def build_tab_analyze():

with gr.Tab("📊 分析"):
gr.Markdown("""
**第二步：选择层范围，计算王氏五定律全指标**
层号 = safetensors key 中 `layers.{N}` 的原始 N，K=V 共享层自动处理。
""")

with gr.Row():
with gr.Column(scale=3):
model_id_input = gr.Textbox(
label="HuggingFace 模型 ID",
placeholder="google/gemma-4-e2b",
value="google/gemma-4-e2b"
)
token_input = gr.Textbox(
label="HF Access Token（公开模型可留空）",
type="password"
)
with gr.Row():
start_input = gr.Number(
label="起始层号（含）",
value=0, minimum=0, maximum=9999, precision=0
)
end_input = gr.Number(
label="结束层号（含）",
value=5, minimum=0, maximum=9999, precision=0
)
analyze_btn = gr.Button("🚀 开始分析", variant="primary")

with gr.Column(scale=1):
gr.Markdown(SIDEBAR_MD)

analyze_log = gr.Textbox(
label="分析日志（逐头详情）",
lines=35, max_lines=300
)
analyze_table = gr.Dataframe(
label="逐头全指标结果表",
headers=[
"prefix", "layer", "kv_head", "q_head", "kv_shared",
"pearson_QK", "spearman_QK", "pearson_QV", "pearson_KV",
"ssr_QK", "ssr_QV", "ssr_KV",
"cosU_QK", "cosU_QV", "cosU_KV",
"cosV_QK", "cosV_QV", "cosV_KV",
"alpha_QK", "alpha_QV", "alpha_KV",
"alpha_res_QK", "alpha_res_QV", "alpha_res_KV",
"sigma_max_Q", "sigma_min_Q",
"sigma_max_K", "sigma_min_K",
"sigma_max_V", "sigma_min_V",
"cond_Q", "cond_K", "cond_V",
"head_dim", "d_model", "n_q_heads", "n_kv_heads",
]
)

analyze_btn.click(
fn=run_analysis,
inputs=[model_id_input, token_input, start_input, end_input],
outputs=[analyze_log, analyze_table]
)

app.py

@@ -11,7 +11,7 @@ from ui.tab_analyze import build_tab_analyze
 
 with gr.Blocks(
     title="Wang's Five Laws — LLM Spectral Analyzer",
-    theme=gr.themes.Soft()
+    # theme=gr.themes.Soft()
 ) as demo:
 
     gr.Markdown("""

core/metrics.py

@@ -1,32 +1,21 @@
 # core/metrics.py
-"""
-王氏五定律全部指标计算
-输入：Q/K/V weight tensors
-输出：结构化指标字典
-"""
-
 import torch
 import numpy as np
 from scipy.stats import spearmanr
 from core.layer_profile import LayerProfile
 
 
-# ─────────────────────────────────────────────
-# 基础指标
-# ─────────────────────────────────────────────
-
 def pearson(a: torch.Tensor, b: torch.Tensor) -> float:
     am, bm = a - a.mean(), b - b.mean()
     den = torch.norm(am) * torch.norm(bm)
     return float(torch.dot(am, bm) / den) if den > 1e-10 else 0.0
 
 
-def spearman(a: torch.Tensor, b: torch.Tensor) -> float:
+def spearman_r(a: torch.Tensor, b: torch.Tensor) -> float:
     return float(spearmanr(a.numpy(), b.numpy())[0])
 
 
 def ssr(a: torch.Tensor, b: torch.Tensor) -> float:
-    """Spectral Shape Residual（第二定律核心）"""
     n  = min(a.shape[0], b.shape[0])
     an = a[:n] / (torch.norm(a[:n]) + 1e-10)
     bn = b[:n] / (torch.norm(b[:n]) + 1e-10)
@@ -34,38 +23,48 @@ def ssr(a: torch.Tensor, b: torch.Tensor) -> float:
 
 
 def svr(a: torch.Tensor, b: torch.Tensor) -> tuple[float, float]:
-    """Singular Value Ratio：alpha 和残差"""
-    n  = min(a.shape[0], b.shape[0])
-    sa, sb = a[:n], b[:n]
-    den = torch.dot(sb, sb)
+    """
+    最小二乘法拟合：alpha = argmin ||s_a - alpha * s_b||^2
+    返回 (alpha, residual)
+    residual = mean((s_a - alpha * s_b)^2)
+    """
+    n       = min(a.shape[0], b.shape[0])
+    sa, sb  = a[:n], b[:n]
+    den     = torch.dot(sb, sb)
     if den < 1e-10:
         return 1.0, 0.0
-    alpha = torch.dot(sa, sb) / den
-    res   = float(torch.mean((sa - alpha * sb) ** 2))
-    return float(alpha), res
+    alpha   = torch.dot(sa, sb) / den
+    residual= float(torch.mean((sa - alpha * sb) ** 2))
+    return float(alpha), residual
 
