File size: 7,400 Bytes
685d968
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
"""
Generate training visualization charts for README
"""
import json
import matplotlib.pyplot as plt
import matplotlib
matplotlib.use('Agg')
import numpy as np
import os

# Create assets directory
os.makedirs('assets', exist_ok=True)

# Load SFT metrics
sft_metrics = []
with open('training/logs/run_20251124_200256/sft_metrics.jsonl', 'r') as f:
    for line in f:
        sft_metrics.append(json.loads(line))

# Load RL metrics
rl_metrics = []
with open('training/logs/run_20251124_200256/rl_metrics.jsonl', 'r') as f:
    for line in f:
        rl_metrics.append(json.loads(line))

# Style settings
plt.style.use('seaborn-v0_8-whitegrid')
colors = {
    'train': '#2563eb',
    'test': '#dc2626',
    'reward': '#059669',
    'f1': '#7c3aed',
    'our_model': '#2563eb',
    'cohere': '#dc2626'
}

# ============ Chart 1: SFT Loss Curve ============
fig, ax = plt.subplots(figsize=(10, 5))

steps = [m['step'] for m in sft_metrics]
train_loss = [m['train_loss'] for m in sft_metrics]

# Get test loss points
test_steps = [m['step'] for m in sft_metrics if 'test_loss' in m]
test_loss = [m['test_loss'] for m in sft_metrics if 'test_loss' in m]

ax.plot(steps, train_loss, color=colors['train'], linewidth=2, label='Train Loss', alpha=0.8)
ax.scatter(test_steps, test_loss, color=colors['test'], s=80, zorder=5, label='Test Loss', marker='o')
ax.plot(test_steps, test_loss, color=colors['test'], linewidth=2, linestyle='--', alpha=0.5)

ax.set_xlabel('Training Step', fontsize=12)
ax.set_ylabel('Loss', fontsize=12)
ax.set_title('SFT Training: Loss Convergence', fontsize=14, fontweight='bold')
ax.legend(loc='upper right', fontsize=10)
ax.set_ylim(0, 6)

# Add annotations
ax.annotate(f'Final: {train_loss[-1]:.3f}', xy=(steps[-1], train_loss[-1]), 
            xytext=(steps[-1]-15, train_loss[-1]+0.5),
            fontsize=9, color=colors['train'])
ax.annotate(f'Best Test: {min(test_loss):.3f}', xy=(test_steps[test_loss.index(min(test_loss))], min(test_loss)), 
            xytext=(test_steps[test_loss.index(min(test_loss))]+5, min(test_loss)+0.3),
            fontsize=9, color=colors['test'])

plt.tight_layout()
plt.savefig('assets/sft_loss.png', dpi=150, bbox_inches='tight')
plt.close()
print("Saved: assets/sft_loss.png")

# ============ Chart 2: RL Reward Progression ============
fig, ax = plt.subplots(figsize=(10, 5))

iterations = [m['iteration'] for m in rl_metrics]
mean_reward = [m['mean_reward'] for m in rl_metrics]
std_reward = [m['std_reward'] for m in rl_metrics]

# Plot with confidence band
ax.fill_between(iterations, 
                [r - s for r, s in zip(mean_reward, std_reward)],
                [r + s for r, s in zip(mean_reward, std_reward)],
                alpha=0.2, color=colors['reward'])
ax.plot(iterations, mean_reward, color=colors['reward'], linewidth=2.5, label='Mean Reward')

ax.set_xlabel('RL Iteration', fontsize=12)
ax.set_ylabel('Reward', fontsize=12)
ax.set_title('RL Training: Reward Progression', fontsize=14, fontweight='bold')
ax.legend(loc='lower right', fontsize=10)
ax.set_ylim(0.5, 1.0)

# Add annotations
ax.annotate(f'Start: {mean_reward[0]:.3f}', xy=(0, mean_reward[0]), 
            xytext=(2, mean_reward[0]-0.05), fontsize=9, color=colors['reward'])
ax.annotate(f'Peak: {max(mean_reward):.3f}', xy=(mean_reward.index(max(mean_reward)), max(mean_reward)), 
            xytext=(mean_reward.index(max(mean_reward))+2, max(mean_reward)+0.02),
            fontsize=9, color=colors['reward'])

plt.tight_layout()
plt.savefig('assets/rl_reward.png', dpi=150, bbox_inches='tight')
plt.close()
print("Saved: assets/rl_reward.png")

