falcon-7b-QLoRA-alpaca-arabic

🚀 Falcon-7b-QLoRA-alpaca-arabic

This repo contains a low-rank adapter for Falcon-7b fit on the Stanford Alpaca dataset Arabic version Yasbok/Alpaca_arabic_instruct.

Model Summary

• Model Type: Causal decoder-only
• Language(s): Arabic
• Base Model: Falcon-7B (License: Apache 2.0)
• Dataset: Yasbok/Alpaca_arabic_instruct
• License(s): Apache 2.0 inherited from "Base Model"

Model Details

The model was fine-tuned in 8-bit precision using 🤗 peft adapters, transformers, and bitsandbytes. Training relied on a method called QLoRA introduced in this paper. The run took approximately 3 hours and was executed on a workstation with a single A100-SXM NVIDIA GPU with 37 GB of available memory.

Model Date

June 10, 2023

Recommendations

We recommend users of this model to develop guardrails and to take appropriate precautions for any production use.

How to Get Started with the Model

Setup

# Install packages
!pip install -q -U bitsandbytes loralib einops
!pip install -q -U git+https://github.com/huggingface/transformers.git 
!pip install -q -U git+https://github.com/huggingface/peft.git
!pip install -q -U git+https://github.com/huggingface/accelerate.git

GPU Inference in 8-bit

This requires a GPU with at least 12 GB of memory.

First, Load the Model

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

# load the model
peft_model_id = "Ali-C137/falcon-7b-chat-alpaca-arabic"
config = PeftConfig.from_pretrained(peft_model_id)

model = AutoModelForCausalLM.from_pretrained(
    config.base_model_name_or_path,
    return_dict=True,
    device_map={"":0},
    trust_remote_code=True,
    load_in_8bit=True,
)

tokenizer = AutoTokenizer.from_pretrained(config.base_model_name_or_path)
tokenizer.pad_token = tokenizer.eos_token

model = PeftModel.from_pretrained(model, peft_model_id)

CUDA Info

• CUDA Version: 12.0
• Hardware: 1 A100-SXM
• Max Memory: {0: "37GB"}
• Device Map: {"": 0}

Package Versions Employed

• torch: 2.0.1+cu118
• transformers: 4.30.0.dev0
• peft: 0.4.0.dev0
• accelerate: 0.19.0
• bitsandbytes: 0.39.0
• einops: 0.6.1

This work is highly inspired from Daniel Furman's work, so Thanks a lot Daniel