|
--- |
|
library_name: transformers |
|
license: apache-2.0 |
|
datasets: |
|
- Abhaykoul/Dhanishtha-R1 |
|
- open-thoughts/OpenThoughts-114k |
|
language: |
|
- en |
|
base_model: |
|
- deepseek-ai/DeepSeek-R1-Distill-Qwen-7B |
|
--- |
|
# Dhanishtha Overview |
|
|
|
Dhanishtha is a cutting-edge reasoning AI model developed by **HelpingAI**, designed for deep introspection and structured logical analysis. Unlike traditional models that generate immediate responses, Dhanishtha employs a unique **deep-thinking process** process—an internal deliberation phase that enhances reasoning depth before presenting refined answers. |
|
|
|
## Model Capabilities |
|
Dhanishtha operates in **Dhanishtha Mode**, inspired by the **Dhanishtha Nakshatra**, known for wisdom, rhythm, and intellectual depth. The model engages in a multi-step thought process before providing responses, ensuring high accuracy and coherence. |
|
|
|
### Key Features: |
|
- **Structured Internal Reasoning:** Engages in self-dialogue within `<think></think>` tags, iterating through ideas and refining its thought process before responding. |
|
- **Progressive Thought Refinement:** Evaluates multiple perspectives, making logical connections and ensuring a well-rounded answer. |
|
- **Emotionally Intelligent Conversational Style:** Responses are expressive, engaging, and tailored for natural human interaction. |
|
- **Optimized for Critical Thinking & Problem-Solving:** Excels in analytical reasoning, debate, and deep philosophical discussions. |
|
- **Context Awareness:** Maintains logical coherence in extended interactions, avoiding contradictions and ensuring smooth thought progression. |
|
|
|
## Training & Architecture |
|
- **Model Size:** Optimized for high-performance reasoning with balanced efficiency. |
|
- **Training Approach:** Fine-tuned using advanced structured learning techniques to enhance deliberative thinking and introspective processing. |
|
- **Data Sources:** Trained on a diverse dataset covering philosophy, critical reasoning, and problem-solving scenarios to develop a deep intellectual foundation. |
|
|
|
## Performance & Benchmarks |
|
Dhanishtha outperforms conventional models in structured reasoning and contextual depth. The model has been rigorously evaluated across various metrics, demonstrating significant improvements in: |
|
- **Logical Coherence & Argumentation:** Enhanced ability to follow complex discussions and construct persuasive arguments. |
|
- **Depth of Analysis:** Excels in breaking down intricate topics into clear, structured responses. |
|
- **Adaptive Conversational Flow:** Seamlessly shifts between casual and analytical tones based on user input. |
|
|
|
## Deployment & Use Cases |
|
Dhanishtha is designed for: |
|
- **High-precision academic and philosophical discussions** |
|
- **Deep problem-solving and strategic reasoning** |
|
- **Engaging and thought-provoking conversations** |
|
- **Use in AI-driven research and advanced dialogue systems** |
|
|
|
## Benchmarks |
|
We report Pass@1 accuracy averaged over 16 samples for each problem. |
|
|
|
| Model | AIME 2024 | MATH 500 | AMC 2023 | Minerva Math | OlympiadBench | Avg. | |
|
|-------------------------------|-----------|----------|----------|--------------|----------------|------| |
|
| 2.5-7B-Instruct | 13.3 | 79.8 | 50.6 | 34.6 | 40.7 | 43.8 | |
|
| rStar-Math-7B | 26.7 | 78.4 | 47.5 | - | 47.1 | - | |
|
| Eurus-2.7B-PRIME | 26.7 | 79.2 | 57.8 | 38.6 | 42.1 | 48.9 | |
|
| Qwen2.5-7B-SimpleRL | 26.7 | 82.4 | 62.5 | 39.7 | 43.3 | 50.9 | |
|
| DeepSeek-R1-Distill-Qwen-1.5B | 28.8 | 82.8 | 62.9 | 26.5 | 43.3 | 48.9 | |
|
| Still-1.5B | 32.5 | 84.4 | 66.7 | 29.0 | 45.4 | 51.6 | |
|
| DeepScaleR-1.5B-Preview | 43.1 | 87.8 | 73.6 | 30.2 | 50.0 | 57.0 | |
|
| O1-Preview | 40.0 | 81.4 | - | - | - | - | |
|
| **Dhanishta** | 38.2 | 85.1 | 70.3 | 30.5 | 42.0 | 53.2 | |
|
| **Dhanishta-Large** | - | - | - | - | - | - | |
|
|
|
|
|
## Credits & License |
|
Dhanishtha is developed and maintained by **HelpingAI**, pushing the boundaries of AI-driven introspection and structured reasoning. The model is open-source and community-driven, encouraging contributions and collaborative innovation. |
