S4m2357/Sci-pi

Model Card for Model ID

Model Details

Model Description

• Developed by: Muhammad Samuel Qudus
• Model type: Text Generation
• Language(s) (NLP): English
• License: apache-2.0
• Finetuned from model : microsoft/phi-4

Sci-π leverages Phi-4’s reasoning capabilities, enhanced with retrieval using FAISS and embedding-based filtering via allenai/specter. This enables it to outperform baselines in factual consistency and semantic alignment when generating summaries for scientific content.

Model Sources [optional]

• Repository: Sci-pi
• Paper : Ongoing
• Demo : Coming Soon

Uses

Direct Use

Generate scientific summaries for academic papers

Answer domain-specific questions in science and engineering

Create abstractive highlights for scientific abstracts

Downstream Use [optional]

Plug into research assistants or knowledge base generators

Integrate into academic Q&A tools or automated tutoring systems

[More Information Needed]

Out-of-Scope Use

Non-English content

Informal or casual language generation

Legal, medical, or sensitive policy decision-making without human supervision

Bias, Risks, and Limitations

it May hallucinate facts if retrieval fails or context is insufficient

For English-only; performance may degrade in multilingual settings

Sci-π only Trained on scientific content; not intended for general-purpose chatbot use

Recommendations

Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.

How to Get Started with the Model

Use the code below to get started with the model.

[More Information Needed]

Training Details

Training Data

Domain-specific scientific abstracts from arXiv (math, cs, physics)

Filtered using keywords and metadata to ensure relevance and quality

Training Procedure

Preprocessing: Tokenized using phi-4 tokenizer; context window capped at 2048 tokens

Mixed Precision: fp32

Hardware: Google Colab L4 GPU (22 hours)

Preprocessing [optional]

[More Information Needed]

Training Hyperparameters

• Training regime: FP32

Speeds, Sizes, Times [optional]

[More Information Needed]

Evaluation

Testing Data, Factors & Metrics

Testing Data

[More Information Needed]

Factors

[More Information Needed]

Metrics

ROUGE-1 F1

ROUGE-2 F1

ROUGE-L F1

BERTScore F1

Sentence-BERT Cosine Similarity

Precision@3

Recall@3

mAP@3

NDCG@3

Results

Model Performance

🔍 Retrieval Performance

Metric	Value
Precision@3	1.0000 ✅
Recall@3	100.00% ✅

🧠 Generation Performance

Metric	Score
ROUGE-1 F1	0.5452
ROUGE-2 F1	0.2121
ROUGE-L F1	0.2207
BERTScore F1	0.7795
Sentence-BERT Sim	0.8104

Summary

These scores suggest high semantic and factual alignment between generated summaries and reference academic abstracts.

Model Examination [optional]

[More Information Needed]

Environmental Impact

Carbon emissions can be estimated using the Machine Learning Impact calculator presented in Lacoste et al. (2019).

• Hardware Type: L4 GPU
• Hours used: 22 Hours
• Cloud Provider: Google Colab
• Compute Region: Unknown
• Carbon Emitted: Unknown

Technical Specifications

Architecture: Phi-4 (decoder-only transformer)

Retriever: FAISS with allenai/specter

Generation Mode: RAG-style pipeline

Model Architecture and Objective

[More Information Needed]

Compute Infrastructure

[More Information Needed]

Hardware

[More Information Needed]

Software

[More Information Needed]

Citation [optional]

BibTeX:

[More Information Needed]

APA:

[More Information Needed]

Glossary [optional]

[More Information Needed]

More Information

If you have questions, ideas, or want to collaborate:

Hugging Face Profile: @S4m2357

Model Card Authors [optional]

[More Information Needed]

Model Card Contact

[More Information Needed]