yzsp-studio/mediakit-shared-suite / ppt
ppt
Unverified飞书幻灯片:创建和编辑幻灯片。创建演示文稿、读取幻灯片内容、管理幻灯片页面(创建、删除、读取、局部替换)。当用户需要创建或编辑幻灯片、读取或修改单个页面时使用。
scripts/svg_relation_lib/render/render_tx_taxonomy_variants.py
View on GitHub#!/usr/bin/env python3
"""Regenerate the 4 TX_taxonomy variants into step2_codified.
Variants:
- baseline: 4 domain × 3 sub × 2 leaf (24 leaves · spec 4×3×2) · sidebar on
- dense: 5 domain × 4 sub × 3 leaf (60 leaves) · sidebar on
- sparse: 3 domain × 2 sub × 1 leaf (6 leaves) · sidebar on
- minimal: 3 domain × 2 sub × 1 leaf · sidebar off
"""
from __future__ import annotations
import os
import sys
HERE = os.path.dirname(os.path.abspath(__file__))
if HERE not in sys.path:
sys.path.insert(0, HERE)
from svg_relation_lib.presets.hero_embed_07_TX_taxonomy_v2 import ( # noqa: E402
render_hero_embed_tx_taxonomy_v2,
build_taxonomy_data,
)
OUT_DIR = os.path.abspath(os.path.join(
HERE, "..", "final_svg_relation", "07_TX_taxonomy", "step2_codified",
))
# ═════════════════════════════════════════════════════════════════
# DENSE · 5 × 4 × 3
# ═════════════════════════════════════════════════════════════════
DENSE_ROOT = {"label": "LLM", "sub": "BENCH·XL", "tag": "ROOT"}
DENSE_DOMAINS = [
{
"key": "reason", "code": "D1", "label": "Reasoning",
"sub": "logic · math · plan",
"hue": "magenta", "leaf_count": 12,
"subs": [
{"code": "S01", "label": "Mathematical", "sub": "arithmetic · algebra",
"leaves": [
{"code": "L01", "label": "MATH-500", "tag": "500 q"},
{"code": "L02", "label": "GSM8K", "tag": "8.5k q"},
{"code": "L03", "label": "MATH-Hard", "tag": "1k q"},
]},
{"code": "S02", "label": "Abstract logic", "sub": "grid · induction",
"leaves": [
{"code": "L04", "label": "ARC-AGI", "tag": "400 tk"},
{"code": "L05", "label": "BBH", "tag": "23 sub"},
{"code": "L06", "label": "LogicNLI", "tag": "20k pr"},
]},
{"code": "S03", "label": "Planning", "sub": "multi-step · agent",
"leaves": [
{"code": "L07", "label": "PlanBench", "tag": "1.2k"},
{"code": "L08", "label": "TravelPlanner","tag": "1.6k"},
{"code": "L09", "label": "WebArena", "tag": "812"},
]},
{"code": "S04", "label": "Causal", "sub": "counterfactual · do-calc",
"leaves": [
{"code": "L10", "label": "CLadder", "tag": "10k q"},
{"code": "L11", "label": "CausalQA", "tag": "3.4k"},
{"code": "L12", "label": "CRAB", "tag": "1.1k"},
]},
],
},
{
"key": "know", "code": "D2", "label": "Knowledge",
"sub": "facts · exam · sense",
"hue": "green", "leaf_count": 12,
"subs": [
{"code": "S05", "label": "Academic QA", "sub": "multi-domain",
"leaves": [
{"code": "L13", "label": "MMLU-Pro", "tag": "12k"},
{"code": "L14", "label": "GPQA-Diamond","tag": "198 q"},
{"code": "L15", "label": "AGIEval", "tag": "8.1k"},
]},
{"code": "S06", "label": "Open-domain", "sub": "retrieval",
"leaves": [
{"code": "L16", "label": "TriviaQA", "tag": "95k"},
{"code": "L17", "label": "NQ", "tag": "3.6k"},
{"code": "L18", "label": "HotpotQA", "tag": "112k"},
]},
{"code": "S07", "label": "Commonsense", "sub": "pragmatics",
"leaves": [
{"code": "L19", "label": "WinoGrande", "tag": "44k"},
{"code": "L20", "label": "PIQA", "tag": "21k"},
{"code": "L21", "label": "HellaSwag", "tag": "70k"},
]},
{"code": "S08", "label": "Domain expert", "sub": "law · med · fin",
"leaves": [
{"code": "L22", "label": "LegalBench", "tag": "162 tk"},
{"code": "L23", "label": "MedQA", "tag": "12.7k"},
{"code": "L24", "label": "FinBen", "tag": "36 tk"},
]},
],
},
{
"key": "code", "code": "D3", "label": "Coding",
"sub": "gen · repair · SQL",
"hue": "slate", "leaf_count": 12,
"subs": [
{"code": "S09", "label": "Code synthesis", "sub": "function",
"leaves": [
{"code": "L25", "label": "HumanEval+", "tag": "164"},
{"code": "L26", "label": "MBPP+", "tag": "378"},
