---
license: apache-2.0
language:
- en
tags:
- qwen3
- fine-tuned
- hito
- hitonet
- reasoning
- conversational
- thinking
- adaptive-reasoning
- tree-of-thought
- hierarchical-reasoning
- cognitive-framework
- self-aware-ai
- anti-hallucination
- synthetic-data
- llama-cpp
- ollama
- text-generation-inference
pipeline_tag: text-generation
base_model: Qwen/Qwen3-1.7B
---

# Hito 1.7B

### Nested Cognitive Reasoning for Thoughtful AI

<p align="center">
  <img src="https://hitonet.com/assets/img/hitonet-logo-dark.png" alt="Hitonet" width="300"/>
</p>

<p align="center">
  <a href="https://huggingface.co/hitonet/hito-1.7b-GGUF">GGUF Downloads</a> •
  <a href="https://chat.hitonet.com">Web Chat</a> •
  <a href="https://platform.hitonet.com">API</a> •
  <a href="https://hitonet.com/research">Research Paper</a>
</p>

---

| Status | Parameters | Context | License |
|--------|-----------|---------|---------|
| Production | 1.7B | 32K | Apache 2.0 |

---

## Overview

Hito is a 1.7B parameter language model fine-tuned with **Nested Cognitive Reasoning (NCR)** - a novel architecture that enables structured, self-correcting thinking patterns. Unlike traditional models that produce linear outputs, Hito thinks in branching, hierarchical structures that mirror human cognition.

### Key Features

- **Structured Reasoning**: Uses cognitive tags (`<think>`, `<doubt>`, `<verify>`) for transparent thought processes
- **Self-Correction**: Built-in mechanisms to catch and correct errors mid-reasoning
- **Humble AI**: Acknowledges uncertainty and limitations rather than hallucinating
- **Efficient**: Runs on consumer hardware with GGUF quantizations available

---

## Benchmark Results

| Model | Params | Overall | Counting | Math | Cognitive Bias |
|-------|--------|---------|----------|------|----------------|
| GPT-5-mini | ~8B | **100%** | 100% | 100% | ✅ |
| Claude Haiku 4.5 | ~8B | 90% | 67% | 100% | ✅ |
| **Hito 1.7B** | **1.7B** | **80%** | **67%** | **100%** | **✅** |
| GPT-4o-mini | ~8B | 80% | 33% | 100% | ❌ |
| Qwen3 1.7B (base) | 1.7B | 17% | 0% | 17% | ❌ |

### The Bat and Ball Test

*"A bat and a ball cost $1.10 together. The bat costs $1.00 more than the ball. How much does the ball cost?"*

Most AI models (and humans) answer 10 cents. **That's wrong.**

| Model | Answer | Correct |
|-------|--------|---------|
| **Hito 1.7B** | **$0.05** | ✅ |
| Qwen3 1.7B (base) | $0.10 | ❌ |
| GPT-4o-mini | $0.10 | ❌ |

**Why Hito gets it right:**

```xml
<think>
<understand>Ball + Bat = $1.10, Bat = Ball + $1.00</understand>
<doubt>Intuition says 10 cents... but let me verify.</doubt>
<logic>
If ball = $0.10, bat = $1.10, total = $1.20. WRONG.
Let ball = x: x + (x + 1) = 1.10, 2x = 0.10, x = 0.05
</logic>
<verify>Ball $0.05 + Bat $1.05 = $1.10 ✓</verify>
</think>
The ball costs five cents.
```

---

## Cognitive Architecture

Hito uses a tree-structured reasoning system with four cognitive states:

| State | Focus | Tags Used |
|-------|-------|-----------|
| **Analytical** | Logic, accuracy | `<logic>`, `<verify>`, `<compare>` |
| **Creative** | Imagination, exploration | `<imagine>`, `<brainstorm>`, `<wild>` |
| **Empathetic** | Feelings, perspectives | `<emotion>`, `<empathize>`, `<mood>` |
| **Reflective** | Depth, meaning | `<reflect>`, `<doubt>`, `<honest>` |

