This model is a specialized fine-tune of Qwen2.5-7B-Instruct, optimized for generating high-engagement social media content. It has been trained on a curated dataset of over 5,300 professional examples covering Instagram, LinkedIn, YouTube, and Facebook formats. The model utilizes Low-Rank Adaptation (LoRA) to achieve high performance while maintaining parameter efficiency.

System Architecture

The model implements a QLoRA (Quantized Low-Rank Adaptation) architecture, injecting trainable rank-decomposition matrices into the frozen 4-bit quantized base model.

Quick Start

Prerequisites

pip install transformers peft torch

Inference Code

python

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

# Configuration
BASE_MODEL_ID = "Qwen/Qwen2.5-7B-Instruct"
ADAPTER_ID = "YOUR_USERNAME/qwen2.5-social-content-generator"

# Load Base Model
base_model = AutoModelForCausalLM.from_pretrained(
    BASE_MODEL_ID,
    torch_dtype=torch.bfloat16,
    device_map="auto"
)

# Load LoRA Adapter
model = PeftModel.from_pretrained(base_model, ADAPTER_ID)
tokenizer = AutoTokenizer.from_pretrained(BASE_MODEL_ID)

def generate_content(prompt, max_tokens=150):
    # Format prompt with ChatML template
    formatted_prompt = f"<|im_start|>system\nYou are a professional social media content generator.<|im_end|>\n<|im_start|>user\n{prompt}<|im_end|>\n<|im_start|>assistant\n"
    
    inputs = tokenizer(formatted_prompt, return_tensors="pt").to(model.device)
    
    with torch.no_grad():
        outputs = model.generate(
            **inputs,
            max_new_tokens=max_tokens,
            temperature=0.7,
            top_p=0.9,
            do_sample=True
        )
    
    response = tokenizer.decode(outputs[0], skip_special_tokens=True)
    return response.split("assistant")[-1].strip()

# Example Usage
print(generate_content("Write a professional LinkedIn post about AI trends in 2026"))

Training Details

Dataset Composition

The model was trained on a dataset of 5,334 samples with the following distribution:

Content Type	Count	Percentage	Description
Short-form	1,468	27%	8-30 words (Reels, Shorts)
Medium-form	1,831	34%	31-100 words (Captions, Updates)
Long-form	1,987	37%	100+ words (Articles, Threads)

Hyperparameters

Parameter	Value
Base Model	Qwen2.5-7B-Instruct
Quantization	4-bit (NF4)
LoRA Rank (r)	16
LoRA Alpha	32
Target Modules	q_proj, k_proj, v_proj, o_proj, gate_proj, up_proj, down_proj
Batch Size	16 (Effective)
Learning Rate	2e-4
Epochs	3
Optimizer	paged_adamw_32bit

Performance & Capabilities

The model demonstrates strong capabilities in the following areas:

Hook Generation: 73.8% of generated samples include engagement hooks.
Platform Optimization: Automatically adjusts tone and format for LinkedIn (Professional), Instagram (Visual/Casual), and Twitter (Concise).
Call-to-Action (CTA): consistently integrates relevant CTAs in 60% of outputs.

Example Outputs

Prompt: "Write an Instagram reel about productivity tips (20 words max)"

Output: Stop scrolling. 🛑 This productivity hack changed everything for me. Focus on one thing at a time. Save this. 📌

Prompt: "Create a professional LinkedIn post about AI trends"

Output: AI is not just a buzzword; it's a paradigm shift. In 2026, we are seeing a move towards agentic workflows that autonomously solve complex problems. Organizations that adapt now will lead the market. #AI #Innovation #FutureOfWork

Deployment Options

CPU / Mobile Inference

This model supports CPU-only inference for edge deployment.

python

# Load on CPU
model = AutoModelForCausalLM.from_pretrained(
    BASE_MODEL_ID,
    torch_dtype=torch.float32,
    device_map="cpu",
    low_cpu_mem_usage=True
)
model = PeftModel.from_pretrained(model, ADAPTER_ID)

Desktop CPU (i7/Ryzen 7): ~5-10 tokens/sec
Mobile/Edge: ~2-3 seconds per response via API

License

This model is released under the Apache 2.0 license, inheriting terms from Qwen2.5.

Citation

bibtex

@misc{qwen2.5-social-content-generator,
  author = {Your Name},
  title = {Social Media Content Generator - Qwen2.5 7B LoRA},
  year = {2026},
  publisher = {HuggingFace},
  howpublished = {\url{https://huggingface.co/YOUR_USERNAME/qwen2.5-social-content-generator}},
}

shivam909067/Social-media-creation

Social Media Content Generator - Qwen2.5 7B LoRA

Model Overview