Add evaluation metrics for bonus2-multitask

README.md

@@ -1,30 +1,13 @@
-# Bonus 2: Multi-Task MoE (Summarization + Classification)
 
-## Model
-
-Multi-task transformer with shared MoE layers for:
-1. **Text Summarization** (XSum)
-2. **Document Classification** (length categories)
-
-## Architecture
-
-- Shared MoE encoder (4 layers)
-- Shared MoE decoder (4 layers)
-- 6 experts per MoE layer
-- Task-specific heads
-
-## Performance
-
-### Classification
-- Validation Accuracy: 0.3420
-
-### Summarization
-- ROUGE-1: 0.2250
-- ROUGE-2: 0.0333
-- ROUGE-L: 0.2078
-
-## Benefits
-
-1. **Parameter Efficiency**: Shared experts reduce total parameters
-2. **Knowledge Transfer**: Tasks benefit from shared representations
-3. **Better Generalization**: Multi-task learning improves robustness
+# Bonus 2: Multitask MoE (XSum)
+
+## Metrics
+- ROUGE-1: 0.0000  
+- ROUGE-2: 0.0000  
+- ROUGE-L: 0.0000  
+- ROUGE-Lsum: 0.0000  
+- SacreBLEU: 0.0000  
+- BERTScore (P/R/F1): 0.7473 / 0.8181 / 0.7809  
+- Compression ratio: 0.2472  
+- Extractiveness: 0.6044  
+- NLI factual consistency: 0.4948