up

plot_bench.py

@@ -1,36 +1,59 @@
 #!/usr/bin/env python3
 import matplotlib.pyplot as plt
 import numpy as np
+import seaborn as sns
 
-batch_size_4 = {
-   "A100": [4.75, 3.26, 3.24, 3.10],  # those values are made up
-   "A10": [13.94, 9.81, 10.01, 9.35],
-   "T4": [38.81, 30.09, 29.74, 27.55],
-   "V100": [9.84, 8.16, 8.09, 7.65],
-   "3090": [10.04, 7.82, 7.89, 7.47],
-   "3090TI": [9.07, 7.14, 7.15, 6.81],
-}
+sns.set(font_scale=1.1)
 
-batch_size_16 = {
-   "A100": [18.95, 13.57, 13.67, 12.25],
-   "A10": [0, 37.55, 38.31, 36.81],
-   "T4": [0, 111.47, 113.26, 106.93],
-   "V100": [0, 30.29, 29.84, 28.22],
-   "3090": [0, 29.06, 29.06, 28.2],
-   "3090TI": [0, 26.1, 26.28, 25.46],
+batch_sizes = {
+    "fp16_4": {
+       "A100": [4.75, 3.26, 3.24, 3.10],  # those values are made up
+       "A10": [13.94, 9.81, 10.01, 9.35],
+       "T4": [38.81, 30.09, 29.74, 27.55],
+       "V100": [9.84, 8.16, 8.09, 7.65],
+       "3090": [10.04, 7.82, 7.89, 7.47],
+       "3090TI": [9.07, 7.14, 7.15, 6.81],
+    },
+    "fp16_16": {
+       "A100": [18.95, 13.57, 13.67, 12.25],
+       "A10": [0, 37.55, 38.31, 36.81],
+       "T4": [0, 111.47, 113.26, 106.93],
+       "V100": [0, 30.29, 29.84, 28.22],
+       "3090": [0, 29.06, 29.06, 28.2],
+       "3090TI": [0, 26.1, 26.28, 25.46],
+    },
+    "fp32_4": {
+       "A100": [16.56, 12.42, 12.2, 11.84],
+       "A10": [34.77, 27.63, 22.77, 22.07],
+       "T4": [0, 85.72, 85.78, 84.48],
+       "V100": [0, 25.73, 25.31, 24.7],
+       "3090": [22.69, 21.45, 18.67, 18.09],
+       "3090TI": [20.32, 19.31, 16.9, 16.37],
+    },
+    "fp32_16": {
+       "A100": [0, 47.08, 46.27, 44.8],
+       "A10": [0, 116.49, 88.56, 86.64],
+       "T4": [0, 276.47, 280.26, 270.93],  # numbers are made up
+       "V100": [0, 84.99, 84.73, 82.55],
+       "3090": [0, 85.35, 72.37, 70.25],
+       "3090TI": [0, 75.37, 65.25, 64.32],
+    },
 }
 
-batch_sizes = batch_size_16
+batch_size = 16
+dtype = "fp32"
+
+key = f"{dtype}_{batch_size}"
 
 methods = {
-    "Vanilla Attention": [x[0] for x in batch_sizes.values()],
-    "xFormers": [x[1] for x in batch_sizes.values()],
-    "PyTorch2.0 SDPA": [x[2] for x in batch_sizes.values()],
-    "SDPA + torch.compile": [x[3] for x in batch_sizes.values()],
+    "Vanilla Attention": [x[0] for x in batch_sizes[key].values()],
+    "xFormers": [x[1] for x in batch_sizes[key].values()],
+    "PyTorch2.0 SDPA": [x[2] for x in batch_sizes[key].values()],
+    "SDPA + torch.compile": [x[3] for x in batch_sizes[key].values()],
 }
 
-x = np.arange(len(batch_size_4))  # the label locations
-width = 0.15  # the width of the bars
+x = np.arange(len(batch_sizes[key]))  # the label locations
+width = 0.1  # the width of the bars
 multiplier = 0
 
 fig, ax = plt.subplots(constrained_layout=True)
@@ -38,14 +61,12 @@ fig, ax = plt.subplots(constrained_layout=True)
 for attribute, measurement in methods.items():
     offset = width * multiplier
     rects = ax.bar(x + offset, measurement, width, label=attribute)
-    ax.bar_label(rects, padding=3)
     multiplier += 1
 
 # Add some text for labels, title and custom x-axis tick labels, etc.
 ax.set_ylabel('Time (s)')
-ax.set_title('Inference Speed at Batch Size=4')
-ax.set_xticks(x + width, batch_size_4)
+ax.set_title(f'Inference Speed at Batch Size={batch_size} for {dtype}')
+ax.set_xticks(x + width, batch_sizes[key])
 ax.legend(loc='upper left')
-ax.set_ylim(0, 250)
 
 plt.show()