PrithviAI9998/BiblioVecMistral7b

Mistral 7B – Original Research Title Recovery

Overview

This model is a fine-tuned version of Mistral 7B Instruct, developed for the task of original research paper title recovery.

The objective is to reconstruct the original title of a research paper from an LLM-generated reformulation of that title. The model explores a generative approach to title recovery, learning to map reformulated titles back to their corresponding original titles.

This work was developed in the context of the FIRE 2026 Find It, Fix It (FIFI) Shared Task.

Model Details

• Base model: Mistral 7B Instruct
• Architecture: Mistral 7B
• Fine-tuning framework: Unsloth
• Training approach: Supervised Fine-Tuning (SFT)
• Task: Original research paper title recovery
• Input: LLM-generated reformulated research title
• Output: Reconstructed original research title

Methodology

The model was fine-tuned using supervised instruction-based training.

Training examples consist of reformulated research paper titles paired with their corresponding original titles. The model learns to generate the original title while preserving the key technical concepts, terminology, and semantic information contained in the input.

The task explores whether a generative language model can recover original research titles without relying on an external retrieval corpus during inference.

Intended Use

• Original research paper title recovery
• Research title reconstruction
• Academic text generation
• Research-oriented text transformation
• Experiments on generative approaches to semantic title matching

Usage

Install the required library:

pip install transformers accelerate torch

Load the model:

from transformers import AutoTokenizer, AutoModelForCausalLM
import torch

model_id = "YOUR_USERNAME/YOUR_MODEL_NAME"

tokenizer = AutoTokenizer.from_pretrained(model_id)

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

Generate a prediction:

prompt = """Recover the original research paper title
from the following reformulated title:

Input: Deep learning techniques for detecting
anomalies in computer networks.

Original title:"""

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

outputs = model.generate(
    **inputs,
    max_new_tokens=100,
    do_sample=False
)

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

print(response)

Adapt the prompt to match the exact instruction format used during training.

Training Configuration

The model was fine-tuned using parameter-efficient training techniques.

• Training framework: Unsloth
• Fine-tuning method: Supervised Fine-Tuning
• Trainable parameters: To be specified
• Training dataset size: To be specified
• Number of epochs: To be specified
• Precision and quantization: To be specified

Limitations

• The model may generate plausible but factually incorrect research titles.
• Exact title reconstruction can be challenging when the reformulated input omits distinctive technical terminology.
• Generated outputs are not guaranteed to match the original publication title exactly.
• Performance may vary across research domains and title-generation styles.

Research Context

This model was developed as part of experiments for the FIRE 2026 Find It, Fix It (FIFI) Shared Task on Original Title Recovery.

Acknowledgements

We acknowledge the developers of Mistral, Unsloth, and the open-source machine learning ecosystem used in this work.

Citation

Please cite the associated research paper when referencing this model. Citation details will be added upon publication.