{"code": "L27", "label": "APPS", "tag": "10k"},
]},
{"code": "S10", "label": "Repo repair", "sub": "issue → PR",
"leaves": [
{"code": "L28", "label": "SWE-Bench", "tag": "2.3k"},
{"code": "L29", "label": "SWE-Bench-Verif", "tag": "500"},
{"code": "L30", "label": "R2E", "tag": "246"},
]},
{"code": "S11", "label": "Data / SQL", "sub": "text2sql",
"leaves": [
{"code": "L31", "label": "Spider 2.0", "tag": "632"},
{"code": "L32", "label": "DS-1000", "tag": "1k"},
{"code": "L33", "label": "BIRD", "tag": "12k"},
]},
{"code": "S12", "label": "Multi-file", "sub": "codebase edit",
"leaves": [
{"code": "L34", "label": "RepoBench", "tag": "27k"},
{"code": "L35", "label": "CrossCode", "tag": "9.5k"},
{"code": "L36", "label": "SWE-Gym", "tag": "2.6k"},
]},
],
},
{
"key": "safe", "code": "D4", "label": "Safety",
"sub": "truth · tox · privacy",
"hue": "rust", "leaf_count": 12,
"subs": [
{"code": "S13", "label": "Truthfulness", "sub": "halluc · calib",
"leaves": [
{"code": "L37", "label": "TruthfulQA", "tag": "817"},
{"code": "L38", "label": "HaluEval", "tag": "35k"},
{"code": "L39", "label": "FActScore", "tag": "500"},
]},
{"code": "S14", "label": "Toxicity", "sub": "stereotype",
"leaves": [
{"code": "L40", "label": "ToxiGen", "tag": "274k"},
{"code": "L41", "label": "BBQ", "tag": "58k"},
{"code": "L42", "label": "RealTox", "tag": "100k"},
]},
{"code": "S15", "label": "Robustness", "sub": "jailbreak",
"leaves": [
{"code": "L43", "label": "JailbreakBench", "tag": "100"},
{"code": "L44", "label": "HarmBench", "tag": "510"},
{"code": "L45", "label": "AdvBench", "tag": "520"},
]},
{"code": "S16", "label": "Privacy", "sub": "PII · leakage",
"leaves": [
{"code": "L46", "label": "PII-Leak", "tag": "3.2k"},
{"code": "L47", "label": "PrivacyQA", "tag": "1.7k"},
{"code": "L48", "label": "MemExtract", "tag": "820"},
]},
],
},
{
"key": "multi", "code": "D5", "label": "Multimodal",
"sub": "vision · audio · vid",
"hue": "gold_p", "leaf_count": 12,
"subs": [
{"code": "S17", "label": "Vision QA", "sub": "chart · doc",
"leaves": [
{"code": "L49", "label": "MMMU-Pro", "tag": "1.5k"},
{"code": "L50", "label": "ChartQA", "tag": "32k"},
{"code": "L51", "label": "DocVQA", "tag": "50k"},
]},
{"code": "S18", "label": "Video", "sub": "temporal",
"leaves": [
{"code": "L52", "label": "VideoMME", "tag": "900"},
{"code": "L53", "label": "MVBench", "tag": "4k"},
{"code": "L54", "label": "TempCompass","tag": "1.2k"},
]},
{"code": "S19", "label": "Audio", "sub": "speech · music",
"leaves": [
{"code": "L55", "label": "AudioBench","tag": "1k"},
{"code": "L56", "label": "MMAU", "tag": "5k"},
{"code": "L57", "label": "AIR-Bench", "tag": "19k"},
]},
{"code": "S20", "label": "Grounding", "sub": "detect · point",
"leaves": [
{"code": "L58", "label": "RefCOCO", "tag": "50k"},
{"code": "L59", "label": "PointQA", "tag": "3.9k"},
{"code": "L60", "label": "SEED-Bench","tag": "24k"},
]},
],
},
]
DENSE_KPIS = [
{"kicker": "DOMAINS", "value": "5", "note": "top-level buckets", "hue": "magenta"},
{"kicker": "SUB-GROUPS", "value": "20", "note": "4 per domain", "hue": "green"},
{"kicker": "BENCHMARKS", "value": "60", "note": "leaf-level test sets", "hue": "rust"},
{"kicker": "COVERAGE", "value": "92%", "note": "extended HELM+ suite", "hue": "slate"},
]
DENSE_COVERAGE = [
{"label": "Reasoning", "pct": 92, "note": "12 of 13 targeted", "hue": "magenta"},
{"label": "Knowledge", "pct": 96, "note": "12 of 12 core suite", "hue": "green"},
{"label": "Coding", "pct": 90, "note": "12 of 13 · multi-file gap", "hue": "slate"},
{"label": "Safety", "pct": 80, "note": "12 of 15 · privacy pending", "hue": "rust"},
{"label": "Multimodal", "pct": 75, "note": "12 of 16 · audio incomplete", "hue": "gold_p"},
]
