tests/test_shape_issue.py

import torch
from components.tiny_confessional_layer import TinyConfessionalLayer

def diagnose_shape_issue():
    """Diagnose the shape mismatch issue in TinyConfessionalLayer"""
    print("🔍 Diagnosing shape issue...")
    
    # Test with the exact parameters from the error
    d_model = 512
    model = TinyConfessionalLayer(d_model=d_model)
    
    # Create input with proper dimensions (batch=1, seq=10, d_model=512)
    x = torch.randn(1, 10, d_model)
    
    print(f"Input shape: {x.shape}")
    print(f"Think net first layer: {model.think_net[0].weight.shape}")
    print(f"Act net first layer: {model.act_net[0].weight.shape}")
    
    # Test forward pass step by step
    y_state = torch.zeros_like(x)
    z_state = torch.zeros_like(x)
    
    # Think step
    think_input = torch.cat([x, y_state, z_state], dim=-1)
    print(f"\nThink input shape: {think_input.shape}")
    print(f"Expected: (1, 10, {d_model*3}) = (1, 10, {3*d_model})")
    
    try:
        z_state = model.think_net(think_input)
        print("✅ Think step passed")
    except Exception as e:
        print(f"❌ Think step failed: {e}")
    
    # Act step  
    act_input = torch.cat([y_state, z_state], dim=-1)
    print(f"\nAct input shape: {act_input.shape}")
    print(f"Expected: (1, 10, {d_model*2}) = (1, 10, {2*d_model})")
    
    try:
        y_state = model.act_net(act_input)
        print("✅ Act step passed")
    except Exception as e:
        print(f"❌ Act step failed: {e}")

if __name__ == "__main__":
    diagnose_shape_issue()