NexusLLM-Math-1B-v1

NexusLLM-Math-1B-v1

Model Details

NexusLLM-Math-1B-v1 is a fine-tuned version of Llama 3.2 (1B parameters) optimized specifically for solving advanced high-school mathematics problems, with a focus on JEE Main and Advanced syllabus topics.

• Developed by: ZentithLLM
• Model Type: Causal Language Model (Fine-tuned with LoRA)
• Language: English
• Base Model: meta-llama/Llama-3.2-1B-Instruct
• Precision: FP16

Intended Use

This model is designed to act as an educational assistant for 11th-grade mathematics. It is trained to provide step-by-step reasoning and explanations for complex topics, rather than just outputting the final answer.

Primary Topics Covered:

• Binomial Theorem
• Geometry (Circle Theorems, cyclic quadrilaterals, tangents, etc.)

Training Data

The model was trained on a custom dataset of structured mathematics Q&A pairs. The dataset maps specific mathematical prompts to detailed completions, heavily utilizing an explanation field to teach the model the underlying mathematical logic and derivation steps.

Training Procedure

The model was fine-tuned using the standard Hugging Face trl and peft libraries on a single NVIDIA T4 GPU, utilizing strictly native FP16 precision to ensure mathematical gradient stability.

• Training Framework: Pure Hugging Face (No Unsloth/Quantization)
• Method: LoRA (Low-Rank Adaptation)
• Rank (r): 32
• Alpha: 32
• Optimizer: adamw_torch
• Learning Rate: 2e-4
• Max Sequence Length: 2048

How to Use

Because this model was trained on a specific dataset structure, you must wrap your prompts in the ### Instruction: and ### Response: format for it to output the correct mathematical explanations.

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer

model_id = "ZentithLLM/NexusLLM-Math-1B-v1"

tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    torch_dtype=torch.float16,
    device_map="auto" 
)

question = "What is the general term in the expansion of (x+y)^n?"
formatted_prompt = f"### Instruction:\\n{question}\\n\\n### Response:\\n"

inputs = tokenizer(formatted_prompt, return_tensors="pt").to(model.device)

outputs = model.generate(
    **inputs,
    max_new_tokens=250,
    temperature=0.3,
    do_sample=True,
    pad_token_id=tokenizer.eos_token_id
)

print(tokenizer.decode(outputs[0], skip_special_tokens=True))