Tohidichi/gemma3-genz16-270m

Gemma 3 270M Fine-Tuned with LoRA

This model is a fine-tuned derivative of Google's Gemma 3 270M using LoRA. fp16 version
It was fine-tuned by on a small dataset of Gen Z conversations in Hinglish, focusing on casual interactions among college students.

fp32 one is here : link to fp32

Model Details

• Developed by: Toheed Akhtar
• Model type: Causal Language Model (text-generation)
• Language(s): Multilingual (Hinglish focus)
• License: Subject to Gemma Terms of Use
• Finetuned from model: google/gemma-3-270m-it

Intended Use

This model is designed for casual text generation, simulating informal Gen Z conversations in Hinglish. It is mainly intended for personal experimentation.

Out-of-Scope Use

• The model may not produce accurate or safe content for critical applications.

How to Get Started

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline
from peft import PeftModel

# Set device for pipeline
device = 0 if torch.cuda.is_available() else -1  # 0 = first GPU, -1 = CPU

# Load base model
base_model_name = "google/gemma-3-270m-it"
base_model = AutoModelForCausalLM.from_pretrained(
    base_model_name,
    torch_dtype=torch.float16
)

# Load PEFT LoRA fine-tuned model from Hugging Face Hub
peft_model_hf = "Tohidichi/gemma3-genz16-270m"
model = PeftModel.from_pretrained(base_model, peft_model_hf)
model.eval()

# Load tokenizer from the PEFT model repo
tokenizer = AutoTokenizer.from_pretrained(peft_model_hf)

# Create text-generation pipeline
text_gen_pipeline = pipeline(
    "text-generation",
    model=model,
    tokenizer=tokenizer,
    device=device
)