README.md

---
license: apache-2.0
tags:
- adam
- curious-architecture
- instruction-tuned
- conversational-ai
- 2b-parameters
library_name: transformers
pipeline_tag: text-generation
---

# Adam: Instruction-Tuned Conversational AI

<div align="center">
  <img src="https://img.shields.io/badge/Parameters-2B-blue" alt="2B Parameters">
  <img src="https://img.shields.io/badge/Architecture-CuriousForCausalLM-green" alt="Curious Architecture">
  <img src="https://img.shields.io/badge/Instruction%20Tuned-Yes-orange" alt="Instruction Tuned">
  <img src="https://img.shields.io/badge/Context%20Length-8K-purple" alt="8K Context">
</div>

## 🚀 Model Overview

**Adam** is a powerful 2 billion parameter language model built with the Curious architecture, specifically instruction-tuned for high-quality conversational AI and task completion. This model represents the next generation of efficient, instruction-tuned language models optimized for natural conversations.

## ✨ Key Features

- **🏗️ Native Curious Architecture**: Custom `CuriousForCausalLM` architecture with Curious-specific optimizations
- **🎯 Instruction-Tuned**: Fine-tuned for conversational AI and task completion
- **⚡ Efficient**: 2B parameters with optimized inference
- **💬 Conversational**: Specialized for natural dialogue and helpful responses
- **🔧 Advanced Features**: Sliding window attention, logit softcapping, and enhanced activations

## 📊 Model Specifications

| Parameter | Value |
|-----------|-------|
| **Architecture** | CuriousForCausalLM |
| **Model Type** | curious_text |
| **Parameters** | ~2.6B |
| **Context Length** | 8,192 tokens |
| **Vocabulary** | 256,000 tokens |
| **Training** | Instruction-tuned |
| **Curious Version** | 2.0 |

## 🎯 Capabilities

- **Natural Conversations**: Engaging and contextually aware dialogue
- **Question Answering**: Accurate responses to diverse queries
- **Creative Writing**: Poetry, stories, and creative content generation
- **Code Assistance**: Programming help and code generation
- **Mathematical Reasoning**: Problem-solving and calculations
- **Instruction Following**: Precise task execution and completion

## 🚀 Quick Start


### Interactive Chat

```python
pip install requirements.txt
```

```python
# Use the included chat interface
python chat_with_adam.py to talk to adam.
```

## 🏗️ Curious Architecture Features

- **Enhanced Attention**: Advanced attention mechanisms for better context understanding
- **Sliding Window**: Efficient processing of long sequences
- **Logit Softcapping**: Improved generation stability
- **Optimized Activations**: GELU with PyTorch tanh for better performance
- **Instruction Tuning**: Specialized for conversational AI tasks

## 📈 Performance

- **Quality**: High-quality instruction-tuned responses
- **Speed**: Optimized for efficient inference
- **Memory**: ~5GB model size
- **Hardware**: GPU recommended for best performance
- **Context**: 8K token context window

## 🔧 Technical Details

### Model Configuration

```json
{
  "architectures": ["CuriousForCausalLM"],
  "model_type": "curious_text",
  "hidden_size": 2304,
  "num_attention_heads": 8,
  "num_hidden_layers": 26,
  "max_position_embeddings": 8192,
  "curious_version": "2.0",
  "curious_instruction_tuned": true
}
```

### Generation Parameters



## 🎨 Use Cases

- **Chatbots**: Conversational AI applications
- **Assistants**: Task-oriented AI helpers
- **Creative Writing**: Content generation and editing
- **Education**: Tutoring and explanation
- **Coding**: Programming assistance
- **Research**: Information synthesis and analysis

## ⚠️ Limitations

- **Context Length**: Limited to 8K tokens
- **Training Data**: Cutoff date applies to training data
- **Bias**: May reflect biases in training data
- **Factual Accuracy**: Should be verified for critical applications


## 🙏 Acknowledgments

- Built with the Curious Architecture Framework
- Instruction-tuned for conversational AI
- Powered by the Curious Architecture Framework v2.0

---

<div align="center">
  <strong>Adam: The Future of Conversational AI</strong><br>
  <em>Built with ❤️ using the Curious Architecture Framework</em>
</div>