File size: 5,567 Bytes
756cf82
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
#!/usr/bin/env python3
from __future__ import annotations

import os
from pathlib import Path

os.environ.setdefault("MPLCONFIGDIR", "/tmp/matplotlib")

import matplotlib.pyplot as plt


OUT_DIR = Path("figures")
OUT_DIR.mkdir(exist_ok=True)

FINAL_LONG_RUN_PPL = 19.7822

EXPERIMENTS = [
    {
        "name": "Naive\nbaseline",
        "legend": "baseline",
        "color": "#6b7280",
        "public_ppl": 42.2650,
        "train_loss_4500": 3.6290,
        "internal_val_loss_4500": 3.6823,
        "curve": [
            (0, 10.9696), (250, 6.0841), (500, 5.2423), (750, 4.6500),
            (1000, 4.3752), (1250, 4.2232), (1500, 4.1024),
            (1750, 3.9847), (2000, 3.9412), (2250, 3.8667),
            (2500, 3.8559), (2750, 3.7679), (3000, 3.7503),
            (3250, 3.7585), (3500, 3.7046), (3750, 3.7635),
            (4000, 3.6673), (4250, 3.6389), (4500, 3.6823),
        ],
    },
    {
        "name": "Mixed\ndata",
        "legend": "data",
        "color": "#2563eb",
        "public_ppl": 38.7702,
        "train_loss_4500": 3.4646,
        "internal_val_loss_4500": 3.4312,
        "curve": [
            (0, 10.9799), (250, 5.8726), (500, 5.1043), (750, 4.5490),
            (1000, 4.1688), (1250, 4.0293), (1500, 3.9846),
            (1750, 3.8147), (2000, 3.7881), (2250, 3.7224),
            (2500, 3.6965), (2750, 3.6746), (3000, 3.5739),
            (3250, 3.5122), (3500, 3.5903), (3750, 3.4440),
            (4000, 3.5392), (4250, 3.4584), (4500, 3.4312),
        ],
    },
    {
        "name": "Muon\noptimizer",
        "legend": "Muon",
        "color": "#dc2626",
        "public_ppl": 40.0987,
        "train_loss_4500": 3.5600,
        "internal_val_loss_4500": 3.6192,
        "curve": [
            (0, 10.9696), (250, 5.9106), (500, 5.1345), (750, 4.4705),
            (1000, 4.2422), (1250, 4.1028), (1500, 3.9927),
            (1750, 3.8881), (2000, 3.8523), (2250, 3.7771),
            (2500, 3.7735), (2750, 3.6887), (3000, 3.6751),
            (3250, 3.6859), (3500, 3.6309), (3750, 3.6952),
            (4000, 3.6025), (4250, 3.5744), (4500, 3.6192),
        ],
    },
    {
        "name": "Lyra\narchitecture",
        "legend": "arch",
        "color": "#059669",
        "public_ppl": 36.5445,
        "train_loss_4500": 3.4135,
        "internal_val_loss_4500": 3.4889,
        "curve": [
            (0, 10.9630), (250, 5.5041), (500, 4.6324), (750, 4.2915),
            (1000, 4.0727), (1250, 3.9969), (1500, 3.8429),
            (1750, 3.7813), (2000, 3.7209), (2250, 3.7253),
            (2500, 3.6427), (2750, 3.6593), (3000, 3.5319),
            (3250, 3.6220), (3500, 3.5259), (3750, 3.5657),
            (4000, 3.5040), (4250, 3.5018), (4500, 3.4889),
        ],
    },
    {
        "name": "Combined\nshort run",
        "legend": "combined",
        "color": "#7c3aed",
        "public_ppl": 32.1195,
        "train_loss_4500": 3.2819,
        "internal_val_loss_4500": 3.3663,
        "curve": [
            (0, 10.9511), (250, 5.5112), (500, 4.4614), (750, 4.0177),
            (1000, 3.8477), (1250, 3.7237), (1500, 3.7700),
            (1750, 3.6963), (2000, 3.5705), (2250, 3.5041),
            (2500, 3.4608), (2750, 3.3681), (3000, 3.3872),
            (3250, 3.4139), (3500, 3.3327), (3750, 3.3329),
            (4000, 3.3754), (4250, 3.2846), (4500, 3.3663),
        ],
    },
]


def main() -> None:
    plt.style.use("seaborn-v0_8-whitegrid")
    fig, axes = plt.subplots(1, 2, figsize=(13.5, 5.2), constrained_layout=True)

    ax = axes[0]
    labels = [exp["name"] for exp in EXPERIMENTS]
    ppls = [exp["public_ppl"] for exp in EXPERIMENTS]
    colors = [exp["color"] for exp in EXPERIMENTS]
    bars = ax.bar(labels, ppls, color=colors, width=0.68)
    ax.axhline(FINAL_LONG_RUN_PPL, color="#111827", linewidth=1.8, linestyle="--")
    ax.text(
        0.02,
        FINAL_LONG_RUN_PPL + 0.3,
        f"final long run: {FINAL_LONG_RUN_PPL:.2f}",
        transform=ax.get_yaxis_transform(),
        ha="left",
        va="bottom",
        fontsize=9,
        color="#111827",
    )
    ax.set_title("Course Public Validation Perplexity", fontsize=13, weight="bold")
    ax.set_ylabel("perplexity, lower is better")
    ax.set_ylim(FINAL_LONG_RUN_PPL * 0.85, max(ppls) * 1.12)
    for bar, val in zip(bars, ppls):
        ax.text(
            bar.get_x() + bar.get_width() / 2,
            val,
            f"{val:.1f}",
            ha="center",
            va="bottom",
            fontsize=9,
        )

    ax = axes[1]
    for exp in EXPERIMENTS:
        steps = [x for x, _ in exp["curve"]]
        losses = [y for _, y in exp["curve"]]
        ax.plot(
            steps,
            losses,
            marker="o",
            linewidth=2.0,
            markersize=4,
            color=exp["color"],
            label=exp["legend"],
        )
    ax.set_title("Short-Run Heldout Loss Curves", fontsize=13, weight="bold")
    ax.set_xlabel("training iteration")
    ax.set_ylabel("validation loss")
    ax.legend(frameon=True, fontsize=9)
    ax.text(
        0.02,
        -0.18,
        "Each run changes one variable and uses the same 4,500-iteration budget; public PPL is the comparable metric.",
        transform=ax.transAxes,
        fontsize=8.5,
        color="#4b5563",
    )

    fig.suptitle("Ablation Summary for the Presentation", fontsize=15, weight="bold")
    for suffix in ("png", "pdf"):
        out = OUT_DIR / f"presentation_ablation_summary_standalone.{suffix}"
        fig.savefig(out, dpi=220)
        print(out)


if __name__ == "__main__":
    main()