|
|
--- |
|
|
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** |
