Update app.py

app.py

@@ -1,47 +1,81 @@
-import io, traceback, numpy as np, gradio as gr, matplotlib
+import io, json, traceback, numpy as np, matplotlib
 matplotlib.use("Agg")
+
 from PIL import Image
+import gradio as gr
 from transformers import AutoTokenizer, AutoModelForCausalLM
 from wfgy_sdk import get_engine
 from wfgy_sdk.evaluator import compare_logits, plot_histogram
-from tabulate import tabulate
 
-MODEL = "sshleifer/tiny-gpt2"
-tok   = AutoTokenizer.from_pretrained(MODEL)
-mdl   = AutoModelForCausalLM.from_pretrained(MODEL)
-eng   = get_engine()
+# model + engine
+MODEL_ID = "sshleifer/tiny-gpt2"
+tok  = AutoTokenizer.from_pretrained(MODEL_ID)
+mdl  = AutoModelForCausalLM.from_pretrained(MODEL_ID)
+eng  = get_engine()
+
+# runtime history
+hist = {"step": [], "var": [], "kl": []}
+
+# paper benchmark numbers
+paper = {
+    "benchmark": [
+        "MMLU", "GSM8K", "BBH", "MathBench",
+        "TruthfulQA", "XNLI", "MLQA", "LongBench",
+        "VQAv2", "OK-VQA"
+    ],
+    "baseline": [61.0, 78.0, 79.3, 72.2, 62.4, 59.5, 78.1, 51.4, 69.1, 65.7],
+    "wfgy":     [89.8, 98.7, 100.7, 87.4, 90.4, 77.3, 106.6, 69.6, 86.6, 86.8]
+}
 
 def run(prompt: str):
     prompt = prompt.strip()
     if not prompt:
-        return "", "", "", None, None
+        return "", "", "", "", None, update_history()
     try:
         ids  = tok(prompt, return_tensors="pt").input_ids
-        rawL = mdl(ids).logits[0, -1].detach().cpu().numpy()
+        raw  = mdl(ids).logits[0, -1].detach().cpu().numpy()
         G    = np.random.randn(256).astype(np.float32)
         I    = G + np.random.normal(scale=0.05, size=256).astype(np.float32)
-        modL = eng.run(I, G, rawL)
-
-        m = compare_logits(rawL, modL)
-        tbl = tabulate(
-            [[f"{m['std_ratio']:.3f}",
-              f"{m['var_drop']*100:4.1f} %",
-              f"{m['kl']:.3f}",
-              "✔" if m['top1'] else "✘"]],
-            headers=["std_ratio", "▼ var", "KL", "top-1"],
-            tablefmt="github")
-        headline = f"▼ var {m['var_drop']*100:4.1f} % | KL {m['kl']:.3f}"
-
-        fig = plot_histogram(rawL, modL)
+        mod  = eng.run(I, G, raw)
+        m    = compare_logits(raw, mod)
+
+        # update history
+        step = len(hist["step"]) + 1
+        hist["step"].append(step)
+        hist["var"].append(m["var_drop"] * 100)
+        hist["kl"].append(m["kl"])
+
+        headline = f"▼ var {m['var_drop']*100:4.1f}% | KL {m['kl']:.3f}"
+        metrics_md = (
+            "| metric | value |\n"
+            "|--------|-------|\n"
+            f"| std_ratio | {m['std_ratio']:.3f} |\n"
+            f"| var_drop  | {m['var_drop']*100:.1f}% |\n"
+            f"| KL        | {m['kl']:.3f} |\n"
+            f"| top-1 same| {'yes' if m['top1'] else 'no'} |"
+        )
+
+        fig = plot_histogram(raw, mod)
         buf = io.BytesIO(); fig.savefig(buf, format="png"); buf.seek(0)
         img = Image.open(buf)
 
-        raw_txt = prompt + tok.decode(int(rawL.argmax()))
-        mod_txt = prompt + tok.decode(int(modL.argmax()))
-        return raw_txt, mod_txt, headline, tbl, img
+        raw_txt = prompt + tok.decode(int(raw.argmax()))
+        mod_txt = prompt + tok.decode(int(mod.argmax()))
+        return raw_txt, mod_txt, headline, metrics_md, img, update_history()
     except Exception:
         tb = traceback.format_exc()
-        return "runtime error", tb, "runtime error", "", None
+        return "runtime error", tb, "runtime error", "", None, update_history()
+
+def update_history():
+    return {
+        "step": hist["step"],
+        "var":  hist["var"],
+        "kl":   hist["kl"]
+    }
+
+def clear_history():
+    hist["step"].clear(); hist["var"].clear(); hist["kl"].clear()
+    return update_history()
 
 with gr.Blocks(title="WFGY variance gate") as demo:
     gr.Markdown("# 🧠 WFGY simulation demo")
@@ -53,15 +87,29 @@ with gr.Blocks(title="WFGY variance gate") as demo:
         mod_box = gr.Textbox(label="After WFGY")
 
     headline = gr.Markdown()
-    metrics  = gr.Markdown()           #  ← 新增數值表
+    metrics  = gr.Markdown()
     img      = gr.Image(label="Logit histogram", type="pil")
 
-    btn.click(run, prompt,
-              [raw_box, mod_box, headline, metrics, img])
+    hist_plot = gr.LinePlot(
+        label="History (var% ↓ & KL)",
+        x="step", y=["var", "kl"],
+        overlay=True, height=250
+    )
+    clear_btn = gr.Button("Clear history")
 
-    gr.Markdown("---\n"
-                "### ⭐ 10 000 stars → unlock **WFGY 2.0** by 2025-08-01")
+    with gr.Accordion("Paper benchmarks", open=False):
+        bench_df  = gr.DataFrame(paper, interactive=False)
+        bench_bar = gr.BarPlot(paper, x="benchmark", y=["baseline", "wfgy"],
+                               overlay=False, height=300)
+
+    gr.Markdown(
+        "---\n"
+        "### ⭐ 10 000 GitHub stars before **2025-08-01** unlock **WFGY 2.0**"
+    )
+
+    btn.click(run, prompt,
+              [raw_box, mod_box, headline, metrics, img, hist_plot])
+    clear_btn.click(fn=clear_history, inputs=None, outputs=hist_plot)
 
 if __name__ == "__main__":
     demo.queue().launch()
-