Vinay-11/Llama-lora-instruct-finetuning

LLaMA LoRA Instruction Fine-Tuning (FP16)

This repository contains LoRA adapter weights for a LLaMA language model fine-tuned on instruction-style question–answer data using FP16 precision. The goal is to improve instruction following and reasoning while keeping training efficient via parameter-efficient fine-tuning (PEFT).

Model Details

Base Model: LLaMA

Fine-Tuning Method: LoRA (Low-Rank Adaptation)

Precision: FP16

Task: Instruction Following / Question Answering

Frameworks: Hugging Face Transformers, PEFT, PyTorch

Checkpoint Format: .safetensors

🔹 This repository contains LoRA adapter weights only. 🔹 The base LLaMA model must be loaded separately.

Dataset

The model was fine-tuned using the FineTome-100k dataset, a curated collection of high-quality instruction–response pairs designed for supervised fine-tuning (SFT) of large language models.

Dataset Name: FineTome-100k

Type: Instruction / Q&A pairs

Size: ~100K samples

🔗 Dataset Link: https://huggingface.co/datasets/mlabonne/FineTome-100k

Training Procedure

Objective: Causal Language Modeling

Trainable Parameters: LoRA adapters only

Frozen Parameters: All base model weights

Optimizer: AdamW

Precision: FP16

Approach: Supervised Fine-Tuning (SFT)

This setup allows efficient adaptation of a large model without updating all parameters.

How to Use Requirements pip install transformers peft accelerate torch

Load Model with LoRA Adapters from transformers import AutoModelForCausalLM, AutoTokenizer from peft import PeftModel import torch

base_model_id = "meta-llama/Llama-2-7b-hf" adapter_id = "Vinay-11/Llama-lora-instruct-finetuning"

tokenizer = AutoTokenizer.from_pretrained(base_model_id)

base_model = AutoModelForCausalLM.from_pretrained( base_model_id, torch_dtype=torch.float16, device_map="auto" )

model = PeftModel.from_pretrained(base_model, adapter_id) model.eval()

Example Inference prompt = "What is LoRA fine-tuning?"

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

outputs = model.generate( **inputs, max_new_tokens=120, temperature=0.7, do_sample=True )

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