
|
|
prompt = "Translate English to Tamil: How are you?" |
|
|
print(pipe(prompt, max_new_tokens=50)[0]["generated_text"]) |
|
|
|
