add Efficientnet on xpu

training/benchmarks/efficientnet/pytorch/train/trainer_adapter.py

@@ -88,17 +88,20 @@ def create_grad_scaler(args):
 
 def backward(args, step: int, epoch: int, loss: torch.Tensor, model: nn.Module,
              optimizer: Optimizer, scaler):
-    optimizer.zero_grad()
     if scaler is not None:
         scaler.scale(loss).backward()
-        if args.clip_grad_norm is not None:
-            # we should unscale the gradients of optimizer's assigned params if do gradient clipping
-            scaler.unscale_(optimizer)
-            nn.utils.clip_grad_norm_(model.parameters(), args.clip_grad_norm)
-        scaler.step(optimizer)
-        scaler.update()
+        if step % args.gradient_accumulation_steps == 0:
+            if args.clip_grad_norm is not None:
+                # we should unscale the gradients of optimizer's assigned params if do gradient clipping
+                scaler.unscale_(optimizer)
+                nn.utils.clip_grad_norm_(model.parameters(), args.clip_grad_norm)
+            scaler.step(optimizer)
+            optimizer.zero_grad()
+            scaler.update()
     else:
         loss.backward()
-        if args.clip_grad_norm is not None:
-            nn.utils.clip_grad_norm_(model.parameters(), args.clip_grad_norm)
-        optimizer.step()
+        if step % args.gradient_accumulation_steps == 0:
+            if args.clip_grad_norm is not None:
+                nn.utils.clip_grad_norm_(model.parameters(), args.clip_grad_norm)
+            optimizer.step()
+            optimizer.zero_grad()

training/kunlunxin/efficientnet-pytorch/README.md

@@ -20,14 +20,17 @@
 
 
 ### 运行情况
-| 训练资源 | 配置文件        | 运行时长(s) | 目标精度 | 收敛精度 | Steps数 | 性能（samples/s) |
+| 训练资源 | 配置文件        | 运行时长(s) | 目标精度 | 收敛精度 | Steps数 | 性能 (samples/s)|
 | -------- | --------------- | ----------- | -------- | -------- | ------- | ---------------- |
 | 单机1卡  | config_R300x1x1 |      |       |    |     |             |
 | 单机2卡  | config_R300x1x2 |      |       |    |     |             |
 | 单机4卡  | config_R300x1x4 |      |       |    |     |             |
-| 单机8卡  | config_R300x1x8 |      |     |    |     |           |
+| 单机8卡  | config_R300x1x8 |      |   82.672    | 72.666   |  868540   |             |
 | 两机8卡  | config_R300x2x8 |      |       |    |     |             |
 
+### 收敛曲线
+![acc](acc.png)
+
 ### 许可证
 
 Apache 2.0 license。

training/kunlunxin/efficientnet-pytorch/config/config_R300x1x8.py

@@ -1,4 +1,5 @@
 from config_common import *
 
 train_batch_size = 64
-eval_batch_size = 64
+eval_batch_size = 128
+gradient_accumulation_steps = 2