Soumyajit-7/adv-mathematics-reasoning-8b

Adv mathematics reasoning Model

This model is a fine-tuned version of unsloth/DeepSeek-R1-Distill-Llama-8B specialized for mathematical reasoning and problem-solving.

Model Description

• Base Model: DeepSeek-R1-Distill-Llama-8B
• Fine-tuning Method: LoRA (Low-Rank Adaptation)
• Dataset: Mathematical reasoning dataset with chain-of-thought explanations
• Specialization: Mathematical problem-solving with step-by-step reasoning

Features

• Chain-of-Thought Reasoning: The model thinks through problems step-by-step before providing answers
• Mathematical Expertise: Trained on mathematical problems and solutions
• Structured Responses: Provides both reasoning process and final answers

Usage

Direct Usage

from transformers import AutoTokenizer, AutoModelForCausalLM
import torch

# Load the model and tokenizer
model = AutoModelForCausalLM.from_pretrained("Soumyajit-7/adv-mathematics-reasoning-8b")
tokenizer = AutoTokenizer.from_pretrained("Soumyajit-7/adv-mathematics-reasoning-8b")

# Define the prompt format
prompt = '''Below is an instruction that describes a task, paired with an input that provides further context. 
Write a response that appropriately completes the request. 
Before answering, think carefully about the question and create a step-by-step chain of thoughts to ensure a logical and accurate response.

### Instruction:
You are a mathematics expert with advanced knowledge in problem-solving, logical reasoning, and mathematical concepts. 
Please solve the following mathematics problem. 

### Question:
{}

### Response:
<think>'''

# Example usage
question = "If x + 5 = 12, what is the value of x?"
inputs = tokenizer([prompt.format(question)], return_tensors="pt")

with torch.no_grad():
    outputs = model.generate(
        **inputs,
        max_new_tokens=500,
        temperature=0.7,
        do_sample=True,
        pad_token_id=tokenizer.eos_token_id
    )

response = tokenizer.decode(outputs[0], skip_special_tokens=True)
print(response.split("### Response:")[1])

Using with Unsloth

from unsloth import FastLanguageModel

model, tokenizer = FastLanguageModel.from_pretrained(
    model_name="Soumyajit-7/adv-mathematics-reasoning-8b",
    max_seq_length=2048,
    dtype=None,
    load_in_4bit=True,
)

FastLanguageModel.for_inference(model)
# Use the model for inference...

Training Details

• Training Framework: Unsloth + TRL
• LoRA Rank: 16
• LoRA Alpha: 16
• Target Modules: q_proj, k_proj, v_proj, o_proj, gate_proj, up_proj, down_proj
• Learning Rate: 2e-4
• Batch Size: 2 (with gradient accumulation)
• Optimizer: AdamW 8-bit

Model Performance

This model excels at:

• Mathematical problem-solving
• Step-by-step reasoning
• Chain-of-thought explanations
• Arithmetic and algebraic problems
• Logical reasoning tasks

Limitations

• Specialized for mathematical reasoning; may not perform as well on general tasks
• Requires specific prompt format for optimal performance
• Limited to problems similar to the training data

License

This model is released under the Apache 2.0 license.

Citation

If you use this model, please cite:

@misc{adv-mathematics-reasoning,
  title={Adv Mathematics Reasoning Model},
  author={Soumyajit Biswas},
  year={2025},
  howpublished={\url{https://huggingface.co/Soumyajit-7/adv-mathematics-reasoning-8b}},
}

Acknowledgments

• Base model: DeepSeek-AI
• Fine-tuning framework: Unsloth
• Training framework: TRL