llama3-8b-lora-finetune

LLaMA 3 8B - LoRA Fine-Tuned for SAP ABAP Documentation

This model is a LoRA fine-tuned version of meta-llama/Meta-Llama-3-8B, specialized for generating technical documentation for SAP ABAP code, including custom classes and reports.

It was trained using the PEFT framework with QLoRA and DeepSpeed optimizations.

---

Model Details

Model Description

This model is designed to help ABAP developers by automatically generating technical documentation for legacy and custom ABAP code artifacts. It was trained on proprietary, structured ABAP class and report files.

• Developed by: Anu Reddy
• Model type: Causal Language Model (LoRA on LLaMA 3 8B)
• Language(s): English (output), SAP ABAP (input)
• License: apache-2.0
• Finetuned from model: meta-llama/Meta-Llama-3-8B

Model Sources

• GitHub Codebase: Fine-tuning Scripts

---

Uses

Direct Use

• Generate technical documentation for SAP ABAP classes and reports.
• Summarize the purpose and structure of ABAP methods and modules.
• Assist in legacy code understanding and documentation automation.

Downstream Use

• Integrate into RAG systems for SAP ABAP documentation search.
• Use in internal tools for automated doc generation pipelines.

Out-of-Scope Use

• General-purpose chat tasks
• Code generation outside ABAP domain
• Production environments without further evaluation

---

How to Get Started

from transformers import AutoTokenizer, AutoModelForCausalLM
from peft import PeftModel

base_model = AutoModelForCausalLM.from_pretrained("meta-llama/Meta-Llama-3-8B")
model = PeftModel.from_pretrained(base_model, "Anu123/llama3-8b-lora-finetune")
tokenizer = AutoTokenizer.from_pretrained("meta-llama/Meta-Llama-3-8B")

input_ids = tokenizer("CLASS zcl_invoice_handler DEFINITION PUBLIC CREATE PUBLIC .", return_tensors="pt").input_ids
output = model.generate(input_ids, max_new_tokens=300)
print(tokenizer.decode(output[0], skip_special_tokens=True))