Apppublic

deepsodha/axionx-demo

sourceHugging Facemitupdated 6mo agoView on Hugging Face

0likes

App README

AxionX Digital — Model Training & Evaluation Suite

A live platform for training, evaluating, and comparing domain-specific language models and coding agents — running on Hugging Face Spaces.

Demos

Module	Description	Tech
Coding Agent Evaluator	Compare outputs from GPT-4o, Claude, Gemini, DeepSeek across coding tasks. Radar charts, leaderboards, side-by-side diffs.	Streamlit, Plotly
FinanceGPT	Fine-tuned FLAN-T5 for financial Q&A and summarization	LoRA, HF Transformers
LegalDoc Summarizer	Clause-level summarization using the CUAD dataset	FLAN-T5, PEFT
RetailGPT Evaluator	Benchmarking retail-QA models + interactive leaderboard	Evaluation, Plotly
Model Comparison	Side-by-side inference comparison for any two models	Streamlit

Live App

https://huggingface.co/spaces/deepsodha/axionx-demo

Use the sidebar to navigate between demos.

Project Structure

axionx-demo/
├── pages/                     → New pages (home, coding agent evaluator, model comparison)
├── financegpt/                → Finance fine-tuning demo
├── legaldoc_summarizer/       → Legal text summarization demo
├── retailgpt_evaluator/       → Retail chatbot evaluator with Plotly leaderboard
├── shared/                    → Common helpers (metrics, model loading, config)
├── datasets/                  → Sample JSONL datasets
├── streamlit_hub.py           → Unified Streamlit dashboard (entry point)
└── Dockerfile                 → Container deployment

Tech Stack

—Training: Transformers, PEFT (LoRA), bitsandbytes, Accelerate
—Evaluation: ROUGE, BLEU, Factuality, Code Quality metrics
—Dashboard: Streamlit, Plotly, Pandas
—Deployment: Hugging Face Spaces, Docker

About AxionX Digital

We fine-tune, evaluate, and benchmark domain-specific LLMs for enterprises. Our platform helps teams train models and compare agent outputs at scale across Finance, Legal, and Retail verticals.

License

Released under the MIT License.