Update README.md

README.md

@@ -32,7 +32,7 @@ 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 0.3% of the scores of the unquantized model for MMLU, ARC-Challenge, GSM-8k, Hellaswag, Winogrande and TruthfulQA.
 
 ### Model Optimizations
 
@@ -145,6 +145,8 @@ 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
@@ -162,11 +164,11 @@ 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>87.47
+   <td>87.59
    </td>
-   <td>100.1%
+   <td>100.2%
    </td>
   </tr>
   <tr>
@@ -174,9 +176,9 @@ This version of the lm-evaluation-harness includes versions of MMLU, ARC-Challen
    </td>
    <td>88.26
    </td>
-   <td>88.23
+   <td>88.19
    </td>
-   <td>100.0%
+   <td>99.9%
    </td>
   </tr>
   <tr>
@@ -184,9 +186,9 @@ This version of the lm-evaluation-harness includes versions of MMLU, ARC-Challen
    </td>
    <td>94.97
    </td>
-   <td>94.88
+   <td>94.80
    </td>
-   <td>99.9%
+   <td>99.8%
    </td>
   </tr>
   <tr>
@@ -196,7 +198,7 @@ This version of the lm-evaluation-harness includes versions of MMLU, ARC-Challen
    </td>
    <td>96.13
    </td>
-   <td>99.7%
+   <td>100.8%
    </td>
   </tr>  
   <tr>
@@ -204,7 +206,7 @@ This version of the lm-evaluation-harness includes versions of MMLU, ARC-Challen
    </td>
    <td>88.33
    </td>
-   <td>88.50
+   <td>88.52
    </td>
    <td>100.2%
    </td>
@@ -214,9 +216,9 @@ This version of the lm-evaluation-harness includes versions of MMLU, ARC-Challen
    </td>
    <td>87.21
    </td>
-   <td>87.61
+   <td>87.92
    </td>
-   <td>100.5%
+   <td>100.8%
    </td>
   </tr>
   <tr>
@@ -224,7 +226,7 @@ This version of the lm-evaluation-harness includes versions of MMLU, ARC-Challen
    </td>
    <td>64.64
    </td>
-   <td>65.42
+   <td>65.41
    </td>
    <td>101.2%
    </td>
@@ -234,7 +236,7 @@ This version of the lm-evaluation-harness includes versions of MMLU, ARC-Challen
    </td>
    <td><strong>86.75</strong>
    </td>
-   <td><strong>86.89</strong>
+   <td><strong>86.94</strong>
    </td>
    <td><strong>100.2%</strong>
    </td>
@@ -249,7 +251,7 @@ The results were obtained using the following commands:
 ```
 lm_eval \
   --model vllm \
-  --model_args pretrained="neuralmagic/Meta-Llama-3.1-405B-Instruct-quantized.w8a16",dtype=auto,add_bos_token=True,max_model_len=3850,max_gen_toks=10,enable_chunked_prefill=True,tensor_parallel_size=8 \
+  --model_args pretrained="neuralmagic/Meta-Llama-3.1-405B-Instruct-quantized.w8a16",dtype=auto,max_model_len=3850,max_gen_toks=10,enable_chunked_prefill=True,tensor_parallel_size=8 \
   --tasks mmlu_llama_3.1_instruct \
   --fewshot_as_multiturn \
   --apply_chat_template \
@@ -261,7 +263,7 @@ lm_eval \
 ```
 lm_eval \
   --model vllm \
-  --model_args pretrained="neuralmagic/Meta-Llama-3.1-405B-Instruct-quantized.w8a16",dtype=auto,add_bos_token=True,max_model_len=4064,max_gen_toks=1024,enable_chunked_prefill=True,tensor_parallel_size=8 \
+  --model_args pretrained="neuralmagic/Meta-Llama-3.1-405B-Instruct-quantized.w8a16",dtype=auto,max_model_len=4064,max_gen_toks=1024,enable_chunked_prefill=True,tensor_parallel_size=8 \
   --tasks mmlu_cot_0shot_llama_3.1_instruct \
   --apply_chat_template \
   --num_fewshot 0 \
@@ -272,7 +274,7 @@ lm_eval \
 ```
 lm_eval \
   --model vllm \
-  --model_args pretrained="neuralmagic/Meta-Llama-3.1-405B-Instruct-quantized.w8a16",dtype=auto,add_bos_token=True,max_model_len=3940,max_gen_toks=100,enable_chunked_prefill=True,tensor_parallel_size=8 \
+  --model_args pretrained="neuralmagic/Meta-Llama-3.1-405B-Instruct-quantized.w8a16",dtype=auto,max_model_len=3940,max_gen_toks=100,enable_chunked_prefill=True,tensor_parallel_size=8 \
   --tasks arc_challenge_llama_3.1_instruct \
   --apply_chat_template \
   --num_fewshot 0 \
@@ -283,7 +285,7 @@ lm_eval \
 ```
 lm_eval \
   --model vllm \
-  --model_args pretrained="neuralmagic/Meta-Llama-3.1-405B-Instruct-quantized.w8a16",dtype=auto,add_bos_token=True,max_model_len=4096,max_gen_toks=1024,enable_chunked_prefill=True,tensor_parallel_size=8 \
+  --model_args pretrained="neuralmagic/Meta-Llama-3.1-405B-Instruct-quantized.w8a16",dtype=auto,max_model_len=4096,max_gen_toks=1024,enable_chunked_prefill=True,tensor_parallel_size=8 \
   --tasks gsm8k_cot_llama_3.1_instruct \
   --fewshot_as_multiturn \
   --apply_chat_template \