update

src/display/about.py

@@ -18,10 +18,13 @@ Columns and Metrics:
 - Method: The MOE LLMs inference framework.
 - E2E(s): Average End to End generation time in seconds.
 - PRE(s): Prefilling Time of input prompt in seconds.
-- T/s: Tokens throughout per second.
-- S-MBU(%): Sparse Model Bandwidth Utilization.
-- S-MFU(%): Sparse Model FLOPs Utilization.
-- Precision: The precison of used model.
+- Decoding T/s: Tokens throughout per second for decoding.
+- Decoding S-MBU(%): Sparse Model Bandwidth Utilization for decoding.
+- Decoding S-MFU(%): Sparse Model FLOPs Utilization for decoding.
+- Prefill T/s: Tokens throughout per second for Prefilling.
+- Prefill S-MBU(%): Sparse Model Bandwidth Utilization for Prefilling.
+- Prefill S-MFU(%): Sparse Model FLOPs Utilization for Prefilling.
+- Precision: The precision of used model.
 
 """
 

src/display/utils.py

@@ -125,9 +125,9 @@ for task in Tasks:
     # auto_eval_column_dict.append([f"{task.name}_gpu_mem", ColumnContent, ColumnContent(f"{task.value.col_name} {GPU_Mem}", "number", True, hidden=True)])
     auto_eval_column_dict.append([f"{task.name}_gpu", ColumnContent, ColumnContent(f"{task.value.col_name} {GPU_Name}", "str", True, hidden=True)])
     # auto_eval_column_dict.append([f"{task.name}_gpu_util", ColumnContent, ColumnContent(f"{task.value.col_name} {GPU_Util}", "number", True, hidden=True)])
+    auto_eval_column_dict.append([f"{task.name}_prefilling_time", ColumnContent, ColumnContent(f"{task.value.col_name} {PREs}", "number", False, hidden=True)])
     if task.value.benchmark in MULTIPLE_CHOICEs:
         continue
-    auto_eval_column_dict.append([f"{task.name}_prefilling_time", ColumnContent, ColumnContent(f"{task.value.col_name} {PREs}", "number", False, hidden=True)])
     auto_eval_column_dict.append([f"{task.name}_decoding_throughput", ColumnContent, ColumnContent(f"{task.value.col_name} {TS}", "number", True, hidden=True)])
     # if task.value.benchmark != "gsm8k_custom":
     #     continue