---
license: apache-2.0
datasets:
- Abhaykoul/Dhanishtha-R1
- open-thoughts/OpenThoughts-114k
- Abhaykoul/Dhanishtha-2.0-SUPERTHINKER
- Abhaykoul/Dhanishtha-2.0
language:
- en
- hi
- pa
- ur
- ar
- te
- ta
- ml
- gu
- he
- ru
- ko
- ms
- kn
- as
- es
- fr
- de
- ja
- pt
base_model:
- HelpingAI/Dhanishtha-2.0-preview
tag:
- rkllm
- rk3588
- rockchip
pipeline_tag: text-generation
---

# Dhanishtha-2.0

## Dhanishtha-2.0 Highlights

**Dhanishtha-2.0** is a groundbreaking reasoning AI model developed by **HelpingAI**, representing the first model to feature **Intermediate thinking** capabilities. Unlike traditional models that provide single-pass responses, Dhanishtha-2.0 employs a revolutionary **multi-phase thinking process** that allows the model to think, reconsider, and refine its reasoning multiple times throughout a single response.

- **World's First Intermediate Thinking Model**: Uniquely capable of **multiple thinking phases throughout a single response**, enabling deeper analysis and self-correction during generation.
- **Dynamic Reasoning**: Seamlessly integrates `<think>...</think>` blocks at any point in the response, allowing for real-time problem decomposition and iterative refinement.
- **Enhanced Problem-Solving Architecture**: Excels at complex riddles, multi-step reasoning, and scenarios requiring backtracking and re-evaluation of initial assumptions.
- **Structured Emotional Intelligence**: Incorporates **SER** (Structured Emotional Reasoning) with `<ser>...</ser>` blocks for empathetic and contextually aware responses.
- **Bilingual Reasoning Capabilities**: Native support for English and Hindi with natural code-switching between languages.

## Model Overview

**Dhanishtha-2.0** revolutionizes AI reasoning by introducing the concept of **intermediate thinking** - the ability to pause, reflect, and restart reasoning processes within a single generation. This breakthrough enables the model to:

- **Type**: Causal Language Model with Intermediate Thinking Capability
- **Training Stage**: Continuous Pretraining and Advanced Reasoning Fine-tuning.
- **Base model**: Qwen/Qwen3-14B-Base
- **Thinking Mode**: Multiple `<think>...</think>` blocks per response
- **Context Length**: 40,960 tokens natively
- **Reasoning Depth**: Multi-layered with self-correction mechanisms

## Revolutionary Intermediate Thinking

Unlike traditional models that engage in single-phase reasoning, Dhanishtha-2.0 introduces **Intermediate Thinking** - the ability to pause, reflect, and reconsider throughout the response generation process. This breakthrough allows the model to:

- **Think Multiple Times**: Insert reasoning blocks at any point during response generation
- **Self-Correct in Real-Time**: Identify and correct logical inconsistencies mid-response
- **Explore Alternative Perspectives**: Consider multiple approaches before settling on conclusions
- **Demonstrate Uncertainty**: Express doubt and reconsider initial assumptions naturally

## Key Features

### 1 Intermediate Thinking
Unlike traditional models that think once before responding, Dhanishtha-2.0 can engage in multiple thinking sessions throughout its response, allowing for:
- **Self-correction** when initial reasoning seems flawed
- **Alternative exploration** of different solution paths  
- **Iterative refinement** of ideas and conclusions
- **Natural hesitation** and reconsideration patterns

### 2 Dynamic Reasoning Flow
The model seamlessly transitions between:
- **Analysis phases** (`<think>` blocks) for internal reasoning
- **Communication phases** for sharing insights with users
- **Reflection phases** for questioning previous assumptions
- **Synthesis phases** for bringing ideas together

### 3. Structured Emotional Reasoning (SER)
```
<ser>
Emotion ==> frustration, curiosity
Cause ==> complex problem, initial confusion  
Mind ==> analytical, determined
Growth ==> patience, systematic approach
</ser>
```