 
 def cos_U(U_a: torch.Tensor, U_b: torch.Tensor) -> float:
-    """左奇异向量对齐（第四定律）"""
-    r = min(U_a.shape[0], U_b.shape[0])
-    c = min(U_a.shape[1], U_b.shape[1])
+    r  = min(U_a.shape[0], U_b.shape[0])
+    c  = min(U_a.shape[1], U_b.shape[1])
     Ua = U_a[:r, :c] / (torch.norm(U_a[:r, :c], dim=0, keepdim=True) + 1e-10)
     Ub = U_b[:r, :c] / (torch.norm(U_b[:r, :c], dim=0, keepdim=True) + 1e-10)
     return float(torch.diag(torch.abs(Ua.T @ Ub)).mean())
 
 
 def cos_V(Vt_a: torch.Tensor, Vt_b: torch.Tensor) -> float:
-    """右奇异向量对齐（第五定律）"""
-    r = min(Vt_a.shape[0], Vt_b.shape[0])
-    c = min(Vt_a.shape[1], Vt_b.shape[1])
+    r  = min(Vt_a.shape[0], Vt_b.shape[0])
+    c  = min(Vt_a.shape[1], Vt_b.shape[1])
     Va = Vt_a[:r, :c] / (torch.norm(Vt_a[:r, :c], dim=1, keepdim=True) + 1e-10)
     Vb = Vt_b[:r, :c] / (torch.norm(Vt_b[:r, :c], dim=1, keepdim=True) + 1e-10)
     return float(torch.abs((Va * Vb).sum(dim=1)).mean())
 
 
-# ─────────────────────────────────────────────
-# 逐头分析
-# ─────────────────────────────────────────────
+def sigma_stats(s: torch.Tensor) -> tuple[float, float, float]:
+    """
+    返回 (sigma_max, sigma_min, cond)
+    sigma_min 过滤接近零的奇异值，避免条件数虚高
+    """
+    s_max  = float(s.max())
+    valid  = s[s > 1e-10]
+    s_min  = float(valid.min()) if valid.numel() > 0 else 0.0
+    cond   = s_max / (s_min + 1e-10)
+    return s_max, s_min, cond
+
 
 def analyze_layer(
     W_q:     torch.Tensor,
@@ -73,12 +72,7 @@ def analyze_layer(
     W_v:     torch.Tensor,
     profile: LayerProfile,
 ) -> tuple[list[dict], str]:
-    """
-    对一个 LayerProfile 做逐头全指标分析。
-    返回 (records列表, 日志字符串)
 
-    K=V 共享层：KV 指标直接填理论值，不重复计算
-    """
     n_q       = profile.n_q_heads
     n_kv      = profile.n_kv_heads
     d_head    = profile.head_dim
@@ -88,7 +82,6 @@ def analyze_layer(
     records: list[dict] = []
     lines:   list[str]  = []
 
-    # 日志头
     kv_tag = " [K=V共享]" if kv_shared else ""
     lines.append(
         f"\n{'─'*80}\n"
@@ -111,9 +104,11 @@ def analyze_layer(
         U_k, s_k, Vt_k = torch.linalg.svd(k_t, full_matrices=False)
         U_v, s_v, Vt_v = torch.linalg.svd(v_t, full_matrices=False)
 
+        smxk, smnk, cond_k = sigma_stats(s_k)
+        smxv, smnv, cond_v = sigma_stats(s_v)
+
         # KV 指标
         if kv_shared:
-            # W_v = W_k → 理论值
             ssr_kv   = 0.0
             pkv      = 1.0
             cosU_KV  = 1.0
@@ -121,11 +116,9 @@ def analyze_layer(
             alpha_kv = 1.0
             res_kv   = 0.0
         else:
+            n_kv_sv  = min(len(s_k), len(s_v))
             ssr_kv   = ssr(s_k, s_v)
-            pkv      = pearson(
-                s_k[:min(len(s_k), len(s_v))],
-                s_v[:min(len(s_k), len(s_v))]
-            )
+            pkv      = pearson(s_k[:n_kv_sv], s_v[:n_kv_sv])
             cosU_KV  = cos_U(U_k, U_v)
             cosV_KV  = cos_V(Vt_k, Vt_v)
             alpha_kv, res_kv = svr(s_k, s_v)
@@ -135,81 +128,72 @@ def analyze_layer(
             q_t = W_q[h * d_head:(h + 1) * d_head, :]
             U_q, s_q, Vt_q = torch.linalg.svd(q_t, full_matrices=False)
 
