Update README.md (#1)
Browse files- Update README.md (9692d6350cc82096d838fa93a537d1f512fcee33)
Co-authored-by: laixinn <[email protected]>
README.md
CHANGED
|
@@ -46,6 +46,10 @@ library_name: transformers
|
|
| 46 |
<a href="https://github.com/deepseek-ai/DeepSeek-R1/blob/main/DeepSeek_R1.pdf"><b>Paper Link</b>👁️</a>
|
| 47 |
</p>
|
| 48 |
|
|
|
|
|
|
|
|
|
|
|
|
|
| 49 |
|
| 50 |
## 1. Introduction
|
| 51 |
|
|
|
|
| 46 |
<a href="https://github.com/deepseek-ai/DeepSeek-R1/blob/main/DeepSeek_R1.pdf"><b>Paper Link</b>👁️</a>
|
| 47 |
</p>
|
| 48 |
|
| 49 |
+
## 0. INT8 Quantization
|
| 50 |
+
|
| 51 |
+
We apply a INT8 quantization on the BF16 checkpoints, where weight scales are determined by dividing he block-wise maximum of element values by the INT8 type maximum.
|
| 52 |
+
The quantization script is provided in inference/bf16_case_int8.py.
|
| 53 |
|
| 54 |
## 1. Introduction
|
| 55 |
|