qwen2-webshop-federated

This is a federated learning model based on Qwen2-1.5B-Instruct, trained using GRPO (Generalized Reward Policy Optimization) on the WebShop dataset.

Model Details

• Base Model: Qwen2-1.5B-Instruct
• Training Method: Federated Learning with GRPO
• Dataset: WebShop
• Architecture: Qwen2ForCausalLM
• Parameters: ~1.5B
• Hidden Size: 1536
• Layers: 28
• Attention Heads: 12

Usage

from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("qwen2-webshop-federated")
model = AutoModelForCausalLM.from_pretrained("qwen2-webshop-federated")

# Example usage
inputs = tokenizer("Hello, how are you?", return_tensors="pt")
outputs = model.generate(**inputs, max_length=100)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

Training Details

This model was trained using federated learning with the following configuration:

• Total clients: 100
• Clients per round: 2
• Rounds: 70
• Epochs per client: 3
• Minimum goals per client: 100

The model was aggregated after round 70, global step 0.