🌟 SmolTransformer Zero - Translation App

A compact Encoder-Decoder Transformer for English to Hindi translation deployed with ZeroGPU.

Features:

  • ZeroGPU Integration: Dynamic GPU allocation with NVIDIA H200 slices
  • Advanced Generation: Top-K sampling and beam search methods
  • Real-time Translation: Fast inference with automatic scaling
  • Model Download: Automatically downloads pre-trained model from HuggingFace

💬 Translation
Generation Method
10 200
0.1 2
1 100
1 10

🎯 Translation Output

📊 Model Information
  • Model Size: ~25M parameters
  • Architecture: 6-layer Encoder-Decoder Transformer
  • Context Length: 128 tokens
  • Vocabulary: IndicBARTSS tokenizer
  • ZeroGPU: NVIDIA H200 slice (70GB VRAM)
  • Max Duration: 120 seconds per translation