STACKS

STACKS

STACKS (Specialized Text AI for Creative Knowledge Synthesis) is a fine-tuned version of google/gemma-3-270m specifically trained to generate creative and detailed prompts for role-playing scenarios. Instead of responding to prompts, STACKS creates prompts when given a role or character type.

🎯 Model Description

STACKS transforms any role or profession into detailed, engaging prompts perfect for:

• Role-playing games and creative writing
• Educational scenarios and training simulations
• Content creation and storytelling
• Character development exercises

🚀 Quick Start

with torch.no_grad(): outputs = model.generate( **inputs, max_new_tokens=200, temperature=0.8, do_sample=True, top_p=0.9, repetition_penalty=1.1, pad_token_id=tokenizer.eos_token_id )

response = tokenizer.decode(outputs, skip_special_tokens=True) return response[len(input_text):].strip()

📋 Example Outputs

Input: "chef" STACKS Output: "Act as a world-renowned chef who specializes in fusion cuisine. You've just opened a new restaurant that combines traditional techniques with molecular gastronomy. Explain your signature dish and the story behind its creation."

Input: "time traveler" STACKS Output: "You are a time traveler who has just arrived in the year 2024 from 1850. Describe your amazement at modern technology, but also share wisdom from the past that people today have forgotten."

Input: "detective" STACKS Output: "Act as a private detective in 1940s noir-style setting. You've been hired to investigate a mysterious disappearance at a high-society party. Describe the case and your first impressions of the suspects."

🎭 Supported Role Categories

STACKS excels at generating prompts for:

• Professional Roles: doctors, lawyers, teachers, engineers, scientists
• Creative Roles: artists, writers, musicians, designers, filmmakers
• Historical Figures: philosophers, explorers, inventors, rulers
• Fictional Characters: superheroes, fantasy characters, sci-fi roles
• Specialized Experts: consultants, coaches, advisors, mentors
• Adventure Roles: explorers, adventurers, survivalists, travelers

🔧 Technical Details

Training Configuration

• Base Model: google/gemma-3-270m (268M parameters)
• Training Type: Complete fine-tuning (all parameters trainable)
• Dataset: fka/awesome-chatgpt-prompts
• Format: Role → Prompt generation patterns
• Precision: BF16 optimized
• Context Length: 768 tokens
• Training Date: 2025-08-20

Model Specifications

• Architecture: Gemma-3
• Parameters: 268,098,176
• Format: Safetensors
• Size: ~536MB
• Hardware: Optimized for GPU inference
• Attention: Eager implementation (required for Gemma-3)

📊 Performance & Quality

STACKS generates:

• Coherent prompts that match the requested role
• Creative scenarios with engaging storylines
• Detailed instructions for effective role-playing
• Varied outputs avoiding repetitive patterns
• Contextually appropriate content for each role

🎯 Usage Patterns

Basic Generation

📋 License

This model is released under the Gemma License. Please see the Gemma License for complete terms and conditions.

---

Built with ❤️ by gouthamsai78 Transforming roles into creative prompts, one generation at a time.