🧃 Model Card: Ad Generator for Marketing Copy

This is a fine-tuned version of Microsoft's phi-2 language model, adapted for generating high-quality marketing content such as ad copy, slogans, and promotional text. It uses prompt-response training to structure outputs fluently and persuasively.

Model Details

Model Description

A fine-tuned Causal Language Model (CLM) based on microsoft/phi-2, optimized to produce structured marketing text with consistent formatting and clarity.

Developed by: Adnane Touiyate
Shared by : Adnane10
Model type: Causal Language Model (phi-2)
Language(s): English
License: MIT
Finetuned from model: microsoft/phi-2

Uses

✅ Direct Use

Marketing teams can input a product name and short description to generate ad copy Copywriters seeking inspiration or quick content drafts Startup founders, product teams, or solopreneurs generating headlines and taglines

🚫 Out-of-Scope Use

Not intended for factual, academic, or scientific content generation Not suitable for generating personal, sensitive, or confidential information May not generalize well to domains outside of marketing or product promotion

⚠️ Bias, Risks, and Limitations

While the model generates fluent and persuasive marketing text, it may:

Include overly generic, exaggerated, or unverifiable claims Mimic clichés or stereotypes from marketing-focused training data Lack fact-checking for health-related, numerical, or product safety statements

🔍 Recommendations

Use human review and editing before publishing outputs Consider further fine-tuning the model on your brand voice, domain, or regulatory constraints if needed

🚀 How to Get Started with the Model

from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("Adnane10/AdsGeniusAI")
model = AutoModelForCausalLM.from_pretrained("Adnane10/AdsGeniusAI")

prompt = "Create an ad for a vegan skincare brand that emphasizes natural ingredients and sustainability."
inputs = tokenizer(prompt, return_tensors='pt').to('cuda')
output = model.generate(**inputs, max_length=256, num_beams=5, temperature=0.7)
print(tokenizer.decode(output[0], skip_special_tokens=True))

📊 Training Details

Training Data

Fine-tuned on a dataset of curated product advertisements and promotional templates, covering sectors such as:

Food & Beverage Tech & Gadgets Beauty & Skincare Fitness & Wellness

Training Procedure

Precision: fp16 mixed precision Quantization: 4-bit (nf4) using BitsAndBytes Optimizer: AdamW Scheduler: Linear warmup + cosine decay Epochs: 3–6 (early stopping used) Framework: Hugging Face transformers, peft, accelerate, and bitsandbytes

📈 Evaluation

Metrics

BLEU / ROUGE: For structural and surface evaluation Human Evaluation: Based on fluency, creativity, and relevance Manual Checks: On repetition and prompt adherence

🌱 Environmental Impact

Carbon emissions can be estimated using the Machine Learning Impact calculator presented in Lacoste et al. (2019).

Hardware Type: NVIDIA Tesla T4
Hours used: ~2 hours
Cloud Provider: Kaggle
Estimated Carbon Emission: < 0.5 kg CO₂

🔧 Technical Specifications

Base Model: microsoft/phi-2 (~2.7B parameters)
Tokenizer: AutoTokenizer from phi-2
Quantization: 4-bit (NF4 with FP16 compute)

Libraries Used:

transformers
peft
accelerate
bitsandbytes

Model Architecture and Objective

The model is based on Microsoft’s phi-2, a small-scale language model focused on reasoning and general-purpose NLP tasks. It was fine-tuned as a Causal Language Model (CLM) to generate high-quality, structured advertising copy using prompt-response style formatting. Quantized to 4-bit using bitsandbytes for efficiency.

📚 Citation [optional]

@misc{freshpress-adgen, title={FreshPress Ad Generator}, author={Adnane Touiyate}, year={2025}, url={https://huggingface.co/Adnane10/phi2-marketing-generator}, note={Fine-tuned Phi-2 model for marketing and ad copy generation} }

✍️ Model Card Authors

Adnane Touiyate (@Adnane10)

📬 Contact

For questions or collaborations, reach out via LinkedIn or email: adnanetouiayte11@gmail.com