---
datasets:
- BAAI/Infinity-Instruct
base_model:
- nvidia/Llama-3.1-Minitron-4B-Depth-Base
---

We fine-tune nvidia/Llama-3.1-Minitron-4B-Depth-Base with LLM-Neo method，which combines LoRA and KD in one. Training data is sampling from BAAI/Infinity-Instruct for 100k lines.



## Benchmarks 

In this section, we report the results for Llama-3.1-Minitron-4B-Depth-Neo-10w on standard automatic benchmarks. For all the evaluations, we use [lm-evaluation-harness](https://github.com/EleutherAI/lm-evaluation-harness) library. 

### Evaluation results

<table>
  <tr>
   <td><strong>Category</strong>
   </td>
   <td><strong>Benchmark</strong>
   </td>
   <td><strong>Version</strong>
   </td>
   <td><strong>n-shot</strong>
   </td>
   <td><strong>Metric</strong>
   </td>
   <td><strong>Value</strong>
   </td>
   <td><strong>Stderr</strong>
   </td>
  </tr>
  <tr>
   <td rowspan="3" >BBH
   </td>
   <td>BBH (General)</td>
   <td>N/A</td>
   <td>3</td>
   <td>exact_match</td>
   <td>0.4729</td>
   <td>± 0.0055</td>
  </tr>
  <tr>
   <td>BBH (Boolean Expressions)</td>
   <td>2</td>
   <td>3</td>
   <td>exact_match</td>
   <td>0.8120</td>
   <td>± 0.0248</td>
  </tr>
  <tr>
   <td>BBH (Date Understanding)</td>
   <td>2</td>
   <td>3</td>
   <td>exact_match</td>
   <td>0.6600</td>
   <td>± 0.0300</td>
  </tr>
  <tr>
   <td rowspan="4" >CEVAL
   </td>
   <td>CEVAL (General)</td>
   <td>N/A</td>
   <td>0</td>
   <td>acc</td>
   <td>0.4413</td>
   <td>± 0.0135</td>
  </tr>
  <tr>
   <td>CEVAL (Accountant)</td>
   <td>1</td>
   <td>0</td>
   <td>acc</td>
   <td>0.3469</td>
   <td>± 0.0687</td>
  </tr>
  <tr>
   <td>CEVAL (Advanced Mathematics)</td>
   <td>1</td>
   <td>0</td>
   <td>acc</td>
   <td>0.4737</td>
   <td>± 0.1177</td>
  </tr>
  <tr>
   <td>CEVAL (Art Studies)</td>
   <td>1</td>
   <td>0</td>
   <td>acc</td>
   <td>0.4545</td>
   <td>± 0.0880</td>
  </tr>
  <tr>
   <td rowspan="3" >MMLU
   </td>
   <td>MMLU (General)</td>
   <td>N/A</td>
   <td>0</td>
   <td>acc</td>
   <td>0.6048</td>
   <td>± 0.0039</td>
  </tr>
  <tr>
   <td>MMLU (Humanities)</td>
   <td>N/A</td>
   <td>0</td>
   <td>acc</td>
   <td>0.5552</td>
   <td>± 0.0067</td>
  </tr>
  <tr>
   <td>MMLU (STEM)</td>
   <td>N/A</td>
   <td>0</td>
   <td>acc</td>
   <td>0.5214</td>
   <td>± 0.0086</td>
  </tr>
  <tr>
   <td rowspan="2" >CMMLU
   </td>
   <td>CMMLU (General)</td>
   <td>N/A</td>
   <td>0</td>
   <td>acc</td>
   <td>0.3548</td>
   <td>± 0.0044</td>
  </tr>
  <tr>
   <td>CMMLU (Normalized)</td>
   <td>N/A</td>
   <td>0</td>
   <td>acc_norm</td>
   <td>0.3548</td>
   <td>± 0.0044</td>
  </tr>
</table>