Add sample_data.py

2026-06-02 00:44:52 +08:00
parent 9746adc05a
commit 103a5bd046
1 changed files with 65 additions and 0 deletions
--- a/sample_data.py
+++ b/sample_data.py
@@ -0,0 +1,65 @@
+# 示例数据：复现 Qwen3.6 vs Gemma4 benchmark 对比图
+# 用法: python benchmark_chart.py -c sample_data.py -o output.png
+
+TITLE = "AI Model Benchmark Comparison"
+
+COLORS = {
+    "qwen36": "#5B2D8E",
+    "qwen35": "#8B5CF6",
+    "gemma26": "#06B6D4",
+    "qwen27": "#B0B5BD",
+    "gemma31": "#34D399",
+}
+
+BG_COLOR = "#E8F5E9"
+BAR_WIDTH = 0.55
+NCOLS = 4
+
+# 快捷定义：减少重复代码
+QW36 = {"name": "Qwen3.6-35B-A3B", "short_label": "Qwen\n3.6", "group": "moe", "score": 0, "color_key": "qwen36"}
+QW35 = {"name": "Qwen3.5-35B-A3B", "short_label": "Qwen\n3.5", "group": "moe", "score": 0, "color_key": "qwen35"}
+GM26 = {"name": "Gemma4-26B-A4B",   "short_label": "Gemma4\n26B", "group": "moe", "score": 0, "color_key": "gemma26"}
+QW27 = {"name": "Qwen3.5-27B",      "short_label": "Qwen3.5\n27B", "group": "dense", "score": 0, "color_key": "qwen27"}
+GM31 = {"name": "Gemma4-31B",       "short_label": "Gemma4\n31B", "group": "dense", "score": 0, "color_key": "gemma31"}
+
+
+def mk(qw36, qw35, gm26, qw27, gm31):
+    """快捷构建 models 列表"""
+    return [
+        {**QW36, "score": qw36},
+        {**QW35, "score": qw35},
+        {**GM26, "score": gm26},
+        {**QW27, "score": qw27},
+        {**GM31, "score": gm31},
+    ]
+
+
+BENCHMARKS = [
+    # Row 1
+    {"title": "Terminal-Bench 2.0", "subtitle": "Agentic Terminal Coding",
+     "models": mk(51.5, 40.5, 34.2, 41.6, 42.9)},
+    {"title": "SWE-bench Pro", "subtitle": "Agentic Coding",
+     "models": mk(49.5, 44.6, 13.8, 51.2, 35.7)},
+    {"title": "SWE-bench Verified", "subtitle": "Agentic Coding",
+     "models": mk(73.4, 70.0, 17.4, 75.0, 52.0)},
+    {"title": "SWE-bench Multilingual", "subtitle": "Multilingual Agentic Coding",
+     "models": mk(67.2, 60.3, 17.3, 69.3, 51.7)},
+    # Row 2
+    {"title": "QwenClawBench", "subtitle": "Real-World Agent",
+     "models": mk(52.6, 47.7, 38.7, 52.2, 41.7)},
+    {"title": "QwenWebBench (Elo)", "subtitle": "Artifacts",
+     "models": mk(1397, 978, 1178, 1068, 1197)},
+    {"title": "NL2Repo", "subtitle": "Long-Horizon Coding",
+     "models": mk(29.4, 20.5, 11.6, 27.3, 15.5)},
+    {"title": "MCPMark", "subtitle": "General Agent",
+     "models": mk(37.0, 27.0, 14.2, 36.3, 18.1)},
+    # Row 3
+    {"title": "GPQA Diamond", "subtitle": "Graduate-level Reasoning",
+     "models": mk(86.0, 84.2, 82.3, 85.5, 84.3)},
+    {"title": "HMMT Feb 26", "subtitle": "Harvard-MIT Math Tournament",
+     "models": mk(83.6, 78.7, 79.0, 84.3, 77.2)},
+    {"title": "MMMU", "subtitle": "Multimodal Reasoning",
+     "models": mk(81.7, 81.4, 78.4, 82.3, 80.4)},
+    {"title": "RealWorldQA", "subtitle": "Image Reasoning",
+     "models": mk(85.3, 84.1, 72.2, 83.7, 72.3)},
+]