Update README.md

README.md

@@ -33,7 +33,7 @@ base_model: meta-llama/Meta-Llama-3.1-8B-Instruct
 - **Model Developers:** Neural Magic
 
 Quantized version of [Meta-Llama-3.1-8B-Instruct](https://huggingface.co/meta-llama/Meta-Llama-3.1-8B-Instruct).
-It achieves an average score of 73.81 on the [OpenLLM](https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard) benchmark (version 1), whereas the unquantized model achieves 74.17.
+It achieves an average score of 73.56 on the [OpenLLM](https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard) benchmark (version 1), whereas the unquantized model achieves 73.79.
 
 ### Model Optimizations
 
@@ -117,11 +117,11 @@ model_stub = "meta-llama/Meta-Llama-3.1-8B-Instruct"
 model_name = model_stub.split("/")[-1]
 
 device_map = calculate_offload_device_map(
-    model_stub, reserve_for_hessians=False, num_gpus=1, torch_dtype=torch.float16
+    model_stub, reserve_for_hessians=False, num_gpus=1, torch_dtype="auto"
 )
 
 model = SparseAutoModelForCausalLM.from_pretrained(
-    model_stub, torch_dtype=torch.float16, device_map=device_map
+    model_stub, torch_dtype="auto", device_map=device_map
 )
 
 output_dir = f"./{model_name}-FP8-dynamic"
@@ -139,7 +139,7 @@ oneshot(
 
 The model was evaluated on MMLU, ARC-Challenge, GSM-8K, Hellaswag, Winogrande and TruthfulQA.
 Evaluation was conducted using the Neural Magic fork of [lm-evaluation-harness](https://github.com/neuralmagic/lm-evaluation-harness/tree/llama_3.1_instruct) (branch llama_3.1_instruct) and the [vLLM](https://docs.vllm.ai/en/stable/) engine.
-This version of the lm-evaluation-harness includes versions of ARC-Challenge and GSM-8K that match the prompting style of [Meta-Llama-3.1-Instruct-evals](https://huggingface.co/datasets/meta-llama/Meta-Llama-3.1-8B-Instruct-evals).
+This version of the lm-evaluation-harness includes versions of ARC-Challenge, GSM-8K, MMLU, and MMLU-cot that match the prompting style of [Meta-Llama-3.1-Instruct-evals](https://huggingface.co/datasets/meta-llama/Meta-Llama-3.1-8B-Instruct-evals).
 
 ### Accuracy
 
@@ -158,71 +158,81 @@ This version of the lm-evaluation-harness includes versions of ARC-Challenge and
   <tr>
    <td>MMLU (5-shot)
    </td>
-   <td>67.94
+   <td>67.95
    </td>
-   <td>68.09
+   <td>68.02
    </td>
-   <td>100.2%
+   <td>100.1%
+   </td>
+  </tr>
+  <tr>
+   <td>MMLU-cot (0-shot)
+   </td>
+   <td>71.24
+   </td>
+   <td>71.64
+   </td>
+   <td>100.5%
    </td>
   </tr>
   <tr>
    <td>ARC Challenge (0-shot)
    </td>
-   <td>83.11
+   <td>82.00
    </td>
-   <td>82.34
+   <td>81.23
    </td>
-   <td>99.07%
+   <td>99.06%
    </td>
   </tr>
   <tr>
-   <td>GSM-8K (CoT, 8-shot, strict-match)
+   <td>GSM-8K-cot (8-shot, strict-match)
    </td>
-   <td>82.03
+   <td>81.96
    </td>
-   <td>82.34
+   <td>82.03
    </td>
-   <td>100.3%
+   <td>100.0%
    </td>
   </tr>
   <tr>
    <td>Hellaswag (10-shot)
    </td>
-   <td>80.01
+   <td>80.46
    </td>
-   <td>79.68
+   <td>80.04
    </td>
-   <td>99.59%
+   <td>99.48%
    </td>
   </tr>
   <tr>
    <td>Winogrande (5-shot)
    </td>
-   <td>77.90
+   <td>78.45
    </td>
-   <td>77.03
+   <td>77.66
    </td>
-   <td>98.88%
+   <td>98.99%
    </td>
   </tr>
   <tr>
    <td>TruthfulQA (0-shot, mc2)
    </td>
-   <td>54.04
+   <td>54.5
    </td>
-   <td>53.37
+   <td>54.28
    </td>
-   <td>98.76%
+   <td>99.60%
    </td>
   </tr>
   <tr>
    <td><strong>Average</strong>
    </td>
-   <td><strong>74.17</strong>
+   <td><strong>73.79</strong>
    </td>
-   <td><strong>73.81</strong>
+   <td><strong>73.56</strong>
    </td>
-   <td><strong>99.48%</strong>
+   <td><strong>99.70%</strong>
    </td>
   </tr>
 </table>
@@ -241,6 +251,17 @@ lm_eval \
   --batch_size auto
 ```
 
+#### MMLU-cot
+```
+lm_eval \
+  --model vllm \
+  --model_args pretrained="neuralmagic/Meta-Llama-3.1-8B-Instruct-FP8-dynamic",dtype=auto,add_bos_token=True,max_model_len=4096,tensor_parallel_size=1 \
+  --tasks mmlu_cot_0shot_llama_3.1_instruct \
+  --apply_chat_template \
+  --num_fewshot 0 \
+  --batch_size auto
+```
+
 #### ARC-Challenge
 ```
 lm_eval \