Update README.md

README.md

@@ -33,7 +33,7 @@ base_model: meta-llama/Meta-Llama-3.1-405B-Instruct
 - **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 scores within 1% of the scores of the unquantized model for MMLU, ARC-Challenge, GSM-8k, Hellaswag, Winogrande and TruthfulQA.
+It achieves scores within 1.3% of the scores of the unquantized model for MMLU, ARC-Challenge, GSM-8k, Hellaswag, Winogrande and TruthfulQA.
 
 ### Model Optimizations
 
@@ -149,6 +149,8 @@ The model was evaluated on MMLU, ARC-Challenge, GSM-8K, Hellaswag, Winogrande an
 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 MMLU, 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-3.1-405B-Instruct-evals).
 
+**Note:** Results have been updated after Meta modified the chat template.
+
 ### Accuracy
 
 #### Open LLM Leaderboard evaluation scores
@@ -158,7 +160,7 @@ This version of the lm-evaluation-harness includes versions of MMLU, ARC-Challen
    </td>
    <td><strong>Meta-Llama-3.1-405B-Instruct </strong>
    </td>
-   <td><strong>Meta-Llama-3.1-405B-Instruct-quantized.w8a8 (this model)</strong>
+   <td><strong>Meta-Llama-3.1-405B-Instruct-quantized.w4a16 (this model)</strong>
    </td>
    <td><strong>Recovery</strong>
    </td>
@@ -166,31 +168,21 @@ This version of the lm-evaluation-harness includes versions of MMLU, ARC-Challen
   <tr>
    <td>MMLU (5-shot)
    </td>
-   <td>87.41
+   <td>87.38
    </td>
    <td>86.76
    </td>
    <td>99.3%
    </td>
-  </tr>  
-  <tr>
-   <td>MMLU (CoT, 0-shot)
-   </td>
-   <td>88.26
-   </td>
-   <td>87.42
-   </td>
-   <td>99.0%
-   </td>
   </tr>
   <tr>
    <td>ARC Challenge (0-shot)
    </td>
    <td>94.97
    </td>
-   <td>94.62
+   <td>94.37
    </td>
-   <td>99.6%
+   <td>99.4%
    </td>
   </tr>
   <tr>
@@ -198,19 +190,19 @@ This version of the lm-evaluation-harness includes versions of MMLU, ARC-Challen
    </td>
    <td>96.44
    </td>
-   <td>96.13
+   <td>95.45
    </td>
-   <td>99.7%
+   <td>99.0%
    </td>
   </tr>
   <tr>
    <td>Hellaswag (10-shot)
-   </td>
+   </td> 
    <td>88.33
    </td>
-   <td>88.08
+   <td>88.15
    </td>
-   <td>99.7%
+   <td>99.8%
    </td>
   </tr>
   <tr>
@@ -218,19 +210,19 @@ This version of the lm-evaluation-harness includes versions of MMLU, ARC-Challen
    </td>
    <td>87.21
    </td>
-   <td>86.42
+   <td>86.11
    </td>
-   <td>99.7%
+   <td>98.7%
    </td>
   </tr>
   <tr>
-   <td>TruthfulQA (0-shot, mc2)
+   <td>TruthfulQA (0-shot)
    </td>
    <td>64.64
    </td>
-   <td>64.44
+   <td>64.39
    </td>
-   <td>99.7%
+   <td>99.6%
    </td>
   </tr>
   <tr>
@@ -238,9 +230,9 @@ This version of the lm-evaluation-harness includes versions of MMLU, ARC-Challen
    </td>
    <td><strong>86.75</strong>
    </td>
-   <td><strong>86.27</strong>
+   <td><strong>86.11</strong>
    </td>
-   <td><strong>99.4%</strong>
+   <td><strong>99.3%</strong>
    </td>
   </tr>
 </table>
@@ -253,7 +245,7 @@ The results were obtained using the following commands:
 ```
 lm_eval \
   --model vllm \
-  --model_args pretrained="neuralmagic/Meta-Llama-3.1-405B-Instruct-quantized.w8a8",dtype=auto,add_bos_token=True,max_model_len=3850,max_gen_toks=10,tensor_parallel_size=8 \
+  --model_args pretrained="neuralmagic/Meta-Llama-3.1-405B-Instruct-quantized.w8a8",dtype=auto,max_model_len=3850,max_gen_toks=10,tensor_parallel_size=8 \
   --tasks mmlu_llama_3.1_instruct \
   --fewshot_as_multiturn \
   --apply_chat_template \
@@ -265,7 +257,7 @@ lm_eval \
 ```
 lm_eval \
   --model vllm \
-  --model_args pretrained="neuralmagic/Meta-Llama-3.1-405B-Instruct-quantized.w8a8",dtype=auto,add_bos_token=True,max_model_len=4064,max_gen_toks=1024,tensor_parallel_size=8 \
+  --model_args pretrained="neuralmagic/Meta-Llama-3.1-405B-Instruct-quantized.w8a8",dtype=auto,max_model_len=4064,max_gen_toks=1024,tensor_parallel_size=8 \
   --tasks mmlu_cot_0shot_llama_3.1_instruct \
   --apply_chat_template \
   --num_fewshot 0 \
@@ -276,7 +268,7 @@ lm_eval \
 ```
 lm_eval \
   --model vllm \
-  --model_args pretrained="neuralmagic/Meta-Llama-3.1-405B-Instruct-quantized.w8a8",dtype=auto,add_bos_token=True,max_model_len=3940,max_gen_toks=100,tensor_parallel_size=8 \
+  --model_args pretrained="neuralmagic/Meta-Llama-3.1-405B-Instruct-quantized.w8a8",dtype=auto,max_model_len=3940,max_gen_toks=100,tensor_parallel_size=8 \
   --tasks arc_challenge_llama_3.1_instruct \
   --apply_chat_template \
   --num_fewshot 0 \
@@ -287,7 +279,7 @@ lm_eval \
 ```
 lm_eval \
   --model vllm \
-  --model_args pretrained="neuralmagic/Meta-Llama-3.1-405B-Instruct-quantized.w8a8",dtype=auto,add_bos_token=True,max_model_len=4096,max_gen_toks=1024,tensor_parallel_size=8 \
+  --model_args pretrained="neuralmagic/Meta-Llama-3.1-405B-Instruct-quantized.w8a8",dtype=auto,max_model_len=4096,max_gen_toks=1024,tensor_parallel_size=8 \
   --tasks gsm8k_cot_llama_3.1_instruct \
   --fewshot_as_multiturn \
   --apply_chat_template \