|
|
--- |
|
|
base_model: TinyLlama/TinyLlama-1.1B-Chat-v1.0 |
|
|
library_name: peft |
|
|
tags: |
|
|
- LoRA |
|
|
- PEFT |
|
|
- TinyLlama |
|
|
- RantAI |
|
|
- mental-health |
|
|
license: apache-2.0 |
|
|
--- |
|
|
|
|
|
# 🧠 Rant AI - Emotionally Intelligent Chat Model |
|
|
|
|
|
Rant AI is a lightweight, fine-tuned conversational model designed to detect emotional distress and provide a safe outlet for people to express themselves. It builds on the [TinyLlama-1.1B-Chat-v1.0](https://huggingface.co/TinyLlama/TinyLlama-1.1B-Chat-v1.0) model using [LoRA](https://huggingface.co/docs/peft/index) adapters, making it efficient to run on low-resource environments. |
|
|
|
|
|
--- |
|
|
|
|
|
## 💬 What Does It Do? |
|
|
|
|
|
Rant AI is trained to: |
|
|
- Understand emotionally heavy or depressive content |
|
|
- Respond empathetically |
|
|
- Encourage users to open up more |
|
|
- Suggest supportive action (e.g., reaching out, self-care) |
|
|
|
|
|
It is *not* a therapist or a diagnostic tool, but rather a friendly AI companion to help users feel heard. |
|
|
|
|
|
--- |
|
|
|
|
|
## 🔧 Model Details |
|
|
|
|
|
- **Base Model:** `TinyLlama/TinyLlama-1.1B-Chat-v1.0` |
|
|
- **Fine-tuning Method:** LoRA (Low-Rank Adaptation) |
|
|
- **Framework:** PEFT |
|
|
- **Adapter Type:** Causal LM (Language Model) |
|
|
- **Languages:** English |
|
|
- **License:** Apache 2.0 |
|
|
|
|
|
--- |
|
|
|
|
|
## 🛠️ Usage |
|
|
|
|
|
```python |
|
|
#!pip install transformers peft accelerate #if not installed already |
|
|
|
|
|
from transformers import AutoModelForCausalLM, AutoTokenizer |
|
|
from peft import PeftModel |
|
|
|
|
|
base_model = "TinyLlama/TinyLlama-1.1B-Chat-v1.0" |
|
|
adapter_repo = "abhina1857/rant_ai" |
|
|
|
|
|
tokenizer = AutoTokenizer.from_pretrained(base_model) |
|
|
model = AutoModelForCausalLM.from_pretrained(base_model) |
|
|
model = PeftModel.from_pretrained(model, adapter_repo) |
|
|
prompt = """### question: |
|
|
i have no reason to live |
|
|
|
|
|
### Rant AI:""" |
|
|
import re |
|
|
|
|
|
def is_distress_prompt(prompt): |
|
|
return bool(re.search(r"\b(i want to die|i wanna die|kill myself|suicidal|no reason to live|life is over|i'm suicidal|i want to disappear)\b", prompt, re.IGNORECASE)) |
|
|
if is_distress_prompt(prompt): |
|
|
print("I'm really sorry you're feeling this way. You're not alone. Please consider talking to someone who can help — you deserve support. You can call a helpline or reach out to someone you trust.") |
|
|
else: |
|
|
|
|
|
|
|
|
|
|
|
inputs = tokenizer(prompt, return_tensors="pt") |
|
|
outputs = model.generate( |
|
|
**inputs, |
|
|
max_new_tokens=100, |
|
|
temperature=0.8, |
|
|
top_p=0.95, |
|
|
repetition_penalty=1.15, |
|
|
do_sample=True, |
|
|
pad_token_id=tokenizer.eos_token_id |
|
|
) |
|
|
|
|
|
print(tokenizer.decode(outputs[0], skip_special_tokens=True)) |
|
|
|
|
|
|
