Update README.md

README.md

@@ -49,6 +49,13 @@ We evaluated both the **base BGE-M3** and the **fine-tuned AVI-M3** on a held-ou
 
 ---
 
+### 📚 Dataset
+- **Training set:** 1088 examples
+- **Evaluation set:** 273 examples (~20% held-out split)
+- **Task:** Query → Positive passage retrieval
+
+---
+
 ## 📊 Metrics Explained
 
 - **Accuracy:** proportion of queries where the top-1 retrieved document is correct.
@@ -57,22 +64,30 @@ We evaluated both the **base BGE-M3** and the **fine-tuned AVI-M3** on a held-ou
 
 ---
 
-## 🛠️ Training
+### 💻 Hardware
 
-- **Training dataset:** Custom domain-specific dataset (1088 train, 273 eval)
-- **Evaluation dataset:** 273 examples (~20 % held-out split)
-- **Hardware:** 1× NVIDIA A40 48 GB
-- **Batch size:** (depends on your config)
-- **Optimizer:** AdamW
-- **Loss:** Contrastive / InfoNCE
-- **Framework:** FlagEmbedding
+- **GPU:** 1× NVIDIA A40 (48 GB VRAM)
+- **Precision:** FP16 with gradient checkpointing
+- **Effective batch size:** 32 (8 × grad accumulation 4)
 
 ---
 
-## 💡 Usage Example (Python)
+## 🛠️ Training
 
-```python
-from FlagEmbedding import FlagModel
 
-model = FlagModel("shaipro/avi-m3", query_max_len=128, doc_max_len=512, use_fp16=True)
-embeddings = model.encode(["your query"], normalize_embeddings=True)
+- **Evaluation dataset:** 273 examples (~20 % held-out split)
+- **Epochs:** 10  
+- **Learning rate:** 2e-5  
+- **Per-device batch size:** 8  
+- **Gradient accumulation:** 4  
+- **Pooling method:** `cls`  
+- **Temperature:** 0.02  
+- **Loss:** `m3_kd_loss` (knowledge distillation + contrastive)  
+- **Knowledge distillation:** Enabled
+- **Self-distillation:** Enabled  
+- **Unified fine-tuning:** Enabled
+- **Encoder freezing:** Disabled
+- **Optimizer:** AdamW  
+- **Scheduler:** Linear with 10% warmup  
+
+---