Update README.md

README.md

@@ -24,7 +24,7 @@ language:
 - **License(s):** [llama3.1](https://huggingface.co/meta-llama/Meta-Llama-3.1-8B/blob/main/LICENSE)
 - **Model Developers:** Neural Magic
 
-Quantized version of [Meta-Llama-3.1-405B-Instruct](https://huggingface.co/meta-llama/Meta-Llama-3.1-405B-Instruct). It achieves an average recovery of 100.1% on the [OpenLLM](https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard) benchmark (version 1), compared to the unquantized model.
+Quantized version of [Meta-Llama-3.1-405B-Instruct](https://huggingface.co/meta-llama/Meta-Llama-3.1-405B-Instruct). It achieves an average recovery of 99.97% on the [OpenLLM](https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard) benchmark (version 1), compared to the unquantized model.
 <!-- It achieves an average score of 78.69 on the [OpenLLM](https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard) benchmark (version 1), whereas the unquantized model achieves 78.67. -->
 
 ### Model Optimizations
@@ -130,15 +130,9 @@ oneshot(
 
 ## Evaluation
 
-The model was evaluated on the [OpenLLM](https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard) leaderboard tasks (version 1) with the [lm-evaluation-harness](https://github.com/EleutherAI/lm-evaluation-harness) and the [vLLM](https://docs.vllm.ai/en/stable/) engine, using the following command:
-```
-lm_eval \
-  --model vllm \
-  --model_args pretrained="neuralmagic/Meta-Llama-3.1-405B-Instruct-FP8-dynamic",dtype=auto,tensor_parallel_size=8,gpu_memory_utilization=0.755,add_bos_token=True,max_model_len=4096 \
-  --tasks openllm \
-  --batch_size auto
-```
-Certain benchmarks for the full precision model are still being acquired. Average recovery is calculated only with metrics that both models have been evaluated on.
+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).
 
 ### Accuracy
 
@@ -157,41 +151,41 @@ Certain benchmarks for the full precision model are still being acquired. Averag
   <tr>
    <td>MMLU (5-shot)
    </td>
-   <td>*
+   <td>86.25
    </td>
    <td>86.17
    </td>
-   <td>*
+   <td>99.91%
    </td>
   </tr>
   <tr>
-   <td>ARC Challenge (25-shot)
+   <td>ARC Challenge (0-shot)
    </td>
-   <td>*
+   <td>96.93
    </td>
-   <td>*
+   <td>*being collected
    </td>
    <td>*
    </td>
   </tr>
   <tr>
-   <td>GSM-8K (5-shot, strict-match)
+   <td>GSM-8K-cot (8-shot, strict-match)
    </td>
-   <td>95.07
+   <td>96.44
    </td>
-   <td>95.00
+   <td>95.98
    </td>
-   <td>99.93%
+   <td>99.52%
    </td>
   </tr>
   <tr>
    <td>Hellaswag (10-shot)
    </td>
-   <td>*
+   <td>88.33
    </td>
    <td>88.34
    </td>
-   <td>*
+   <td>100.0%
    </td>
   </tr>
   <tr>
@@ -217,11 +211,11 @@ Certain benchmarks for the full precision model are still being acquired. Averag
   <tr>
    <td><strong>Average</strong>
    </td>
-   <td><strong>*</strong>
+   <td><strong>86.63</strong>
    </td>
    <td><strong>*</strong>
    </td>
-   <td><strong>100.1%</strong>
+   <td><strong>99.97%</strong>
    </td>
   </tr>
 </table>