swami93/gpt-indian-names-generator

Indian Name Generator — GPT Fine-Tuned Model

This model generates Indian girl/boy baby names.
It was fine-tuned on 1.5 million unique Indian names containing both male and female names.

---

Model Details

Model Description

A fine-tuned version of openai-gpt from Hugging Face.
The model learns patterns in Indian names and generates unique, culturally appropriate names when prompted.

Property	Value
Base model	openai-gpt (12-layer transformer decoder)
Model parameters	116,536,320 (~116M)
Tokenizer vocab size	40,480 tokens
Number of transformer layers	12
Task	Causal Language Modeling / Name Generation
Language	Primarily Indian names (Tamil, Hindi, Sanskrit, Telugu, Kannada etc.)
Dataset size	1.5 million names
License	MIT

---

Finetuning Strategy

• Base model weights from openai-gpt
• Fine-tuned with 1.5M Indian names dataset
• Low learning rate applied to most layers
• High learning rate applied to the last transformer + attention heads

Result: Model retains GPT's language understanding but learns strong name-generation bias.

---

Uses

Direct Use

from transformers import AutoModelForCausalLM, OpenAIGPTTokenizer

repo_name = "swami93/gpt-indian-names-generator"

tokenizer = OpenAIGPTTokenizer.from_pretrained(repo_name)
model = AutoModelForCausalLM.from_pretrained(repo_name)
model = model.to(device)

prompt = "name:"

# Controls creativity in generated names: higher % more random/creative names, lower % more predictable names
randomness_percent = 120

input_ids = tokenizer.encode(
    prompt,
    return_tensors="pt"
).to(device)

output = model.generate(
    input_ids=input_ids,
    max_length=50,
    do_sample=True,
    top_k=30,
    top_p=0.9,
    temperature=randomness_percent/100
)

generated_names = tokenizer.decode(output[0])

for line in generated_names.split(separator_token):
  line = line.strip()
  if line and ":" in line and len(line) >= 5:
    print(line)