Venomoussaversai
0.1B INT8 AI Model (Experimental)
This repository contains an experimental 0.1 billion (100M) parameter AI model, optimized for low-memory and edge-device inference.
The model is quantized to INT8 and stored using SafeTensors, making it suitable for deployment on CPU-only systems and mobile-class hardware.
📦 Repository Structure
2b/
├── model-00001-of-000xx.safetensors # INT8 weights (≤100MB shards)
├── config.json # Model architecture config
├── generation_config.json # Inference settings
├── tokenizer.json # Custom tokenizer
├── vocab.json # Vocabulary
├── merges.txt # BPE merges
├── tokenizer_config.json
├── special_tokens_map.json
└── README.md
🔧 Model Specifications
- Parameter Count: ~100,000,000 (0.1B)
- Model Type: Transformer (GPT-style)
- Precision: INT8 (quantized)
- Serialized Size: ~249 MB (including metadata)
- Weight Format: ".safetensors"
- Tokenizer: Custom BPE (domain-trained)
- Framework: PyTorch + Hugging Face
🎯 Intended Use Cases
- Pattern recognition
- Trend and probability forecasting
- Lightweight reasoning tasks
- Offline and edge-device inference
Not intended for:
- Real-time future prediction
- Medical, legal, or financial decision-making
- Autonomous control systems
🚀 Loading the Model
from transformers import AutoModelForCausalLM, AutoTokenizer
model_path = "./2b"
tokenizer = AutoTokenizer.from_pretrained(model_path)
model = AutoModelForCausalLM.from_pretrained(
model_path,
load_in_8bit=True,
device_map="auto"
)
⚠️ Limitations
- Outputs are probabilistic, not deterministic
- Performance depends on tokenizer quality and training data
- Some layers may remain in higher precision for stability
- No real-time data access
📜 License & Disclaimer
This model is provided for research and experimentation purposes only.
The author assumes no responsibility for misuse or misinterpretation of the model’s outputs.
✨ Author
Ananthu Sajeev
AI Research & Edge Model Development
🧠 Notes
Smaller models prioritize efficiency and control over scale.
At this size, data quality > parameter count.