| """Budget-forcing chart for the model card: overall accuracy vs thinking budget at 2 loops, for base Dory, |
| Dory-Beam-4B (default effort) and Dory-Beam-4B-Accurate. |
| |
| python3 scripts/make_budget_chart.py # -> release/model/assets/accuracy_vs_budget.png |
| |
| At each budget, a sample still thinking is closed with "\n</think>" and gets up to 1,024 answer tokens |
| (modal_app/eval_dory.py::budget_force). The last point is the uncapped 64k run. Same emphasis form and colors as |
| scripts/make_chart.py: base in the de-emphasis gray, the two Dory-Beam models in categorical slots 1 and 2. |
| """ |
|
|
| import json |
| from pathlib import Path |
|
|
| import matplotlib |
|
|
| matplotlib.use("Agg") |
| import matplotlib.pyplot as plt |
|
|
| ROOT = Path(__file__).resolve().parent.parent |
| BUDGET = ROOT / "data" / "eval" / "budget" |
| RES = ROOT / "data" / "eval" / "results" |
| OUT = ROOT / "release" / "model" / "assets" / "accuracy_vs_budget.png" |
|
|
| SURFACE, INK, INK2, MUTED, GRID, AXIS = "#fcfcfb", "#0b0b0b", "#52514e", "#898781", "#e1e0d9", "#c3c2b7" |
| BLUE, ORANGE, BASE = "#2a78d6", "#eb6834", "#898781" |
| CAPS = [2048, 4096, 8192, 16384] |
| |
| SERIES = [("Dory1.0-4B (base)", "base", BASE, "base_loops2-64k"), |
| ("Dory-Beam-4B-Accurate", "Accurate", ORANGE, "mix030_loops2-64k"), |
| ("Dory-Beam-4B, default effort", "Dory-Beam", BLUE, "rE-t050_loops2-default-64k")] |
|
|
|
|
| def curve(tag): |
| hit = {c: [0, 0] for c in CAPS} |
| for r in map(json.loads, open(BUDGET / f"{tag}.jsonl")): |
| hit[r["cap"]][0] += r["correct"] |
| hit[r["cap"]][1] += 1 |
| full = [json.loads(l)["correct"] for l in open(RES / f"{tag}.jsonl")] |
| return [100 * a / n for a, n in hit.values()] + [100 * sum(full) / len(full)] |
|
|
|
|
| def main(): |
| plt.rcParams.update({"font.family": ["Helvetica Neue", "Helvetica", "Arial", "DejaVu Sans"], |
| "font.size": 10, "axes.edgecolor": AXIS, "axes.labelcolor": INK2, |
| "xtick.color": MUTED, "ytick.color": MUTED, "text.color": INK}) |
| fig, ax = plt.subplots(figsize=(9, 5.6), facecolor=SURFACE) |
| fig.subplots_adjust(left=0.08, right=0.74, top=0.74, bottom=0.12) |
| ax.set_facecolor(SURFACE) |
| for side in ("top", "right"): |
| ax.spines[side].set_visible(False) |
| ax.grid(True, axis="y", color=GRID, linewidth=1, zorder=0) |
| ax.tick_params(length=0, pad=6) |
| xs = list(range(len(CAPS) + 1)) |
| ends = [] |
| for label, short, color, tag in SERIES: |
| ys = curve(tag) |
| ax.plot(xs, ys, color=color, linewidth=2, solid_capstyle="round", solid_joinstyle="round", zorder=3) |
| ax.plot(xs, ys, "o", color=color, markersize=8, markeredgecolor=SURFACE, markeredgewidth=2, zorder=4, |
| label=label) |
| ends.append((ys[0], short)) |
| |
| for y, label in ends: |
| ax.annotate(label, (0, y), xytext=(-10, 0), textcoords="offset points", ha="right", |
| va="center", fontsize=9, color=MUTED) |
| ax.set_xlim(-0.75, len(CAPS) + 0.25) |
| ax.set_xticks(xs, ["2k", "4k", "8k", "16k", "no cap\n(64k)"]) |
| ax.set_ylim(50, 92) |
| ax.yaxis.set_major_formatter(plt.FuncFormatter(lambda v, _: f"{v:g}%")) |
| ax.set_xlabel("thinking budget (tokens)", fontsize=9.5, color=INK2) |
| ax.set_ylabel("accuracy, all benchmarks", fontsize=9.5, color=INK2) |
|
|
| fig.text(0.08, 0.94, "On a thinking budget, Dory-Beam-4B answers more correctly", fontsize=15, |
| fontweight="semibold", color=INK) |
| fig.text(0.08, 0.895, "At the budget, a model still thinking is told to stop and answer (up to 1,024 more tokens). " |
| "2 loops,\n1,836 samples over GPQA, MATH-500, AIME 24/25, AMC 23 and 300 held-out problems.", |
| fontsize=9.5, color=INK2, va="top") |
| handles, labels = ax.get_legend_handles_labels() |
| fig.legend(handles[::-1], labels[::-1], loc="upper left", bbox_to_anchor=(0.073, 0.825), ncol=3, frameon=False, |
| fontsize=9.5, handletextpad=0.4, columnspacing=1.6) |
| fig.text(0.765, 0.70, "Dory-Beam-4B minus base,\n95% CI\n\n" |
| "2k +8.7 (+6.3 to +11.1)\n4k +7.7 (+5.3 to +10.1)\n8k +3.9 (+2.0 to +5.7)\n" |
| "16k +2.3 (+0.4 to +4.2)\n\nDefault effort. Paired\nbootstrap over problems.", |
| fontsize=8.5, color=INK2, va="top", linespacing=1.35) |
| OUT.parent.mkdir(parents=True, exist_ok=True) |
| fig.savefig(OUT, dpi=200, facecolor=SURFACE) |
| print("wrote", OUT) |
|
|
|
|
| if __name__ == "__main__": |
| main() |
|
|
