File size: 5,379 Bytes
782c694
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
# /// script
# dependencies = [
#     "matplotlib",
#     "numpy",
# ]
# ///

import matplotlib.pyplot as plt
import numpy as np
import os

# get the pathf rom UVNOTE_SETUP env var
setup_path = os.getenv("UVNOTE_INPUT_SETUP", ".")
print(f"Reading benchmark data from: {setup_path}")

num_runs = 5
max_tokens = 64
times = []
with open(os.path.join(setup_path, "benchmark_times.txt"), "r") as f:
    for line in f:
        times.append(float(line.strip()))


avg_time = 0.0
min_time = 0.0
max_time = 0.0
final_mem = {"allocated_gb": 0.0, "peak_gb": 0.0, "reserved_gb": 0.0}

avg_tokens_per_sec = 0.0
with open(os.path.join(setup_path, "benchmark_avg_tokens_per_sec.txt"), "r") as f:
    avg_tokens_per_sec = float(f.read().strip())

times_file = os.path.join(setup_path, "benchmark_times.txt")
memory_file = os.path.join(setup_path, "benchmark_memory.txt")


# Minimal brutalist palette (dark theme): grayscale + 1 accent
ACCENT = '#5ec8f8'   # calm cyan-blue accent
FG = '#e6e6e6'       # light gray text/lines
MUTED = '#9aa0a6'    # muted gray for secondary
GRID = '#333333'     # grid lines

# Styling tuned for clarity, high contrast, few colors
plt.style.use('dark_background')
plt.rcParams['figure.facecolor'] = 'none'
plt.rcParams['axes.facecolor'] = 'none'
plt.rcParams['savefig.facecolor'] = 'none'
plt.rcParams['savefig.transparent'] = True
plt.rcParams['font.family'] = 'monospace'
plt.rcParams['font.weight'] = 'bold'
plt.rcParams['axes.linewidth'] = 3
plt.rcParams['grid.linewidth'] = 2
plt.rcParams['lines.linewidth'] = 3
plt.rcParams['patch.linewidth'] = 2

# Prepare data
runs = list(range(1, len(times) + 1))
tokens_per_sec_all = [max_tokens / t for t in times]

# Chart 1: Throughput Performance
fig1, ax1 = plt.subplots(1, 1, figsize=(12, 6))
fig1.patch.set_alpha(0)
ax1.patch.set_alpha(0)

ax1.plot(runs, tokens_per_sec_all, color=ACCENT, marker='o', markersize=12,
         markerfacecolor=ACCENT, markeredgecolor=FG, markeredgewidth=3, linewidth=5, label='tok/s')
ax1.fill_between(runs, 0, tokens_per_sec_all, alpha=0.2, color=ACCENT)
ax1.axhline(y=avg_tokens_per_sec, color=FG, linestyle='--', linewidth=3,
            label=f'AVG: {avg_tokens_per_sec:.1f}')
ax1.set_title('THROUGHPUT PERFORMANCE', color=FG, fontsize=18, pad=20, fontweight='bold')
ax1.set_xlabel('RUN NUMBER', color=FG, fontsize=14, fontweight='bold')
ax1.set_ylabel('TOKENS/SEC', color=FG, fontsize=14, fontweight='bold')
ax1.grid(True, color=GRID, alpha=0.5, linewidth=2)
ax1.tick_params(colors=FG, labelsize=12)
legend1 = ax1.legend(frameon=False, loc='lower right')
for text in legend1.get_texts():
    text.set_color(FG)
    text.set_fontweight('bold')
plt.tight_layout()
plt.savefig('throughput.png', dpi=150, bbox_inches='tight', transparent=True)
plt.show()

# Chart 2: Generation Latency
fig2, ax2 = plt.subplots(1, 1, figsize=(12, 6))
fig2.patch.set_alpha(0)
ax2.patch.set_alpha(0)

bar_colors = [ACCENT if i % 2 == 0 else MUTED for i in range(len(times))]
bars = ax2.bar(runs, times, color=bar_colors, edgecolor=FG, linewidth=3, width=0.6)
ax2.axhline(y=avg_time, color=FG, linestyle='--', linewidth=3,
            label=f'AVG: {avg_time:.2f}s')
for i, (run, time, bar) in enumerate(zip(runs, times, bars)):
    ax2.text(run, time + 0.02, f'{time:.2f}s', ha='center', va='bottom',
             color=FG, fontweight='bold', fontsize=11)
ax2.set_title('GENERATION LATENCY', color=FG, fontsize=18, pad=20, fontweight='bold')
ax2.set_xlabel('RUN NUMBER', color=FG, fontsize=14, fontweight='bold')
ax2.set_ylabel('TIME (SECONDS)', color=FG, fontsize=14, fontweight='bold')
ax2.grid(True, axis='y', color=GRID, alpha=0.5, linewidth=2)
ax2.tick_params(colors=FG, labelsize=12)
ax2.set_ylim(0, max(times) * 1.15)
legend2 = ax2.legend(frameon=False, loc='upper right')
for text in legend2.get_texts():
    text.set_color(FG)
    text.set_fontweight('bold')
plt.tight_layout()
plt.savefig('latency.png', dpi=150, bbox_inches='tight', transparent=True)
plt.show()

# Chart 3: Memory Usage
fig3, ax3 = plt.subplots(1, 1, figsize=(12, 6))
fig3.patch.set_alpha(0)
ax3.patch.set_alpha(0)

memory_labels = ['ALLOCATED', 'PEAK', 'RESERVED']
memory_values = [final_mem['allocated_gb'], final_mem['peak_gb'], final_mem['reserved_gb']]
colors_mem = [MUTED, ACCENT, FG]
bars = ax3.barh(memory_labels, memory_values, color=colors_mem, edgecolor=FG, linewidth=3, height=0.5)
for i, (label, value, bar) in enumerate(zip(memory_labels, memory_values, bars)):
    ax3.text(value + 0.5, i, f'{value:.1f} GB', va='center',
             color=FG, fontweight='bold', fontsize=13)
ax3.set_title('MEMORY USAGE', color=FG, fontsize=18, pad=20, fontweight='bold')
ax3.set_xlabel('GIGABYTES', color=FG, fontsize=14, fontweight='bold')
ax3.set_xlim(0, max(memory_values) * 1.3)
ax3.grid(True, axis='x', color=GRID, alpha=0.5, linewidth=2)
ax3.tick_params(colors=FG, labelsize=12)
ax3.set_yticks(range(len(memory_labels)))
ax3.set_yticklabels(memory_labels, fontweight='bold')
plt.tight_layout()
plt.savefig('memory.png', dpi=150, bbox_inches='tight', transparent=True)
plt.show()

print(f"\n📊 Charts saved as:")
print(f"  • throughput.png")
print(f"  • latency.png")
print(f"  • memory.png")
print(f"\nBenchmark Summary:")
print(f"  avg tokens/sec: {avg_tokens_per_sec:.1f}")
print(f"  min time: {min_time:.3f}s")
print(f"  max time: {max_time:.3f}s")
print(f"  peak memory: {final_mem['peak_gb']:.2f}GB")