### The Humble Tags

| Tag | Purpose |
|-----|---------|
| `<doubt>` | Question assumptions |
| `<honest>` | Admit errors |
| `<limits>` | Acknowledge knowledge gaps |
| `<confidence>` | Rate certainty level |
| `<verify>` | Double-check work |

---

## Quick Start

### Python (Transformers)

```python
from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained("hitonet/hito-1.7b", torch_dtype="auto", device_map="auto")
tokenizer = AutoTokenizer.from_pretrained("hitonet/hito-1.7b")

messages = [
    {"role": "system", "content": "You are Hito by Hitonet.com."},
    {"role": "user", "content": "What is 15% of 200?"}
]

inputs = tokenizer.apply_chat_template(messages, return_tensors="pt", add_generation_prompt=True).to(model.device)
outputs = model.generate(inputs, max_new_tokens=512, temperature=0.7, do_sample=True)
print(tokenizer.decode(outputs[0], skip_special_tokens=False))
```

### Ollama

```bash
# Download GGUF from hitonet/hito-1.7b-GGUF
ollama create hito -f Modelfile
ollama run hito
```

### API

```bash
curl https://api.hitonet.com/v1/chat/completions \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "hito", "messages": [{"role": "user", "content": "Hello!"}]}'
```

Try the full API at [platform.hitonet.com](https://platform.hitonet.com) - $1 free credit included.

---

## Model Variants

| Repository | Format | Use Case |
|------------|--------|----------|
| [hitonet/hito-1.7b](https://huggingface.co/hitonet/hito-1.7b) | Safetensors | Python/Transformers |
| [hitonet/hito-1.7b-GGUF](https://huggingface.co/hitonet/hito-1.7b-GGUF) | GGUF | Ollama/llama.cpp/LM Studio |

### Recommended GGUF Quantizations

| Quantization | Size | Quality | Use Case |
|--------------|------|---------|----------|
| Q4_K_M | 1.1 GB | ⭐ Best Balance | Most users |
| Q5_K_M | 1.2 GB | Excellent | Quality-focused |
| Q8_0 | 1.8 GB | Highest | Maximum quality |

---

## Training

Hito is fine-tuned from Qwen3-1.7B using Supervised Fine-Tuning (SFT) with synthetic data generated by our flagship Hito-Genius model. The training focuses on:

- **Cognitive Pattern Transfer**: Teaching structured reasoning through demonstration
- **Self-Correction Habits**: Training the model to verify its own work
- **Humility Patterns**: Learning to express uncertainty appropriately

---

## Research

For technical details on Nested Cognitive Reasoning, see our research paper:

**[Nested Cognitive Reasoning: A Tree-Structured Approach to Language Model Thinking](https://hitonet.com/research)**

*Hitonet Research, 2025*

---

## Licensing

| Component | License | Commercial Use |
|-----------|---------|----------------|
| **Model Weights** | Apache 2.0 | ✅ Free |
| **NCR Methodology** | CC BY-NC-ND | ⚠️ License Required |

The model weights are fully open source under Apache 2.0.

The Nested Cognitive Reasoning methodology (cognitive tags, tree-structured thinking, humble tags system) is protected under CC BY-NC-ND. Commercial use of the NCR method requires a license.

**Contact:** legal@hitonet.com

---

## Links

- **Website:** [hitonet.com](https://hitonet.com)
- **Chat:** [chat.hitonet.com](https://chat.hitonet.com)
- **API:** [platform.hitonet.com](https://platform.hitonet.com)
- **Research:** [hitonet.com/research](https://hitonet.com/research)
- **Blog:** [hitonet.com/blog](https://hitonet.com/blog)

---

<p align="center">
  <b>Made with genuine curiosity by Hitonet</b><br>
  <i>Teaching AI to think, doubt, and learn.</i>
</p>