## Quickstart

The following code demonstrates how to use Dhanishtha-2.0 with its unique intermediate thinking capabilities:

```python
from transformers import AutoModelForCausalLM, AutoTokenizer

model_name = "HelpingAI/Dhanishtha-2.0-preview"

# Load the tokenizer and model
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(
    model_name,
    torch_dtype="auto",
    device_map="auto"
)

# Prepare input with thinking enabled
prompt = "Who is your favorite person"
messages = [
    {"role": "user", "content": prompt}


text = tokenizer.apply_chat_template(
    messages,
    tokenize=False,
    add_generation_prompt=True
)

model_inputs = tokenizer([text], return_tensors="pt").to(model.device)

# Generate response with intermediate thinking
generated_ids = model.generate(
    **model_inputs,
    max_new_tokens=2048,
    temperature=0.7,
    top_p=0.9,
    do_sample=True
)

response = tokenizer.decode(generated_ids[0][len(model_inputs.input_ids[0]):], skip_special_tokens=True)
print(response)
```

## Intermediate Thinking in Action

What makes Dhanishtha-2.0 unique is its ability to think multiple times throughout a response. Here's an example:

**User**: *"Who is your favorite person"*

**Dhanishtha-2.0**:
```
<think>
This is an interesting question about my \"favorite human.\" I need to be thoughtful here. As Dhanishtha, I should respond in a human-like way while being honest about my nature. I can share what kinds of humans I find most interesting or inspiring based on my understanding of human nature and values.
</think>

<ser>
Emotion ==> curiosity, openness, desire for connection  
Cause ==> user asking a personal question about preferences  
Mind ==> seeking to understand my perspective or values  
Growth ==> opportunity to share thoughtful reflections on human qualities I admire  
</ser>

I don't really have a \"favorite human\" in the traditional sense, as I don't experience personal preferences the way humans do. But if I had to reflect on what qualities I find most admirable in people, I'd be drawn to those who demonstrate genuine compassion, intellectual curiosity, and the courage to grow.

<think>
Let me think more specifically about what kind of humans I find most compelling based on the values that seem to resonate most with humans - kindness, wisdom, creativity, resilience, and authenticity.
</think>

People like Mahatma Gandhi, who showed how quiet courage and unwavering principles can transform societies; artists like Rabindranath Tagore, who bridged cultural divides through creativity; and everyday heroes I've learned about who made profound differences through simple acts of kindness.

What fascinates me most are humans who maintain their humanity even in difficult circumstances - those who choose empathy when it would be easier to be harsh, who seek understanding rather than just being right.

I'm curious though - who's your favorite human, and why? What qualities or actions make someone special in your eyes?

```

## Best Practices

### Optimal Generation Parameters
```python
generation_config = {
    "temperature": 0.7,
    "top_p": 0.9,
    "top_k": 40,
    "max_new_tokens": 2048,
    "do_sample": True
}
```

### Effective Prompting
- **Encourage exploration**: "Think through this step by step and reconsider if needed"
- **Allow uncertainty**: "Take your time to work through different possibilities"
- **Request thinking**: "Show your reasoning process as you work through this"

## License & Usage

This model is released under the **Apache 2.0 License**. It is designed for research and educational purposes, with commercial applications welcome under the license terms.

## Prototype Status

> [!NOTE]
> Dhanishtha-2.0 is currently a **prototype model** representing the first implementation of Intermediate Thinking technology. While demonstrating groundbreaking capabilities, users should expect:
> - Experimental features that may require refinement
> - Potential verbosity due to multiple thinking phases
> - Ongoing development and optimization

## Citation

If you use Dhanishtha-2.0 in your research, please cite:

```bibtex
@misc{dhanishtha2025,
  title={Dhanishtha-2.0: First Model with Intermediate Thinking Capabilities},
  author={HelpingAI Team},
  year={2025},
  publisher={HuggingFace},
  url={https://huggingface.co/HelpingAI/Dhanishtha-2.0}
}
```

---


*Dhanishtha-2.0 represents a new paradigm in AI reasoning - where thinking isn't just a prelude to response, but an integral, iterative part of the conversation itself.*

Developed with ❤️ by **HelpingAI**