docs: refresh top-models telemetry figure and add uv generator

README.md

@@ -2,7 +2,7 @@
 
 [![CI](https://github.com/NVIDIA-NeMo/DataDesigner/actions/workflows/ci.yml/badge.svg)](https://github.com/NVIDIA-NeMo/DataDesigner/actions/workflows/ci.yml)
 [![License](https://img.shields.io/badge/License-Apache_2.0-blue.svg)](https://opensource.org/licenses/Apache-2.0)
-[![Python 3.10 - 3.14](https://img.shields.io/badge/🐍_Python-3.10_|_3.11_|_3.12_|_3.13_|_3.14-blue.svg)](https://www.python.org/downloads/) [![NeMo Microservices](https://img.shields.io/badge/NeMo-Microservices-76b900)](https://docs.nvidia.com/nemo/microservices/latest/index.html) [![Code](https://img.shields.io/badge/Code-Documentation-8A2BE2.svg)](https://nvidia-nemo.github.io/DataDesigner/) ![Tokens](https://img.shields.io/badge/400+_Billion-Tokens_Generated-76b900.svg?logo=nvidia&logoColor=white)
+[![Python 3.10 - 3.14](https://img.shields.io/badge/🐍_Python-3.10_|_3.11_|_3.12_|_3.13_|_3.14-blue.svg)](https://www.python.org/downloads/) [![NeMo Microservices](https://img.shields.io/badge/NeMo-Microservices-76b900)](https://docs.nvidia.com/nemo/microservices/latest/index.html) [![Code](https://img.shields.io/badge/Code-Documentation-8A2BE2.svg)](https://nvidia-nemo.github.io/DataDesigner/) ![Tokens](https://img.shields.io/badge/2.6T+-Tokens_Processed-76b900.svg?logo=nvidia&logoColor=white)
 
 **Generate high-quality synthetic datasets from scratch or using your own seed data.**
 
@@ -153,11 +153,11 @@ Disable with `NEMO_TELEMETRY_ENABLED=false`. **[More details →](#telemetry-and
 
 ### Top models (YTD)
 
-Aggregate model usage across synthetic data generation jobs, year-to-date 1/1/2026–5/1/2026:
+Aggregate model usage across synthetic data generation jobs, year-to-date 1/1/2026–6/1/2026:
 
 ![Top models used for synthetic data generation](docs/images/top-models.png)
 
-_Last updated on May 1, 2026_
+_Last updated on June 1, 2026_
 
 ---
 

docs/scripts/generate_top_models_figure.py

@@ -0,0 +1,284 @@
+# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+# /// script
+# requires-python = ">=3.10"
+# dependencies = [
+#     "matplotlib==3.9.4",
+# ]
+# ///
+"""Regenerate the "Top Model Usage" telemetry figure.
+
+Renders the ranked input-vs-output token breakdown shown in the README's
+"Top models (YTD)" section, styled to match the Data Designer devnote charts
+(near-black canvas, NVIDIA-green duotone). The same PNG is written to both
+tracked copies so the README and Fern docs site stay in sync:
+
+    docs/images/top-models.png    (rendered by the README)
+    fern/images/top-models.png    (Fern's /images/* mirror)
+
+The source telemetry export lives at docs/scripts/top-model-usage.csv with
+columns: model name, input (context) tokens, output (generated) tokens, plus a
+trailing "Other" aggregate row. Drop in a fresh export to refresh the figure.
+
+Run:
+    # Regenerate from the committed CSV (zero args)
+    uv run docs/scripts/generate_top_models_figure.py
+
+    # Refresh from a new telemetry export
+    uv run docs/scripts/generate_top_models_figure.py --csv ~/Downloads/new-export.csv
+
+    # Options
+    uv run docs/scripts/generate_top_models_figure.py --help
+"""
+
+from __future__ import annotations
+
+import argparse
+import csv
+import shutil
+from pathlib import Path
+
+import matplotlib.pyplot as plt
+from matplotlib import rcParams
+from matplotlib.ticker import FuncFormatter, MaxNLocator
+
+# Repo root is two levels up from docs/scripts/.
+REPO_ROOT = Path(__file__).resolve().parents[2]
+DEFAULT_CSV = REPO_ROOT / "docs" / "scripts" / "top-model-usage.csv"
+# Tracked copies of the figure; first entry is the canonical render target.
+# docs/images/ is what the README renders; fern/images/ is Fern's mirror for
+# /images/* references.
+TARGETS = (
+    REPO_ROOT / "docs" / "images" / "top-models.png",
+    REPO_ROOT / "fern" / "images" / "top-models.png",
+)
+
+# ---------------------------------------------------------------- palette ----
+BG = "#0E0E0E"  # near-black canvas (matches DD devnote charts)
+GREEN = "#76B900"  # NVIDIA green -> input (context) tokens
+LIME = "#C5E86C"  # light NVIDIA-tint green -> output (generated) tokens
+WHITE = "#FFFFFF"
+SUBTLE = "#9A9A9A"
+AXIS = "#B8B8B8"
+MODELNAME = "#ECECEC"
+GRID = "#FFFFFF"
+SPINE = "#4A4A4A"
+INK = "#0E0E0E"  # dark ink for labels sitting on bright bars
+
+B = 1e9  # render token counts in billions
+
+
+def load_rows(csv_path: Path) -> list[tuple[str, float, float]]:
+    """Parse the telemetry CSV into (name, input_tokens, output_tokens) rows."""
+    rows: list[tuple[str, float, float]] = []
+    with csv_path.open(newline="", encoding="utf-8-sig") as fh:
+        reader = csv.reader(fh)
+        next(reader)  # header
+        for name, inp, out in reader:
+            rows.append((name, float(inp.replace(",", "")), float(out.replace(",", ""))))
+    return rows
+
+
+def configure_matplotlib() -> None:
+    """Pin rendering to deterministic settings so the asset is reproducible.
+
+    Forces the Agg backend and matplotlib's bundled DejaVu Sans face rather than
+    opportunistically selecting a system Helvetica/Arial. Combined with the
+    pinned matplotlib version in the script metadata, this keeps the checked-in
+    PNG byte-reproducible across machines and CI.
+    """
+    plt.switch_backend("Agg")
+    rcParams["font.family"] = "DejaVu Sans"
+    rcParams["font.size"] = 13
+
+
+def fmt(v: float) -> str:
+    """Compact billions/trillions label."""
+    if v >= 1e12:
+        return f"{v / 1e12:.2f}T"
+    return f"{v / 1e9:.0f}B"
+
+
+def render(rows: list[tuple[str, float, float]], out_path: Path) -> None:
+    """Render the ranked stacked-bar figure to out_path."""
+    # Split the "Other" aggregate out; sort named models by total descending.
+    other = next((r for r in rows if r[0].lower() == "other"), None)
+    models = [r for r in rows if r[0].lower() != "other"]
+    models.sort(key=lambda r: r[1] + r[2], reverse=True)
+
+    n = len(models)
+    ypos = list(range(n, 0, -1))  # n, n-1, ... 1  (top -> down)
+    labels = [m[0] for m in models]
+    inputs = [m[1] for m in models]
+    outputs = [m[2] for m in models]
+
+    if other is not None:
+        ypos.append(-0.6)  # gap below the named models
+        labels.append("Other models")
+        inputs.append(other[1])
+        outputs.append(other[2])
+
+    fig, ax = plt.subplots(figsize=(14.5, 9.2), dpi=200)
+    fig.patch.set_facecolor(BG)
+    ax.set_facecolor(BG)
+
+    xmax = max(i + o for i, o in zip(inputs, outputs)) / B
+    bar_h = 0.62
+
+    for idx, (y, inp, out) in enumerate(zip(ypos, inputs, outputs)):
+        is_other = other is not None and idx == len(ypos) - 1
+        a = 0.45 if is_other else 1.0
+
+        ax.barh(y, inp / B, height=bar_h, color=GREEN, alpha=a, zorder=3, edgecolor=BG, linewidth=1.2)
+        ax.barh(y, out / B, height=bar_h, left=inp / B, color=LIME, alpha=a, zorder=3, edgecolor=BG, linewidth=1.2)
+
+        total = (inp + out) / B
+        ax.text(
+            total + xmax * 0.008,
+            y,
+            fmt(inp + out),
+            va="center",
+            ha="left",
+            color=SUBTLE if is_other else WHITE,
+            fontweight="bold",
+            fontsize=13.5,
+            zorder=5,
+        )
+
+        # In-segment value labels only where the segment is wide enough.
+        if inp / B > xmax * 0.085:
+            ax.text(
+                (inp / B) / 2,
+                y,
+                fmt(inp),
+                va="center",
+                ha="center",
+                color=INK,
+                fontweight="bold",
+                fontsize=11.5,
+                alpha=a,
+                zorder=5,
+            )
+        if out / B > xmax * 0.085:
+            ax.text(
+                inp / B + (out / B) / 2,
+                y,
+                fmt(out),
+                va="center",
+                ha="center",
+                color=INK,
+                fontweight="bold",
+                fontsize=11.5,
+                alpha=a,
+                zorder=5,
+            )
+
+    # ------------------------------------------------------------- axes -----
+    ax.set_yticks(ypos)
+    ax.set_yticklabels(labels, fontsize=12.5)
+    is_other_flags = [False] * n + ([True] if other else [])
+    for tick, is_other in zip(ax.get_yticklabels(), is_other_flags):
+        tick.set_color(SUBTLE if is_other else MODELNAME)
+        if is_other:
+            tick.set_fontstyle("italic")
+
+    ax.set_xlim(0, xmax * 1.13)
+    ax.set_ylim(-1.3, n + 0.8)
+
+    # Derive ticks from the data so the axis stays sane as totals grow; fmt()
+    # promotes B -> T automatically, so the labels never need hand-editing.
+    ax.xaxis.set_major_locator(MaxNLocator(nbins=8, steps=[1, 2, 2.5, 5, 10]))
+    ax.xaxis.set_major_formatter(FuncFormatter(lambda v, _pos: "0" if v <= 0 else fmt(v * B)))
+    ax.tick_params(axis="y", length=0, pad=10)
+    ax.tick_params(axis="x", colors=AXIS, length=0, pad=8, labelsize=11)
+    ax.set_xlabel("Tokens processed", color=AXIS, fontsize=12.5, labelpad=12)
+
+    ax.xaxis.grid(True, color=GRID, alpha=0.07, linewidth=1, zorder=0)
+    ax.set_axisbelow(True)
+    for s in ("top", "right"):
+        ax.spines[s].set_visible(False)
+    for s in ("bottom", "left"):
+        ax.spines[s].set_color(SPINE)
+        ax.spines[s].set_linewidth(1.0)
+
+    # ---------------------------------------------------------- titling -----
+    fig.subplots_adjust(left=0.235, right=0.965, top=0.83, bottom=0.085)
+    # Signature DD green left-accent rule (mirrors the .devnote-dek element).
+    ax.add_patch(
+        plt.Rectangle(
+            (-0.018, 1.045),
+            0.006,
+            0.135,
+            transform=ax.transAxes,
+            facecolor=GREEN,
+            edgecolor="none",
+            clip_on=False,
+            zorder=6,
+        )
+    )
+    ax.text(
+        0.012,
+        1.145,
+        "Top Model Usage",
+        transform=ax.transAxes,
+        color=WHITE,
+        fontweight="bold",
+        fontsize=26,
+        ha="left",
+        va="bottom",
+    )
+    ax.text(
+        0.012,
+        1.07,
+        "Context vs. generated tokens across the most-used models",
+        transform=ax.transAxes,
+        color=SUBTLE,
+        fontsize=13.5,
+        ha="left",
+        va="bottom",
+    )
+
+    # Manual legend, top-right of the plotting area.
+    leg_x, leg_y = 0.99, 1.115
+    legend = [(GREEN, "Input  ·  context tokens"), (LIME, "Output  ·  generated tokens")]
+    for i, (c, lbl) in enumerate(legend):
+        yy = leg_y - i * 0.052
+        ax.add_patch(
+            plt.Rectangle(
+                (leg_x - 0.205, yy - 0.012),
+                0.022,
+                0.026,
+                transform=ax.transAxes,
+                facecolor=c,
+                edgecolor="none",
+                clip_on=False,
+                zorder=6,
+            )
+        )
+        ax.text(leg_x - 0.172, yy, lbl, transform=ax.transAxes, color=MODELNAME, fontsize=12, ha="left", va="center")
+
+    out_path.parent.mkdir(parents=True, exist_ok=True)
+    fig.savefig(out_path, facecolor=BG, dpi=200, bbox_inches="tight", pad_inches=0.25)
+    plt.close(fig)
+
+
+def main() -> None:
+    parser = argparse.ArgumentParser(description=__doc__.splitlines()[0])
+    parser.add_argument("--csv", type=Path, default=DEFAULT_CSV, help=f"Telemetry export CSV (default: {DEFAULT_CSV})")
+    args = parser.parse_args()
+
+    configure_matplotlib()
+    rows = load_rows(args.csv)
+
+    primary, *mirrors = TARGETS
+    render(rows, primary)
+    for mirror in mirrors:
+        mirror.parent.mkdir(parents=True, exist_ok=True)
+        shutil.copyfile(primary, mirror)
+
+    for target in TARGETS:
+        print(f"wrote {target.relative_to(REPO_ROOT)}")
+
+
+if __name__ == "__main__":
+    main()

docs/scripts/top-model-usage.csv

@@ -0,0 +1,12 @@
+"Top 10 Model Usage","Input Tokens (Context)","Output Tokens (Generated)"
+"openai/gpt-oss-120b","581,991,035,603","69,823,305,523"
+"google/gemma-4-31B-it","305,097,721,372","139,909,403,045"
+"Qwen/Qwen3-VL-235B-A22B-Instruct","252,299,362,661","2,506,282,983"
+"Qwen/Qwen3.5-397B-A17B-FP8","185,392,972,434","72,214,577,833"
+"google/gemma-4-26B-A4B-it","112,014,037,550","16,872,099,656"
+"Qwen/Qwen3.5-122B-A10B","87,216,522,178","41,888,115,144"
+"gcp/google/gemini-3.1-flash-lite-preview","61,793,069,244","7,206,950,344"
+"Qwen/Qwen3-VL-235B-A22B-Thinking-FP8","52,889,942,762","9,031,174,934"
+"Qwen/Qwen3.6-35B-A3B","46,115,903,437","4,269,353,359"
+"Qwen/Qwen3-VL-30B-A3B-Thinking","42,718,861,428","7,201,483,397"
+Other,"394,226,701,751","189,813,318,234"