130 lines
2.7 KiB
Markdown
130 lines
2.7 KiB
Markdown
---
|
|
license: apache-2.0
|
|
language:
|
|
- en
|
|
tags:
|
|
- science
|
|
- reasoning
|
|
- advanced-reasoning
|
|
- thinking
|
|
- small-model
|
|
- efficient
|
|
- education
|
|
- local
|
|
- qwen
|
|
- qwen3
|
|
- qwen3.5
|
|
- 4b
|
|
- small
|
|
- cot
|
|
- chainofthought
|
|
- deep-thinking
|
|
- physics
|
|
- chemistry
|
|
- biology
|
|
- logic
|
|
- daily-use
|
|
- localai
|
|
- ai
|
|
- gpt
|
|
- dqnlabs
|
|
- dqngpt
|
|
- gguf
|
|
- lmstudio
|
|
- ollama
|
|
pipeline_tag: text-generation
|
|
---
|
|
|
|
# dqnScience-v1
|
|
|
|
dqnScience-v1 is a 4B-parameter flagship reasoning model designed for deep thinking, scientific problem solving, and complex multi-step reasoning.
|
|
|
|
Unlike lightweight fast-response models, dqnScience-v1 is built to **think longer, reason deeper, and solve harder problems**—often performing far above its size class.
|
|
|
|
|
|
|
|
---
|
|
|
|
## Model Description
|
|
|
|
- **Model type:** Causal Language Model
|
|
- **Parameters:** 4B
|
|
- **Primary use:** Scientific reasoning and advanced problem solving
|
|
- **Style:** Deep, structured, step-by-step reasoning
|
|
|
|
dqnScience-v1 prioritizes **reasoning quality over speed**, making it ideal for problems that require careful thought, abstraction, and layered logic.
|
|
|
|
---
|
|
|
|
## Intended Uses
|
|
|
|
### Direct Use
|
|
|
|
- Solving physics, chemistry, and biology problems
|
|
- Logical and analytical reasoning tasks
|
|
- Multi-step problem solving
|
|
- Conceptual understanding of scientific topics
|
|
- Competitive exam-style questions (college level to moderate)
|
|
|
|
---
|
|
|
|
## Key Characteristics
|
|
|
|
- Strong multi-step reasoning ability
|
|
- Produces structured and detailed explanations
|
|
- Excels at breaking down complex problems
|
|
- Performs above typical 4B models in reasoning capability
|
|
- Designed for consistency and logical correctness
|
|
- Handles abstract and conceptual questions effectively
|
|
|
|
---
|
|
|
|
## Usage
|
|
|
|
dqnScience-v1 is available in multiple formats:
|
|
|
|
- **GGUF** → llama.cpp, LM Studio
|
|
- **MLX** → optimized for Apple Silicon (coming soon)
|
|
- **HF Transformers** → universal compatibility
|
|
|
|
---
|
|
|
|
## Training Details
|
|
|
|
dqnScience-v1 is fine-tuned with a strong focus on reasoning-heavy datasets, emphasizing:
|
|
|
|
- Deep chain-of-thought reasoning
|
|
- Scientific and logical problem solving
|
|
- Conceptual clarity over memorization
|
|
- Robust multi-step inference
|
|
|
|
---
|
|
|
|
## Limitations
|
|
|
|
- Slower than lightweight models due to deeper reasoning
|
|
- May over-explain simple questions
|
|
- Not optimized for casual or short-form responses
|
|
- Performance may vary on highly specialized or research-level topics
|
|
|
|
---
|
|
|
|
## Efficiency
|
|
|
|
Despite its strong reasoning capabilities, dqnScience-v1 is optimized to run moderately efficiently on consumer hardware, with support for quantized formats.
|
|
|
|
---
|
|
|
|
## License
|
|
|
|
Apache 2.0
|
|
|
|
---
|
|
|
|
## Author
|
|
|
|
Developed by DQN Labs.
|
|
|
|
Special thanks to Ram2 for quantization.
|
|
|
|
This model card was generated with the help of dqnGPT v1. |