boopathiraj/gemma-3-instruct-small

Gemma-3 Instruct Small (LoRA Merged)

Model Summary

Gemma-3 Instruct Small is a lightweight instruction-following language model fine-tuned from Google’s Gemma-3-270M-IT using LoRA and later merged into the base model for efficient inference.

The model is optimized for:

• Instruction following
• Basic mathematical reasoning
• Short-form question answering
• Educational and experimental use

---

Model Details

Model Description

• Developed by: Boopathiraj
• Organization: Self (Independent)
• Model type: Causal Language Model (Instruction-tuned)
• Language(s): English
• License: Apache 2.0
• Finetuned from: google/gemma-3-270m-it

This model was trained using parameter-efficient fine-tuning (LoRA) and later merged into the base weights for standalone inference without PEFT dependencies.

---

Model Sources

• Base Model: https://huggingface.co/google/gemma-3-270m-it
• Repository: https://huggingface.co/boopathiraj/gemma-3-instruct-small

---

Uses

Direct Use

This model can be used directly for:

• Instruction-based text generation
• Simple math word problems
• Educational demos
• Lightweight inference on limited hardware

Example use cases:

• Chatbots
• Teaching assistants
• Rapid prototyping

---

Downstream Use

The model may be further fine-tuned for:

• Domain-specific Q&A
• Educational datasets
• Small-scale reasoning benchmarks

---

Out-of-Scope Use

This model is not intended for:

• Medical, legal, or financial advice
• High-stakes decision making
• Safety-critical applications
• Long-context reasoning

---

Bias, Risks, and Limitations

• Inherits biases from the base Gemma model and training data
• Limited reasoning depth due to small parameter count (270M)
• May produce incorrect or hallucinated answers
• Performance degrades on long or multi-step reasoning tasks

Recommendations

Users should:

• Validate outputs before use
• Avoid high-risk domains
• Treat results as assistive, not authoritative

---

How to Get Started

Inference Example

from transformers import AutoTokenizer, AutoModelForCausalLM
import torch

tokenizer = AutoTokenizer.from_pretrained(
    "boopathiraj/gemma-3-instruct-small",
    use_fast=False
)

model = AutoModelForCausalLM.from_pretrained(
    "boopathiraj/gemma-3-instruct-small",
    device_map="auto",
    dtype=torch.float16
)

model.eval()

prompt = "Solve the problem: What is 7 minus 3?"

inputs = tokenizer(prompt, return_tensors="pt").to(model.device)

with torch.no_grad():
    output = model.generate(**inputs, max_new_tokens=32)

print(tokenizer.decode(output[0], skip_special_tokens=True))