+            smxq, smnq, cond_q = sigma_stats(s_q)
+
             nqk = min(len(s_q), len(s_k))
             nqv = min(len(s_q), len(s_v))
 
-            # QK 指标
-            pqk    = pearson(s_q[:nqk], s_k[:nqk])
-            spqk   = spearman(s_q[:nqk], s_k[:nqk])
-            ssr_qk = ssr(s_q, s_k)
+            # QK
+            pqk        = pearson(s_q[:nqk], s_k[:nqk])
+            spqk       = spearman_r(s_q[:nqk], s_k[:nqk])
+            ssr_qk     = ssr(s_q, s_k)
             a_qk, r_qk = svr(s_q, s_k)
-            cU_QK  = cos_U(U_q, U_k)
-            cV_QK  = cos_V(Vt_q, Vt_k)
+            cU_QK      = cos_U(U_q, U_k)
+            cV_QK      = cos_V(Vt_q, Vt_k)
 
-            # QV 指标
-            pqv    = pearson(s_q[:nqv], s_v[:nqv])
-            ssr_qv = ssr(s_q, s_v)
+            # QV
+            pqv        = pearson(s_q[:nqv], s_v[:nqv])
+            ssr_qv     = ssr(s_q, s_v)
             a_qv, r_qv = svr(s_q, s_v)
-            cU_QV  = cos_U(U_q, U_v)
-            cV_QV  = cos_V(Vt_q, Vt_v)
-
-            # 奇异值范围
-            smxq = float(s_q.max())
-            smnq = float(s_q[s_q > 1e-10].min()) if (s_q > 1e-10).any() else 0.
-            smxk = float(s_k.max())
-            smnk = float(s_k[s_k > 1e-10].min()) if (s_k > 1e-10).any() else 0.
-            smxv = float(s_v.max())
-            smnv = float(s_v[s_v > 1e-10].min()) if (s_v > 1e-10).any() else 0.
-
-            rec = {
-                # 位置
-                "prefix":       profile.prefix,
-                "layer":        profile.layer_idx,
-                "kv_head":      kv_h,
-                "q_head":       h,
-                "kv_shared":    kv_shared,
+            cU_QV      = cos_U(U_q, U_v)
+            cV_QV      = cos_V(Vt_q, Vt_v)
+
+            records.append({
+                "prefix":        profile.prefix,
+                "layer":         profile.layer_idx,
+                "kv_head":       kv_h,
+                "q_head":        h,
+                "kv_shared":     kv_shared,
                 # 第一定律
-                "pearson_QK":   round(pqk,   6),
-                "spearman_QK":  round(spqk,  6),
-                "pearson_QV":   round(pqv,   6),
-                "pearson_KV":   round(pkv,   6),
+                "pearson_QK":    round(pqk,    6),
+                "spearman_QK":   round(spqk,   6),
+                "pearson_QV":    round(pqv,    6),
+                "pearson_KV":    round(pkv,    6),
                 # 第二定律
-                "ssr_QK":       round(ssr_qk, 8),
-                "ssr_QV":       round(ssr_qv, 8),
-                "ssr_KV":       round(ssr_kv, 8),
+                "ssr_QK":        round(ssr_qk,  8),
+                "ssr_QV":        round(ssr_qv,  8),
+                "ssr_KV":        round(ssr_kv,  8),
                 # 第四定律
-                "cosU_QK":      round(cU_QK,  6),
-                "cosU_QV":      round(cU_QV,  6),
-                "cosU_KV":      round(cosU_KV,6),
+                "cosU_QK":       round(cU_QK,   6),
+                "cosU_QV":       round(cU_QV,   6),
+                "cosU_KV":       round(cosU_KV, 6),
                 # 第五定律
-                "cosV_QK":      round(cV_QK,  6),
-                "cosV_QV":      round(cV_QV,  6),
-                "cosV_KV":      round(cosV_KV,6),
-                # 尺度
-                "alpha_QK":     round(a_qk,   4),
-                "alpha_QV":     round(a_qv,   4),
-                "alpha_KV":     round(alpha_kv,4),
-                "alpha_res_QK": round(r_qk,   6),
-                "alpha_res_QV": round(r_qv,   6),
-                "alpha_res_KV": round(res_kv, 6),
-                # 奇异值范围
-                "sigma_max_Q":  round(smxq, 4),
-                "sigma_min_Q":  round(smnq, 4),
-                "sigma_max_K":  round(smxk, 4),
-                "sigma_min_K":  round(smnk, 4),
-                "sigma_max_V":  round(smxv, 4),
-                "sigma_min_V":  round(smnv, 4),
-                # 第三定律
-                "cond_Q":       round(smxq / (smnq + 1e-10), 2),
-                "cond_K":       round(smxk / (smnk + 1e-10), 2),
-                "cond_V":       round(smxv / (smnv + 1e-10), 2),
+                "cosV_QK":       round(cV_QK,   6),
+                "cosV_QV":       round(cV_QV,   6),
+                "cosV_KV":       round(cosV_KV, 6),
+                # 尺度因子 + 最小二乘残差
+                "alpha_QK":      round(a_qk,    4),
+                "alpha_QV":      round(a_qv,    4),
+                "alpha_KV":      round(alpha_kv,4),
+                "alpha_res_QK":  round(r_qk,    6),
+                "alpha_res_QV":  round(r_qv,    6),
+                "alpha_res_KV":  round(res_kv,  6),
+                # 第三定律：奇异值范围 + 条件数
+                "sigma_max_Q":   round(smxq, 4),
+                "sigma_min_Q":   round(smnq, 4),
+                "sigma_max_K":   round(smxk, 4),
+                "sigma_min_K":   round(smnk, 4),
+                "sigma_max_V":   round(smxv, 4),
+                "sigma_min_V":   round(smnv, 4),
+                "cond_Q":        round(cond_q, 2),
+                "cond_K":        round(cond_k, 2),
+                "cond_V":        round(cond_v, 2),
                 # 维度信息
-                "head_dim":     d_head,
-                "d_model":      profile.d_model,
-                "n_q_heads":    n_q,
-                "n_kv_heads":   n_kv,
-            }
-            records.append(rec)
+                "head_dim":      d_head,
+                "d_model":       profile.d_model,
+                "n_q_heads":     n_q,
+                "n_kv_heads":    n_kv,
+            })
 
