#!/usr/bin/env python3
"""Test to reproduce the dimension mismatch error"""
import os
import sys
from pathlib import Path
# Add src to path
sys.path.insert(0, str(Path(__file__).parent.parent / "src"))
def test_dimension_mismatch():
"""Try to reproduce the dimension mismatch error"""
print("=" * 80)
print("TESTING DIMENSION MISMATCH")
print("=" * 80)
# Import after adding to path
from utils.embeddings import get_embeddings_manager
from utils.specialized_embeddings import get_specialized_embedding_manager
# Get embeddings manager
embeddings_manager = get_embeddings_manager()
# Force eviction of CodeRankEmbed by loading other models
print("\n1. Loading models to force eviction...")
embeddings_manager.encode("test", content_type="general") # Load general (384D)
embeddings_manager.encode("test", content_type="config") # Load config (1024D)
if hasattr(embeddings_manager, 'manager'):
actual_manager = embeddings_manager.manager
print(f"Loaded models: {list(actual_manager.loaded_models.keys())}")
# Now directly check what happens when we encode with code content type
print("\n2. Testing encoding with 'code' content type...")
# Get the model that will be used
model_name = actual_manager.get_model_name('code')
print(f"Model name for 'code': {model_name}")
# Check if model is loaded
if model_name not in actual_manager.loaded_models:
print(f"WARNING: {model_name} is NOT loaded!")
print(f"Currently loaded: {list(actual_manager.loaded_models.keys())}")
# Force a situation where general model might be used instead
print("\n3. Simulating fallback scenario...")
# Let's check the fallback logic
print("\nChecking model configs:")
for content_type in ['code', 'general']:
config = actual_manager.model_configs.get(content_type, {})
print(f"{content_type}: {config.get('name')} (dim={config.get('dimension')})")
if 'fallback' in config:
print(f" Fallback: {config['fallback']}")
# Now let's see what dimension we get
print("\n4. Testing actual encoding...")
test_code = "def test(): pass"
# Encode with code content type
code_embedding = embeddings_manager.encode(test_code, content_type="code")
print(f"Code embedding shape: {code_embedding.shape}")
print(f"Code embedding dimension: {code_embedding.shape[-1]}")
# Check loaded models again
print(f"\nLoaded models after encoding: {list(actual_manager.loaded_models.keys())}")
# Let's also check what happens if we manually evict CodeRankEmbed
print("\n5. Manually testing eviction...")
if 'nomic-ai/CodeRankEmbed' in actual_manager.loaded_models:
# Force eviction by loading 2 other models (since max is 2)
embeddings_manager.encode("test", content_type="documentation")
embeddings_manager.encode("test", content_type="general")
print(f"After forced eviction: {list(actual_manager.loaded_models.keys())}")
# Now try encoding as code again
code_embedding2 = embeddings_manager.encode(test_code, content_type="code")
print(f"Code embedding after eviction: {code_embedding2.shape[-1]}D")
print(f"Models after re-encoding: {list(actual_manager.loaded_models.keys())}")
if __name__ == "__main__":
test_dimension_mismatch()
