Update README.md (#2)

- Update README.md (116a0a1da04f0ffe90ff9c7b4f8a3c6f7cc9e812)


Co-authored-by: haoyanli <[email protected]>

README.md

@@ -1,5 +1,89 @@
----
-license: other
-license_name: deepseek
-license_link: https://github.com/deepseek-ai/DeepSeek-MoE/blob/main/LICENSE-MODEL
----
+---
+license: other
+license_name: deepseek
+license_link: https://github.com/deepseek-ai/DeepSeek-MoE/blob/main/LICENSE-MODEL
+---
+
+
+
+
+
+# deepseek-moe-16b-chat-FP8-KV 
+- ## Introduction
+  This model was created by applying [Quark](https://quark.docs.amd.com/latest/index.html) with calibration samples from Pile dataset.
+- ## Quantization Stragegy
+  - ***Quantized Layers***: All linear layers excluding "lm_head", "*gate"
+  - ***Weight***: FP8 symmetric per-tensor
+  - ***Activation***: FP8 symmetric per-tensor
+  - ***KV Cache***: FP8 symmetric  per-tensor
+- ## Quick Start
+1. [Download and install Quark](https://quark.docs.amd.com/latest/install.html)
+2. Run the quantization script in the example folder using the following command line:
+```sh
+export MODEL_DIR = [local model checkpoint folder] or deepseek-ai/deepseek-moe-16b-chat
+# single GPU
+python3 quantize_quark.py \
+        --model_dir $MODEL_DIR \
+        --output_dir deepseek-moe-16b-chat-FP8-KV \
+        --quant_scheme w_fp8_a_fp8 \
+        --kv_cache_dtype fp8 \
+        --num_calib_data 128  \
+        --model_export quark_safetensors
+# If model size is too large for single GPU, please use multi GPU instead.
+python3 quantize_quark.py \
+        --model_dir $MODEL_DIR \
+        --output_dir deepseek-moe-16b-chat-FP8-KV \
+        --quant_scheme w_fp8_a_fp8 \
+        --kv_cache_dtype fp8 \
+        --num_calib_data 128  \
+        --model_export quark_safetensors \
+        --multi_gpu
+```
+## Deployment
+Quark has its own export format and allows FP8 quantized models to be efficiently deployed using the vLLM backend(vLLM-compatible).
+## Evaluation
+Quark currently uses perplexity(PPL) as the evaluation metric for accuracy loss before and after quantization.The specific PPL algorithm can be referenced in the quantize_quark.py.
+The quantization evaluation results are conducted in pseudo-quantization mode, which may slightly differ from the actual quantized inference accuracy. These results are provided for reference only.
+#### Evaluation scores
+<table>
+  <tr>
+   <td><strong>Benchmark</strong>
+   </td>
+   <td><strong>deepseek-moe-16b-chat</strong>
+   </td>
+   <td><strong>deepseek-moe-16b-chat-FP8-KV(this model)</strong>
+   </td>
+  </tr>
+  <tr>
+   <td>Perplexity-wikitext2
+   </td>
+   <td>7.3568
+   </td>
+   <td>7.3929
+   </td>
+  </tr>
+</table>
+
+
+
+#### License
+Copyright (c) 2018-2024 Advanced Micro Devices, Inc. All Rights Reserved.
+
+
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+