🚀 Model Trainer (FREE GPU)

Fine-tune Qwen models on your documents using ZeroGPU - completely free!

Features:

🆓 100% FREE GPU via ZeroGPU (Pro account = priority access!)
🤖 Fine-tune Qwen 0.5B, 1.5B, 3B models
📊 LoRA efficient training (trains 1000x faster)
💾 Auto-save to Model Hub
🎯 No credit card required

Status: 🟢 FULLY OPERATIONAL - Ready to train your AI model!

📊 Training Configuration

Dataset Name

Your processed dataset from doc-processor

Base Model

Smaller = faster training

Output Model Name

Where to save your trained model

⚙️ Training Parameters

Training Epochs

More epochs = better learning (but slower)

1 10

Learning Rate

How fast the model learns

Batch Size

Reduce if out of memory

1 16

Max Sequence Length

Longer = more context (but more memory)

128 2048

🎯 LoRA Parameters

LoRA Rank (r)

Higher = more capacity (but slower)

4 64

LoRA Alpha

Usually 2x the rank

8 128

📊 Training Output

Training Log

📖 How to Use

Prepare Dataset - Use doc-processor to create your dataset first
Configure - Choose base model and training parameters
Start Training - Click "Start Training" and wait
Use Your Model - Open chat-demo and enter your model name

🎯 Recommended Settings

For Quick Testing:

Model: Qwen2.5-0.5B
Epochs: 1-2
Batch Size: 2
Max Length: 256

For Best Quality:

Model: Qwen2.5-1.5B
Epochs: 3-5
Batch Size: 4
Max Length: 512

💡 Tips

First time? Start with Qwen2.5-0.5B and 1 epoch
Out of memory? Reduce batch_size to 1
Pro account? You get longer GPU time (10+ minutes)
Training interrupted? Just run again, ZeroGPU will reallocate

⏱️ Expected Training Time

0.5B model, 100 examples: ~2-3 minutes
1.5B model, 100 examples: ~5-7 minutes
3B model, 100 examples: ~10-15 minutes

With ZeroGPU, it's 100% FREE! 🎉