# ============ Chart 3: Reward Components ============
fig, ax = plt.subplots(figsize=(10, 5))

r_f1 = [m['mean_r_f1'] for m in rl_metrics]
r_temp = [m['mean_r_temp'] for m in rl_metrics]
r_parity = [m['mean_r_parity'] for m in rl_metrics]
r_eff = [m['mean_r_eff'] for m in rl_metrics]

ax.plot(iterations, r_f1, label='R_F1 (60%)', linewidth=2, color='#2563eb')
ax.plot(iterations, r_temp, label='R_temp (20%)', linewidth=2, color='#7c3aed')
ax.plot(iterations, r_parity, label='R_parity (10%)', linewidth=2, color='#059669')
ax.plot(iterations, r_eff, label='R_eff (10%)', linewidth=2, color='#f59e0b')

ax.set_xlabel('RL Iteration', fontsize=12)
ax.set_ylabel('Reward Component', fontsize=12)
ax.set_title('RL Training: Reward Components', fontsize=14, fontweight='bold')
ax.legend(loc='lower right', fontsize=10)
ax.set_ylim(0.5, 1.05)

plt.tight_layout()
plt.savefig('assets/rl_components.png', dpi=150, bbox_inches='tight')
plt.close()
print("Saved: assets/rl_components.png")

# ============ Chart 4: Model Comparison ============
fig, ax = plt.subplots(figsize=(8, 5))

metrics = ['Avg F1', 'Exact Match', 'Any Match']
our_model = [0.68, 0.60, 0.72]
cohere = [0.61, 0.26, 0.82]

x = np.arange(len(metrics))
width = 0.35

bars1 = ax.bar(x - width/2, our_model, width, label='Ours (8B)', color=colors['our_model'])
bars2 = ax.bar(x + width/2, cohere, width, label='Cohere (104B)', color=colors['cohere'])

ax.set_ylabel('Score', fontsize=12)
ax.set_title('Model Comparison: 50 Marketing Scenarios', fontsize=14, fontweight='bold')
ax.set_xticks(x)
ax.set_xticklabels(metrics, fontsize=11)
ax.legend(loc='upper right', fontsize=10)
ax.set_ylim(0, 1.0)

# Add value labels
for bar in bars1:
    height = bar.get_height()
    ax.annotate(f'{height:.0%}',
                xy=(bar.get_x() + bar.get_width() / 2, height),
                xytext=(0, 3), textcoords="offset points",
                ha='center', va='bottom', fontsize=10, fontweight='bold')

for bar in bars2:
    height = bar.get_height()
    ax.annotate(f'{height:.0%}',
                xy=(bar.get_x() + bar.get_width() / 2, height),
                xytext=(0, 3), textcoords="offset points",
                ha='center', va='bottom', fontsize=10)

plt.tight_layout()
plt.savefig('assets/model_comparison.png', dpi=150, bbox_inches='tight')
plt.close()
print("Saved: assets/model_comparison.png")

# ============ Chart 5: Performance by Difficulty ============
fig, ax = plt.subplots(figsize=(8, 5))

difficulties = ['Easy', 'Medium', 'Hard']
our_f1 = [0.86, 0.65, 0.50]
cohere_f1 = [0.48, 0.64, 0.72]

x = np.arange(len(difficulties))
width = 0.35

bars1 = ax.bar(x - width/2, our_f1, width, label='Ours (8B)', color=colors['our_model'])
bars2 = ax.bar(x + width/2, cohere_f1, width, label='Cohere (104B)', color=colors['cohere'])

ax.set_ylabel('F1 Score', fontsize=12)
ax.set_title('F1 Score by Difficulty Level', fontsize=14, fontweight='bold')
ax.set_xticks(x)
ax.set_xticklabels(difficulties, fontsize=11)
ax.legend(loc='upper right', fontsize=10)
ax.set_ylim(0, 1.0)

# Add value labels
for bar in bars1:
    height = bar.get_height()
    ax.annotate(f'{height:.2f}',
                xy=(bar.get_x() + bar.get_width() / 2, height),
                xytext=(0, 3), textcoords="offset points",
                ha='center', va='bottom', fontsize=10, fontweight='bold')

for bar in bars2:
    height = bar.get_height()
    ax.annotate(f'{height:.2f}',
                xy=(bar.get_x() + bar.get_width() / 2, height),
                xytext=(0, 3), textcoords="offset points",
                ha='center', va='bottom', fontsize=10)

plt.tight_layout()
plt.savefig('assets/difficulty_comparison.png', dpi=150, bbox_inches='tight')
plt.close()
print("Saved: assets/difficulty_comparison.png")

print("\nAll charts generated successfully!")