---
library_name: transformers
tags:
- sft
- unsloth
- science
- reasoning
license: apache-2.0
datasets:
- mattwesney/CoT_Reasoning_Scientific_Discovery_and_Research
language:
- en
base_model:
- unsloth/Qwen3-1.7B
pipeline_tag: text-generation
---

![20250421_1023_Scientific Discovery Design_simple_compose_01jscbjwqdetvtd5phh85mwsa7.png](https://cdn-uploads.huggingface.co/production/uploads/65dbedfd2f6d2dfc27763b98/mXuepdf8ZDBDMtdaBx-nV.png)

# Model Card for Qwen3-CoT-Scientific-Research

## Model Details

### Model Description

- **Base Model:** Qwen3-1.7B
- **Task:** Scientific Reasoning with Chain-of-Thought (CoT)
- **Dataset:** CoT_Reasoning_Scientific_Discovery_and_Research (custom dataset focusing on step-by-step scientific reasoning tasks)
- **Training Objective:** Encourage step-by-step logical deductions for scientific reasoning problems

## Uses

### Direct Use

This fine-tuned model is designed for:

 - Assisting in teaching and learning scientific reasoning
 - Supporting educational AI assistants in science classrooms
 - Demonstrating step-by-step scientific reasoning in research training contexts
 - Serving as a resource for automated reasoning systems to better emulate structured scientific logic

It is not intended to replace human researchers, perform advanced analytics, or generate novel scientific discoveries.


## Bias, Risks, and Limitations

- May oversimplify complex or interdisciplinary problems
- Performance limited by the scope of training data (primarily introductory-level scientific reasoning tasks)
- Does not handle real-world experimentation or advanced statistical modeling
- May produce incorrect reasoning if the prompt is highly ambiguous


## How to Get Started with the Model

Use the code below to get started with the model.

```python
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("khazarai/Scie-R1")
model = AutoModelForCausalLM.from_pretrained(
    "khazarai/Scie-R1",
    device_map={"": 0}
)

question = """
How are microfluidic devices revolutionizing laboratory analysis techniques, and what are the primary advantages they offer over traditional methods?
"""

messages = [
    {"role" : "user", "content" : question}
]
text = tokenizer.apply_chat_template(
    messages,
    tokenize = False,
    add_generation_prompt = True,
    enable_thinking = True,
)

from transformers import TextStreamer
_ = model.generate(
    **tokenizer(text, return_tensors = "pt").to("cuda"),
    max_new_tokens = 1800,
    temperature = 0.6,
    top_p = 0.95,
    top_k = 20,
    streamer = TextStreamer(tokenizer, skip_prompt = True),
)
```

## Training Details

### Training Data

**Scope**

This model was fine-tuned on tasks that involve core scientific reasoning:

- Formulating testable hypotheses
- Identifying independent and dependent variables
- Designing simple controlled experiments
- Interpreting graphs, tables, and basic data representations
- Understanding relationships between evidence and conclusions
- Recognizing simple logical fallacies in scientific arguments

**Illustrative Examples**

- Drawing conclusions from experimental results
- Evaluating alternative explanations for observed data
- Explaining step-by-step reasoning behind scientific conclusions

**Emphasis on Chain-of-Thought (CoT)**

- The dataset highlights explicit reasoning steps, making the model better at producing step-by-step explanations when solving scientific reasoning tasks.
- Focus on Foundational Knowledge
- The dataset aims to strengthen models in foundational scientific reasoning skills rather than covering all domains of scientific knowledge.

**Focus on Foundational Knowledge**

The dataset aims to strengthen models in foundational scientific reasoning skills rather than covering all domains of scientific knowledge.

**Dataset:** [moremilk/CoT_Reasoning_Scientific_Discovery_and_Research](https://huggingface.co/datasets/moremilk/CoT_Reasoning_Scientific_Discovery_and_Research)