Update README.md

README.md

@@ -33,7 +33,7 @@ base_model: meta-llama/Meta-Llama-3.1-70B-Instruct
 - **Model Developers:** Neural Magic
 
 Quantized version of [Meta-Llama-3.1-70B-Instruct](https://huggingface.co/meta-llama/Meta-Llama-3.1-70B-Instruct).
-It achieves an average score of 83.14 on the [OpenLLM](https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard) benchmark (version 1), whereas the unquantized model achieves 83.61.
+It achieves an average score of 84.29 on the [OpenLLM](https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard) benchmark (version 1), whereas the unquantized model achieves 84.40.
 
 ### Model Optimizations
 
@@ -118,11 +118,11 @@ model_stub = "meta-llama/Meta-Llama-3.1-70B-Instruct"
 model_name = model_stub.split("/")[-1]
 
 device_map = calculate_offload_device_map(
-    model_stub, reserve_for_hessians=False, num_gpus=2, torch_dtype=torch.float16
+    model_stub, reserve_for_hessians=False, num_gpus=2, 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
 )
 tokenizer = AutoTokenizer.from_pretrained(model_stub)
 
@@ -172,7 +172,7 @@ 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
 
@@ -191,71 +191,81 @@ This version of the lm-evaluation-harness includes versions of ARC-Challenge and
   <tr>
    <td>MMLU (5-shot)
    </td>
-   <td>82.21
+   <td>83.83
    </td>
-   <td>82.24
+   <td>83.73
    </td>
-   <td>100.0%
+   <td>99.88%
+   </td>
+  </tr>
+  <tr>
+   <td>MMLU-cot (0-shot)
+   </td>
+   <td>86.01
+   </td>
+   <td>85.44
+   </td>
+   <td>99.34%
    </td>
   </tr>
   <tr>
    <td>ARC Challenge (0-shot)
    </td>
-   <td>95.05
+   <td>93.26
    </td>
-   <td>94.54
+   <td>92.92
    </td>
-   <td>99.46%
+   <td>99.64%
    </td>
   </tr>
   <tr>
    <td>GSM-8K-cot (8-shot, strict-match)
    </td>
-   <td>93.18
+   <td>94.92
    </td>
-   <td>93.33
+   <td>94.54
    </td>
-   <td>100.1%
+   <td>99.60%
    </td>
   </tr>
   <tr>
    <td>Hellaswag (10-shot)
    </td>
-   <td>86.33
+   <td>86.75
    </td>
-   <td>85.67
+   <td>86.64
    </td>
-   <td>99.24%
+   <td>99.87%
    </td>
   </tr>
   <tr>
    <td>Winogrande (5-shot)
    </td>
-   <td>85.00
+   <td>85.32
    </td>
-   <td>85.79
+   <td>85.95
    </td>
-   <td>100.9%
+   <td>100.7%
    </td>
   </tr>
   <tr>
-   <td>TruthfulQA (0-shot)
+   <td>TruthfulQA (0-shot, mc2)
    </td>
-   <td>59.90
+   <td>60.68
    </td>
-   <td>57.24
+   <td>60.84
    </td>
-   <td>95.56%
+   <td>100.2%
    </td>
   </tr>
   <tr>
    <td><strong>Average</strong>
    </td>
-   <td><strong>83.61</strong>
+   <td><strong>84.40</strong>
    </td>
-   <td><strong>83.14</strong>
+   <td><strong>84.29</strong>
    </td>
-   <td><strong>99.43%</strong>
+   <td><strong>99.88%</strong>
    </td>
   </tr>
 </table>
@@ -274,6 +284,17 @@ lm_eval \
   --batch_size auto
 ```
 
+#### MMLU-cot
+```
+lm_eval \
+  --model vllm \
+  --model_args pretrained="neuralmagic/Meta-Llama-3.1-70B-Instruct-FP8",dtype=auto,add_bos_token=True,max_model_len=4096,tensor_parallel_size=2 \
+  --tasks mmlu_cot_0shot_llama_3.1_instruct \
+  --apply_chat_template \
+  --num_fewshot 0 \
+  --batch_size auto
+```
+
 #### ARC-Challenge
 ```
 lm_eval \