             lines.append(
                 f"  {kv_h:>3d} {h:>3d} │"
@@ -223,12 +207,7 @@ def analyze_layer(
     return records, "".join(lines)
 
 
-# ─────────────────────────────────────────────
-# 全局汇总统计
-# ─────────────────────────────────────────────
-
 def summarize_records(records: list[dict], model_id: str) -> str:
-    """生成全局汇总文本"""
     if not records:
         return "❌ 无记录\n"
 
@@ -253,36 +232,40 @@ def summarize_records(records: list[dict], model_id: str) -> str:
     ]
 
     for pfx in sorted(df["prefix"].unique()):
-        pdf = df[df["prefix"] == pfx]
-        # 排除 kv_shared 的 KV 指标（理论值，不参与统计）
-        real_kv = pdf[~pdf["kv_shared"]]
+        pdf      = df[df["prefix"] == pfx]
+        real_kv  = pdf[~pdf["kv_shared"]]
+        kv_df    = real_kv if len(real_kv) > 0 else pdf
 
         lines.append(
             f"\n▶ {pfx}\n"
             f"  记录：{len(pdf)} 条，"
             f"层：{sorted(pdf['layer'].unique())}\n"
         )
+        if pdf["kv_shared"].any():
+            n_shared = pdf[pdf["kv_shared"]]["layer"].nunique()
+            lines.append(f"  ⚠️  含 {n_shared} 个 K=V共享层，KV指标为理论值\n")
+
         lines += [
             "  【第一定律 Pearson r → 1】\n",
-            stat(pdf["pearson_QK"].values,  "Q-K:"),
-            stat(pdf["pearson_QV"].values,  "Q-V:"),
-            stat(real_kv["pearson_KV"].values if len(real_kv) else pdf["pearson_KV"].values, "K-V:"),
+            stat(pdf["pearson_QK"].values, "Q-K:"),
+            stat(pdf["pearson_QV"].values, "Q-V:"),
+            stat(kv_df["pearson_KV"].values, "K-V(实):"),
             "  【第二定律 SSR → 0】\n",
-            stat(pdf["ssr_QK"].values,  "Q-K:"),
-            stat(pdf["ssr_QV"].values,  "Q-V:"),
-            stat(real_kv["ssr_KV"].values if len(real_kv) else pdf["ssr_KV"].values, "K-V:"),
+            stat(pdf["ssr_QK"].values, "Q-K:"),
+            stat(pdf["ssr_QV"].values, "Q-V:"),
+            stat(kv_df["ssr_KV"].values, "K-V(实):"),
             "  【第四定律 cosU 输出子空间】\n",
-            stat(pdf["cosU_QK"].values,  "cosU Q-K:"),
-            stat(pdf["cosU_QV"].values,  "cosU Q-V:"),
-            stat(real_kv["cosU_KV"].values if len(real_kv) else pdf["cosU_KV"].values, "cosU K-V:"),
+            stat(pdf["cosU_QK"].values, "cosU Q-K:"),
+            stat(pdf["cosU_QV"].values, "cosU Q-V:"),
