IFEval Results - Initial

README.md

@@ -18,6 +18,22 @@ This model was made possible by excellent AMD mi300x compute generously provided
 
 **⚠️ Note: This is an intermediate checkpoint not intended for direct use. For the complete model, use Qwen3-72B-Embiggened.**
 
+## Evaluation Results
+
+The model was evaluated on the **IFEval** (Instruction Following Evaluation) benchmark to test its ability to follow explicit instructions. Despite being an intermediate, un-trained model, it shows promising instruction following capabilities.
+
+Evaluation was performed using the [lm-evaluation-harness](https://github.com/EleutherAI/lm-evaluation-harness) in a 0-shot setting.
+
+| Metric | Accuracy |
+| :--- | :---: |
+| **Prompt-level Strict Accuracy** | **68.75%** |
+| **Instruction-level Strict Accuracy** | **75.00%** |
+| Prompt-level Loose Accuracy | 68.75% |
+| Instruction-level Loose Accuracy | 75.00% |
+
+- **Strict Accuracy**: All instructions in a prompt must be followed perfectly.
+- **Loose Accuracy**: Allows for minor deviations (e.g., formatting, verbosity) as long as the core instruction is met.
+
 ## Architecture Changes
 
 ### Original Qwen3-32B