Update README.md (#6)

- Update README.md (47346e4d832f26979f43a25b069688934c3f4cc8)


Co-authored-by: laixinn <yuanzu@users.noreply.huggingface.co>

README.md

@@ -9,15 +9,15 @@ The INT8 data type is both friendly and efficient for most hardware platforms.
 
 **We provide a block-wise INT8 weight for DeepSeek-R1.**
 
-In benchmarking, we observe **no accuracy loss** and up to **30\%** performance enhancement.
+In benchmarking, we observe **no accuracy loss** and up to **33\%** performance enhancement.
 
 [SGLang](https://github.com/sgl-project/sglang/tree/main) will soon support the block-wise INT8 quantization operation once our [PULL REQUEST](https://github.com/sgl-project/sglang/pull/3730) is merged.
 
 ## 1. Benchmarking Result (detailed in [PULL REQUEST](https://github.com/sgl-project/sglang/pull/3730)):
-| Model  | Config | Accuracy (GSM8K) | Accuracy (MMLU) | Output Throughput(qps=128) | Output Throughput(bs=1) |
-|--------|--------|-------------------|----------------|------------------------------|--------------------------|
-| BF16 R1 | A100\*32  | 95.5              | 87.1           | 3342.29                       | 37.20                     |
-| INT8 R1 | (A100\*16)x2 | **95.8**              | **87.1**           | 4450.02 **(+33%)**                | 44.18 **(+18%)**             |
+| Model  | Config | Accuracy (GSM8K) | Accuracy (MMLU) | Output Throughput(qps=128) |
+|--------|--------|-------------------|----------------|------------------------------|
+| BF16 R1 | A100\*32  | 95.5              | 87.1           | 3342.29                       |
+| INT8 R1 | (A100\*16)x2 | **95.8**              | **87.1**           | 4450.02 **(+33%)**                |
 
 ## 2. Quantization Process