Split project benchmark metrics

Author: tandede

app/globals.css

@@ -267,7 +267,7 @@ html.is-at-top .nav-contact:hover, html.is-at-top .nav-contact:focus-visible { b
 .workstream-card p { margin: 12px 0 0; color: #4b5565; font-size: 12px; line-height: 1.72; }
 
 .project-list { border-top: 1px solid var(--line-strong); }
-.project-row { position: relative; display: grid; grid-template-columns: 54px 280px minmax(0, 1fr) 330px; gap: 28px; align-items: start; padding: 32px 54px 32px 14px; border-bottom: 1px solid var(--line); transition: color 180ms ease, background 180ms ease; }
+.project-row { position: relative; display: grid; grid-template-columns: 54px 280px minmax(0, 1fr) 410px; gap: 28px; align-items: start; padding: 32px 54px 32px 14px; border-bottom: 1px solid var(--line); transition: color 180ms ease, background 180ms ease; }
 .project-row:hover { color: var(--blue); background: #f8faff; }
 .project-id { display: grid; }
 .project-id span { color: var(--blue); font-family: var(--font-geist-mono), monospace; font-size: 13px; font-weight: 700; }
@@ -276,7 +276,7 @@ html.is-at-top .nav-contact:hover, html.is-at-top .nav-contact:focus-visible { b
 .project-copy { display: grid; gap: 14px; }
 .project-copy p { margin: 0; color: #4b5565; font-size: 12px; line-height: 1.65; }
 .project-copy strong { display: block; margin-bottom: 4px; color: var(--ink); font-size: 10px; }
-.metric-grid { display: grid; grid-template-columns: repeat(3, minmax(0, 1fr)); border-left: 1px solid var(--line); }
+.metric-grid { display: grid; grid-template-columns: repeat(auto-fit, minmax(76px, 1fr)); border-left: 1px solid var(--line); }
 .metric-grid > div { display: grid; gap: 7px; align-content: center; min-height: 76px; padding: 10px 14px; border-right: 1px solid var(--line); }
 .metric-grid span { color: var(--muted); font-size: 8px; line-height: 1.4; }
 .metric-grid strong { color: var(--ink); font-family: var(--font-geist-mono), monospace; font-size: 16px; }

app/page.tsx

@@ -135,7 +135,7 @@ const projects = [
     overview: '面向数学、代码与科学推理任务，以 DeepSeek-R1-Distill-Qwen-32B 为底座构建领域增强模型。项目采用 Branch–Merge Distillation：各领域分别训练专家，再通过参数融合把专项能力合并到同一模型，降低多领域联合训练中的相互干扰。',
     technical: '训练链路覆盖领域 CoT 数据收集与质量清洗、三类 SFT 专家训练、统一离线评测以及融合权重搜索；通过分支训练保留各领域的推理模式，再在合并阶段平衡专项增益和通用能力。',
     work: '我参与领域 CoT 数据构建与清洗、数学/代码/科学专家训练、离线评测和融合权重选择，重点验证单领域增益是否真实，以及参数合并后是否出现能力回退。',
-    metrics: [['AIME24 / 25', '90.9 / 82.7'], ['GPQA / LCB', '69.4 / 70.4'], ['IFEval', '89.2']],
+    metrics: [['AIME24', '90.9'], ['AIME25', '82.7'], ['GPQA', '69.4'], ['LCB', '70.4'], ['IFEval', '89.2']],
   },
   {
     index: '02', title: 'Harness-Bench', href: 'https://www.harness-bench.ai/', subtitle: 'Agent Harness Evaluation',