muxi_fix

FlagOpen · Jan 22, 2024 · 4be09ee · 4be09ee
1 parent 13699a0
commit 4be09ee
Show file tree

Hide file tree

Showing 3 changed files with 6 additions and 4 deletions.
diff --git a/README.md b/README.md
@@ -442,7 +442,7 @@
    </tr>
     <tr height="16.80" style='height:16.80pt;'>
     <td class="xl65" x:str>8</td>
-        <td class="xl65" height="33.60" style='height:33.60pt;border-right:none;border-bottom:none;' x:str><a href="https://github.com/FlagOpen/FlagPerf/tree/main/inference/benchmarks/aquila_7b_mmlu" style="text-decoration:none" target="_parent">Aquila-7B-mmlu</td>
+        <td class="xl65" height="33.60" style='height:33.60pt;border-right:none;border-bottom:none;' x:str><a href="https://github.com/FlagOpen/FlagPerf/tree/main/inference/benchmarks/Aquila_7b_mmlu" style="text-decoration:none" target="_parent">Aquila-7B-mmlu</td>
         <td class="xl69" x:str>NLP</td>
     <td class="xl69" x:str>fp16</td>
     <td class="xl69" x:str>N/A</td>
@@ -451,7 +451,7 @@
     </tr>
 <tr height="16.80" style='height:16.80pt;'>
    <td class="xl65" x:str>9</td>
-    <td class="xl65" height="33.60" style='height:33.60pt;border-right:none;border-bottom:none;' x:str><a href="https://github.com/FlagOpen/FlagPerf/tree/main/inference/benchmarks/sam" style="text-decoration:none" target="_parent">SegmentAnything</td>
+    <td class="xl65" height="33.60" style='height:33.60pt;border-right:none;border-bottom:none;' x:str><a href="https://github.com/FlagOpen/FlagPerf/tree/main/inference/benchmarks/sam_h" style="text-decoration:none" target="_parent">SegmentAnything</td>
         <td class="xl69" x:str>MultiModal</td>
     <td class="xl69" x:str>fp16</td>
     <td class="xl69" x:str>W32A16</td>

diff --git a/inference/benchmarks/bertLarge/README.md b/inference/benchmarks/bertLarge/README.md
@@ -15,9 +15,9 @@ bert_reference_results_text_md5.txt
 * 模型实现
   * pytorch：transformers.BertForMaskedLM
 * 权重下载
-  * pytorch：BertForMaskedLM.from_pretrained("bert-large/base-uncased")
+  * pytorch：BertForMaskedLM.from_pretrained("bert-large-uncased")
 * 权重选择
-  * 使用save_pretrained将加载的bert-large或bert-base权重保存到<data_dir>/<weight_dir>路径下
+  * 使用save_pretrained将加载的bert-large权重保存到<data_dir>/<weight_dir>路径下
 
 ### 3. 软硬件配置与运行信息参考
 

diff --git a/training/nvidia/README.md b/training/nvidia/README.md
@@ -6,6 +6,8 @@ NVIDIA 是人工智能计算领域的领导者。率先采用加速计算，以
 在全栈和数据中心级别实现加速计算，英伟达打造的类似于一个计算堆栈或神经网络，其中包含硬件、系统软件、平台软件和应用四层。每一层都对计算机制造商、服务提供商和开发者开放，让他们以更适合的方式集成到其产品当中。
 [来源](https://images.nvidia.cn/nvimages/aem-dam/zh_cn/Solutions/about-us/documents/NVIDIA-Story-zhCN.pdf)
 
+此外，如无特殊说明，使用fp32数制的用户代码，在Ampere芯片上将默认以tf32格式进行计算。
+
 # FlagPerf适配验证环境说明
 ## 环境配置参考
   - 硬件