+            stat(kv_df["cosU_KV"].values, "cosU K-V:"),
             "  【第五定律 cosV 输入子空间】\n",
-            stat(pdf["cosV_QK"].values,  "cosV Q-K:"),
-            stat(pdf["cosV_QV"].values,  "cosV Q-V:"),
-            stat(real_kv["cosV_KV"].values if len(real_kv) else pdf["cosV_KV"].values, "cosV K-V:"),
-            "  【第三定律 条件数】\n",
-            stat(pdf["cond_Q"].values,  "cond Q:"),
-            stat(pdf["cond_K"].values,  "cond K:"),
-            stat(pdf["cond_V"].values,  "cond V:"),
+            stat(pdf["cosV_QK"].values, "cosV Q-K:"),
+            stat(pdf["cosV_QV"].values, "cosV Q-V:"),
+            stat(kv_df["cosV_KV"].values, "cosV K-V:"),
+            "  【第三定律 条件数（sigma_min 已过滤零值）】\n",
+            stat(pdf["cond_Q"].values, "cond Q:"),
+            stat(pdf["cond_K"].values, "cond K:"),
+            stat(pdf["cond_V"].values, "cond V:"),
         ]
 
     lines.append(

ui/tab_analyze.py

@@ -26,6 +26,39 @@ from core.layer_profile import (
 from core.metrics import analyze_layer, summarize_records
 
 
+SIDEBAR_MD = """
+### ✅ 推荐模型
+google/gemma-4-e2b
+google/gemma-4-e4b-it
+google/gemma-4-31b-it
+Qwen/Qwen2.5-14B-Instruct
+deepseek-ai/DeepSeek-R1-Distill-Qwen-14B
+meta-llama/Meta-Llama-3-8B
+
+
+### 层号说明
+- 层号 = safetensors key 中 `layers.{N}` 的 **N**
+- **不按组件重排**，原始值直接输出
+- 混合模态模型（如 Gemma-4）：
+  - `layers.0~11` 同时含 audio/vision/text 层
+  - 全部输出，按前缀区分组件
+
+### 示例：Gemma-4-E2B
+| 组件 | 层范围 |
+|------|--------|
+| audio_tower | 0~11 |
+| language_model | 0~34 |
+| vision_tower | 0~15 |
+
+### 示例：Gemma-4-31B
+| 组件 | 层范围 |
+|------|--------|
+| language(局部层) | 0~59 |
+| language(全局层) | 5,11,17...59 |
+| vision_tower | 0~26 |
+"""
+
+
 def run_analysis(
     model_id:    str,
     hf_token:    str,
@@ -186,16 +219,11 @@ def run_analysis(
 # Tab2 UI 组件
 # ─────────────────────────────────────────────
 
-def build_tab_analyze(
-    shared_model_id: gr.Textbox = None,
-    shared_token:    gr.Textbox = None,
-):
+def build_tab_analyze():
     with gr.Tab("📊 分析"):
         gr.Markdown("""
         **第二步：选择层范围，计算王氏五定律全指标**
-        - 层号 = safetensors key 中 `layers.{N}` 的原始 N
-        - 所有组件（language/vision/audio）同时分析
-        - K=V 共享层自动标注，KV 指标填理论值
+        层号 = safetensors key 中 `layers.{N}` 的原始 N，K=V 共享层自动处理。
         """)
 