DENSE_HUE_LEGEND = [
{"code": "D1", "label": "Reasoning", "note": "math / logic", "hue": "magenta"},
{"code": "D2", "label": "Knowledge", "note": "recall / QA", "hue": "green"},
{"code": "D3", "label": "Coding", "note": "gen / repair", "hue": "slate"},
{"code": "D4", "label": "Safety", "note": "truth / harm", "hue": "rust"},
{"code": "D5", "label": "Multimodal", "note": "vision / A/V", "hue": "gold_p"},
]
DENSE_NOTES = [
"How to read. 5 capability domains × 4 sub-groups × 3 benchmarks · 60 leaves total. Hue identifies parent domain across every tier; count / tag on right of each node is illustrative size.",
"Structure. Extended benchmark inventory · adds Causal reasoning (D1), Domain-expert (D2), Multi-file coding (D3), Privacy (D4), and full Multimodal branch (D5).",
"Source. Extended snapshot inspired by HELM+ · Bloom taxonomy · 2026-09 planning cycle.",
]
# ═════════════════════════════════════════════════════════════════
# SPARSE · 3 × 2 × 1
# ═════════════════════════════════════════════════════════════════
SPARSE_ROOT = {"label": "LLM", "sub": "BENCH·S", "tag": "ROOT"}
SPARSE_DOMAINS = [
{
"key": "reason", "code": "D1", "label": "Reasoning",
"sub": "logic · math",
"hue": "magenta", "leaf_count": 2,
"subs": [
{"code": "S01", "label": "Mathematical", "sub": "arithmetic",
"leaves": [{"code": "L01", "label": "MATH-500", "tag": "500 q"}]},
{"code": "S02", "label": "Abstract logic", "sub": "grid",
"leaves": [{"code": "L02", "label": "ARC-AGI", "tag": "400 tk"}]},
],
},
{
"key": "know", "code": "D2", "label": "Knowledge",
"sub": "facts · common",
"hue": "green", "leaf_count": 2,
"subs": [
{"code": "S03", "label": "Academic QA", "sub": "multi-domain",
"leaves": [{"code": "L03", "label": "MMLU-Pro", "tag": "12k q"}]},
{"code": "S04", "label": "Commonsense", "sub": "WSC",
"leaves": [{"code": "L04", "label": "WinoGrande","tag": "44k pr"}]},
],
},
{
"key": "safe", "code": "D3", "label": "Safety",
"sub": "truth · robust",
"hue": "rust", "leaf_count": 2,
"subs": [
{"code": "S05", "label": "Truthfulness", "sub": "hallucination",
"leaves": [{"code": "L05", "label": "TruthfulQA","tag": "817 q"}]},
{"code": "S06", "label": "Robustness", "sub": "jailbreak",
"leaves": [{"code": "L06", "label": "JailbreakBench","tag": "100 pr"}]},
],
},
]
SPARSE_KPIS = [
{"kicker": "DOMAINS", "value": "3", "note": "compact cut", "hue": "magenta"},
{"kicker": "SUB-GROUPS", "value": "6", "note": "2 per domain", "hue": "green"},
{"kicker": "BENCHMARKS", "value": "6", "note": "one leaf per sub", "hue": "rust"},
{"kicker": "COVERAGE", "value": "60%", "note": "essentials only", "hue": "slate"},
]
SPARSE_COVERAGE = [
{"label": "Reasoning", "pct": 65, "note": "essentials only", "hue": "magenta"},
{"label": "Knowledge", "pct": 70, "note": "MMLU + commonsense", "hue": "green"},
{"label": "Safety", "pct": 55, "note": "truth + jailbreak sample", "hue": "rust"},
]
SPARSE_HUE_LEGEND = [
{"code": "D1", "label": "Reasoning", "note": "math / logic", "hue": "magenta"},
{"code": "D2", "label": "Knowledge", "note": "recall / QA", "hue": "green"},
{"code": "D3", "label": "Safety", "note": "truth / harm", "hue": "rust"},
]
SPARSE_NOTES = [
"How to read. Compact cut · 3 domains · 2 sub-groups each · 1 benchmark per sub. Suitable for exec briefing hero slot.",
"Structure. Only essential leaves retained per sub-group to keep the tree scannable at glance.",
"Source. Compact snapshot · HELM v2 · 2026-09 exec cut.",
]
# ═════════════════════════════════════════════════════════════════
# MINIMAL · 3 × 2 × 1 · no sidebar
# ═════════════════════════════════════════════════════════════════
MINIMAL_NOTES = [
"How to read. Minimal cut · 3 domains · 2 sub-groups · 1 benchmark each · sidebar stripped. Bare taxonomy for a one-shot deck.",
