aditi-rao-123/hf_model_gemma

Gemma 2B Instruction LoRA Adapter (CPU Trained)

Model Description

This repository contains a LoRA (Low-Rank Adaptation) adapter fine-tuned on top of Google’s Gemma 2B Instruct model (google/gemma-2-2b-it).

The adapter was trained entirely on CPU using PEFT and TRL’s SFTTrainer for instruction-following tasks.

⚠️ This repository contains only LoRA adapter weights, not the full base model.

---

Model Details

• Developed by: Aditi Rao
• Shared by: Aditi Rao
• Model type: Causal Language Model (LoRA Adapter)
• Language(s): English
• License: Gemma License
• Finetuned from: google/gemma-2-2b-it

---

Training Details

Training Method

• Fine-tuning approach: LoRA (PEFT)
• Trainer: trl.SFTTrainer
• Precision: float32
• Hardware: CPU only

LoRA Configuration

• Rank (r): 16
• Alpha: 32
• Dropout: 0.05
• Target modules:
  • q_proj
  • k_proj
  • v_proj
  • o_proj
  • gate_proj
  • up_proj
  • down_proj

Training Hyperparameters

• Epochs: 1
• Batch size: 1
• Gradient accumulation: 8
• Learning rate: 1e-4
• Max sequence length: 256
• Optimizer: AdamW

---

Training Data

• Custom instruction-style dataset in JSONL format
• Chat-formatted using Gemma chat template
• System messages normalized into user messages to comply with Gemma formatting rules

---

How to Get Started with the Model

Load the Adapter with the Base Model

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

base_model_id = "google/gemma-2-2b-it"
adapter_id = "aditi-rao-123/hf_model_gemma"

tokenizer = AutoTokenizer.from_pretrained(adapter_id)

base_model = AutoModelForCausalLM.from_pretrained(
    base_model_id,
    torch_dtype=torch.float32,
    device_map="cpu"
)

model = PeftModel.from_pretrained(base_model, adapter_id)

prompt = "Explain what a cryptocurrency wallet is."
inputs = tokenizer(prompt, return_tensors="pt")

outputs = model.generate(**inputs, max_new_tokens=128)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))