Shi-07/Llama3_8b-FineTuned-Gender_Classifier_by_Name

🦙 Llama-3 8B Fine-Tuned Gender Classifier (Indian Names)

This model is a high-performance, fine-tuned version of Meta Llama-3-8B, specifically trained to identify the gender of Indian names based on linguistic patterns and cultural naming conventions. It achieves clean, classification-ready output using Supervised Fine-Tuning (SFT).

🚀 Model Description

• Architecture: Llama-3-8B (Fine-tuned with Unsloth & PEFT/LoRA)

• Primary Task: Binary Gender Classification

• Dataset Source: Kaggle: Indian Names Dataset

• Training Labels:

  → 0: Male

  → 1: Female

🚀 How to Use

Recommended Prompt Template

To get the most accurate results, you must use this exact template:

### Instruction:
Identify the gender of the given Indian name.

### Input:
{name}

### Response:

🧑‍💻 Inference Code( Python )

To try the model, feel free to use the below code:

from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

model_id = "Shi-07/Llama3_8b-FineTuned-Gender_Classifier_by_Name"

tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
    model_id, 
    torch_dtype=torch.float16, 
    device_map="auto"
)

def predict_gender(name):
    prompt = f"### Instruction:\nIdentify the gender of the given Indian name.\n\n### Input:\n{name}\n\n### Response:\n"
    inputs = tokenizer(prompt, return_tensors="pt").to("cuda")
    outputs = model.generate(**inputs, max_new_tokens=2, do_sample=False)
    return tokenizer.decode(outputs[0], skip_special_tokens=True).split("### Response:")[-1].strip()

# Test run
print(predict_gender("Rahul")) # Output: male
print(predict_gender("Priya")) # Output: female

⚠️ Known Limitations & Biases

• Binary Training: The model was trained strictly on Male (0) and Female (1) labels.

• Neutral Names: If you provide a neutral name (e.g., Satya, Surya, Bala), the model will force a binary decision. Due to the training distribution, it typically defaults to male for neutral inputs.

• Out-of-Scope: This model is specialized for Indian names. Its performance on Western or other non-Indian names is not guaranteed.

📊 Evaluation Results

The model was evaluated on a balanced test set of 50 Indian names (25 Male, 25 Female).

Metric	Score
Accuracy	94.00%
F1-Score (Weighted)	0.9398

📜 Technical Specs

• Fine-tuning Tool: Unsloth (2x speed optimization)

• Precision: 4-bit Quantization (bitsandbytes)

• LoRA Hyperparameters: Rank r=16, Alpha 16, Dropout 0.05

• Learning Rate: 2e-4 with Linear Scheduler

🍦 Think of this model as your very own Indian naming expert — it's fast, it's precise

Happy Classifying! 🚀✨