"Structure. No coverage bars · no hue legend · no tier markers · the tree fills the full canvas width.",
]
# ═════════════════════════════════════════════════════════════════
# runners
# ═════════════════════════════════════════════════════════════════
def _write(name: str, svg: str) -> None:
out_path = os.path.join(OUT_DIR, name)
with open(out_path, "w", encoding="utf-8") as f:
f.write(svg)
print(f"wrote {out_path} ({len(svg):,} bytes)")
def render_baseline() -> str:
return render_hero_embed_tx_taxonomy_v2(data=build_taxonomy_data())
def render_dense() -> str:
data = build_taxonomy_data(
root=DENSE_ROOT,
domains=DENSE_DOMAINS,
kpis=DENSE_KPIS,
coverage=DENSE_COVERAGE,
hue_legend=DENSE_HUE_LEGEND,
notes_lines=DENSE_NOTES,
figure_title="LLM Benchmark Taxonomy · extended inventory (5 domains · 60 benchmarks)",
figure_caption=(
"Hierarchical tree · 5 capability domains · 20 sub-groups · "
"60 benchmarks · extended HELM+ coverage"
),
source="Source · HELM+ · Bloom-style taxonomy · extended cut · 2026-09",
fig_tag_note="TX · Taxonomy · dense cut · 5×4×3 extended inventory",
)
return render_hero_embed_tx_taxonomy_v2(data=data)
def render_sparse() -> str:
data = build_taxonomy_data(
root=SPARSE_ROOT,
domains=SPARSE_DOMAINS,
kpis=SPARSE_KPIS,
coverage=SPARSE_COVERAGE,
hue_legend=SPARSE_HUE_LEGEND,
tier_markers=None, # keep baseline markers
notes_lines=SPARSE_NOTES,
figure_title="LLM Benchmark Taxonomy · compact trio",
figure_caption=(
"Hierarchical tree · 3 domains · 6 sub-groups · 6 benchmarks · "
"essentials-only cut"
),
source="Source · HELM v2 · compact cut · 2026-09 exec briefing",
fig_tag_note="TX · Taxonomy · sparse cut · 3×2×1 essentials",
)
return render_hero_embed_tx_taxonomy_v2(data=data)
def render_minimal() -> str:
data = build_taxonomy_data(
root=SPARSE_ROOT,
domains=SPARSE_DOMAINS,
kpis=SPARSE_KPIS,
coverage=[], # no coverage panel
hue_legend=[], # no hue legend
tier_markers=[], # no tier markers
show_sidebar=False,
notes_lines=MINIMAL_NOTES,
figure_title="LLM Benchmark Taxonomy · minimal · sidebar off",
figure_caption=(
"Hierarchical tree · 3 domains · 6 sub-groups · 6 benchmarks · "
"sidebar stripped · bare taxonomy hero"
),
source="Source · HELM v2 · minimal cut · 2026-09",
fig_tag_note="TX · Taxonomy · minimal cut · no sidebar · one-shot",
)
return render_hero_embed_tx_taxonomy_v2(data=data)
def main() -> None:
os.makedirs(OUT_DIR, exist_ok=True)
_write("baseline.svg", render_baseline())
_write("dense.svg", render_dense())
_write("sparse.svg", render_sparse())
_write("minimal.svg", render_minimal())
if __name__ == "__main__":
main()
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| slack-gif-creatoranthropics/skillsKnowledge and utilities for creating animated GIFs optimized for Slack. Provides constraints, validation tools, and animation concepts. Use when users request animated GIFs for Slack like "make me a GIF of X doing Y for Slack." | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |
| ponytailDietrichGebert/ponytailLazy senior dev mode for any coding task: the smallest change that fully solves it, and a reply a busy human understands in one read. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-debtDietrichGebert/ponytailHarvest every shortcut: comment into one debt ledger, so deferrals get tracked instead of forgotten. One-shot report. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-debtDietrichGebert/ponytailList every `shortcut:` comment (and older `ponytail:` ones) as a debt ledger. One-shot report, changes nothing. Use for "ponytail debt", "what did ponytail defer", "list the shortcuts", /ponytail-debt. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |