metadata

title: MGZon Chatbot
emoji: 🤖
colorFrom: blue
colorTo: green
sdk: docker
app_file: main.py
pinned: false

MGZON-AI

A versatile chatbot powered by MGZON/Veltrix for MGZon queries. Supports code generation, analysis, review, web search, and MGZon-specific queries. Licensed under Apache 2.0.

library_name: transformers license: apache-2.0 🌐 Live Demo Live Demo base_model: MGZON/Veltrix tags: - generated_from_trainer model-index: - name: mgzon-flan-t5-base results: []

MGZON/Veltrix

This model is a fine-tuned version of MGZON/Veltrix on the None dataset. It achieves the following results on the evaluation set:

Loss: nan

Features

Text Queries: Ask anything and get detailed responses.
Audio Input/Output: Record audio directly or convert text to speech.
Image Analysis: Capture images from webcam or upload for analysis.
Web Search: Enable DeepSearch for real-time web context.
API Support: Use endpoints like /api/chat, /api/audio-transcription, /api/text-to-speech, /api/image-analysis.

Setup

Add HF_TOKEN and BACKUP_HF_TOKEN as Secrets in Space settings.
Add GOOGLE_API_KEY and GOOGLE_CSE_ID for web search (optional).
Set PORT=7860, QUEUE_SIZE=80, CONCURRENCY_LIMIT=20 as Variables.
Ensure requirements.txt and Dockerfile are configured correctly.

Usage

Access the app at /gradio or use API endpoints. Examples:

Text: "Explain AI history."
Audio: Record audio for transcription.
Image: Capture or upload an image for analysis.

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

learning_rate: 3e-05
train_batch_size: 1
eval_batch_size: 1
seed: 42
gradient_accumulation_steps: 2
total_train_batch_size: 2
optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
lr_scheduler_type: linear
num_epochs: 5
mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss
0.2456	1.0	1488	nan
0.0888	2.0	2976	nan
15.9533	3.0	4464	nan
0.1136	4.0	5952	nan
0.0626	5.0	7440	nan

Framework versions

Transformers 4.55.2
Pytorch 2.8.0+cu126
Datasets 4.0.0
Tokenizers 0.21.4