         with gr.Row():
@@ -205,37 +233,23 @@ def build_tab_analyze(
                     placeholder="google/gemma-4-e2b",
                     value="google/gemma-4-e2b"
                 )
-            with gr.Column(scale=2):
                 token_input = gr.Textbox(
-                    label="HF Access Token",
+                    label="HF Access Token（公开模型可留空）",
                     type="password"
                 )
-
-        with gr.Row():
-            start_input = gr.Number(
-                label="起始层号（含）",
-                value=0, minimum=0, maximum=9999, precision=0
-            )
-            end_input = gr.Number(
-                label="结束层号（含）",
-                value=5, minimum=0, maximum=9999, precision=0
-            )
-            analyze_btn = gr.Button(
-                "🚀 开始分析", variant="primary", size="lg"
-            )
-
-        with gr.Row():
-            gr.Markdown("""
-            ### 层号参考
-            | 模型 | 组件 | 层范围 |
-            |------|------|--------|
-            | Gemma-4-E2B | language | 0~34 |
-            | Gemma-4-E2B | vision | 0~15 |
-            | Gemma-4-31B | language(local) | 0~59（非5的倍数+5） |
-            | Gemma-4-31B | language(global) | 5,11,17...59 |
-            | Qwen2.5-14B | language | 0~47 |
-            | LLaMA-3-8B | language | 0~31 |
-            """)
+                with gr.Row():
+                    start_input = gr.Number(
+                        label="起始层号（含）",
+                        value=0, minimum=0, maximum=9999, precision=0
+                    )
+                    end_input = gr.Number(
+                        label="结束层号（含）",
+                        value=5, minimum=0, maximum=9999, precision=0
+                    )
+                analyze_btn = gr.Button("🚀 开始分析", variant="primary")
+
+            with gr.Column(scale=1):
+                gr.Markdown(SIDEBAR_MD)
 
         analyze_log = gr.Textbox(
             label="分析日志（逐头详情）",

ui/tab_inspect.py

@@ -22,6 +22,38 @@ from core.layer_profile import (
     extract_config_params,
 )
 
+SIDEBAR_MD = """
+### ✅ 推荐模型
+google/gemma-4-e2b
+google/gemma-4-e4b-it
+google/gemma-4-31b-it
+Qwen/Qwen2.5-14B-Instruct
+deepseek-ai/DeepSeek-R1-Distill-Qwen-14B
+meta-llama/Meta-Llama-3-8B
+
+
+### 层号说明
+- 层号 = safetensors key 中 `layers.{N}` 的 **N**
+- **不按组件重排**，原始值直接输出
+- 混合模态模型（如 Gemma-4）：
+  - `layers.0~11` 同时含 audio/vision/text 层
+  - 全部输出，按前缀区分组件
+
+### 示例：Gemma-4-E2B
+| 组件 | 层范围 |
+|------|--------|
+| audio_tower | 0~11 |
+| language_model | 0~34 |
+| vision_tower | 0~15 |
+
+### 示例：Gemma-4-31B
+| 组件 | 层范围 |
+|------|--------|
+| language(局部层) | 0~59 |
+| language(全局层) | 5,11,17...59 |
+| vision_tower | 0~26 |
+"""
+
 
 def inspect_model(
     model_id: str,
@@ -134,11 +166,8 @@ def inspect_model(
 def build_tab_inspect():
     with gr.Tab("🔬 结构探测"):
         gr.Markdown("""
-        **第一步：先探测模型结构**
-        - 自动识别所有组件（language/vision/audio）
-        - 自动推断 head_dim（支持异构层，如 Gemma-4-31B 局部/全局层）
-        - 自动检测 K=V 共享层
-        - 结果供「分析」Tab 使用
+        **第一步：先探测模型结构**，自动识别组件、head_dim、K=V共享层。
+        结果供「分析」Tab 使用。
         """)
 
         with gr.Row():
@@ -148,15 +177,14 @@ def build_tab_inspect():
                     placeholder="google/gemma-4-e2b",
                     value="google/gemma-4-e2b"
                 )
-            with gr.Column(scale=2):
                 inspect_token = gr.Textbox(
                     label="HF Access Token（公开模型可留空）",
                     type="password"
                 )
+                inspect_btn = gr.Button("🔍 探测结构", variant="secondary")
+
             with gr.Column(scale=1):
-                inspect_btn = gr.Button(
-                    "🔍 探测结构", variant="secondary", size="lg"
-                )
+                gr.Markdown(SIDEBAR_MD)
 
         inspect_log = gr.Textbox(
             label="结构探测日志",