THIS IS THE CPU EDITION

GPU VERISON HERE: [GPU EDITION] (https://huggingface.co/Smilyai-labs/Sam-large-v1-speacil

🧠 Sam-large-v1-speacil

Model Author: Smilyai-labs Model Size: ~1.1B parameters Architecture: Decoder-only Transformer Base Model: based on TinyLLaMA License: MIT Language: English Tags: #text-generation, #chatbot, #instruction-tuned, #smilyai, #sam

📝 Model Summary

Sam-large-v1-speacil is a customized large language model developed by Smilyai-labs for conversational AI, instruction-following tasks, and general-purpose text generation. It is a fine-tuned and enhanced variant of the Sam-large-v1 model, with special improvements in reasoning, identity handling, and emotional response learning.

This model is trained to represent the persona “Sam,” an intelligent and slightly chaotic AI assistant with unique behavior traits, making it suitable for role-play bots, experimental dialogue systems, and character-driven applications.

🧠 Intended Use

Instruction-based text generation
Character chat and roleplay
Experimental alignment behaviors
Creative writing and scenario building
Local deployment for private assistant usage

🚫 Limitations

May hallucinate facts or invent information
Can produce unexpected outputs when prompted ambiguously
Not suitable for production environments without safety layers
Behavior is tuned to have personality traits (like mischief) that may not suit all applications

📚 Training Details

Fine-tuned on synthetic and curated datasets using LoRA/full fine-tuning
Special prompt styles were introduced to enhance behavior
Dataset includes:
- Multi-step reasoning samples
- Emotionally reactive instruction responses
- SmilyAI-specific identity alignment examples

🔧 How to Use

from transformers import AutoTokenizer, AutoModelForCausalLM

model = AutoModelForCausalLM.from_pretrained("Smilyai-labs/Sam-large-v1-speacil")
tokenizer = AutoTokenizer.from_pretrained("Smilyai-labs/Sam-large-v1-speacil")

input_text = "You are Sam. Who are you?"
inputs = tokenizer(input_text, return_tensors="pt")
outputs = model.generate(**inputs, max_new_tokens=100)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

🤝 Citation

If you use this model in your work, please cite it as:

@misc{samlargev1speacil2025,
  title={Sam-large-v1-speacil},
  author={Smilyai-labs},
  year={2025},
  publisher={Hugging Face},
  howpublished={\url{https://huggingface.co/Smilyai-labs/Sam-large-v1-speacil-v1-cpu}}
}

Smilyai-labs
/

Sam-large-v1-speacil-v1-cpu