.

meta_ui.py

@@ -42,11 +42,16 @@ def consume_logs(dummy=None):
 
 
 # ───────────────────── 后端调用 ─────────────────────
-def backend(text, sampling):
+def backend(text, sampling, api_suffix):
+    url = f"http://localhost:30000{api_suffix}"
+    if api_suffix == "/generate":
         payload = {"model": MODEL_NAME, "text": text, "sampling_params": sampling}
-    log(f"\n🟡 [{now()}] payload\n{json.dumps(payload, ensure_ascii=False, indent=2)}")
+    else:  # "/v1/completion"
+        payload = {"model": MODEL_NAME, "prompt": text, **sampling}
+
+    log(f"\n🟡 [{now()}] POST {url}\n{json.dumps(payload, ensure_ascii=False, indent=2)}")
     try:
-        r = requests.post(API_URL,
+        r = requests.post(url,
                           headers={"Authorization": f"Bearer {API_KEY}",
                                    "Content-Type": "application/json"},
                           json=payload, timeout=180)
@@ -54,30 +59,38 @@ def backend(text, sampling):
             data = r.json()
         except Exception:
             data = {}
-        fr   = data.get("meta_info", {}).get("finish_reason")
+        if api_suffix == "/generate":
-        ctok = data.get("meta_info", {}).get("completion_tokens")
+            txt = data.get("text", "").strip()
+            meta = data.get("meta_info", {})
+            fr = meta.get("finish_reason")
+            ctok = meta.get("completion_tokens")
+        else:
+            choice = data.get("choices", [{}])[0]
+            txt = choice.get("text", "").strip()
+            fr = choice.get("finish_reason")
+            ctok = data.get("usage", {}).get("completion_tokens")
         log(f"🟢 [{now()}] HTTP {r.status_code}  tokens={ctok}  finish={fr}\n"
             f"🟢 resp800={r.text[:800]!r}")
         if r.status_code != 200:
             return f"[HTTP {r.status_code}] {r.text[:300]}"
-        return data.get("text", "").strip() or "[⚠ 空]"
+        return txt or "[⚠ 空]"
     except Exception as e:
         log(f"[❌ 请求异常] {e}")
         return f"[❌ 请求异常] {e}"
 
+
 # ───────────────────── Chat 回调 ─────────────────────
 def chat(
     user, history,
     max_new, temp, top_p, top_k,
     rep_pen, pres_pen, stop_raw,
-    log_state
+    api_suffix, log_state
 ):
-    import threading
     from queue import Queue, Empty
 
     stop = [s.strip() for s in stop_raw.split(",") if s.strip()] or None
     samp = {
-        "max_new_tokens": int(max_new),
+        ("max_tokens" if api_suffix == "/v1/completion" else "max_new_tokens"): int(max_new),
         "temperature": temp,
         "top_p": top_p,
         "top_k": int(top_k),
@@ -86,21 +99,17 @@ def chat(
         **({"stop": stop} if stop else {})
     }
 
+
     result_q = Queue()
 
-    # 后台线程执行 backend 推理
     def worker():
-        out = backend(user, samp)
+        out = backend(user, samp, api_suffix)
         result_q.put(out)
 
-    thread = threading.Thread(target=worker)
+    threading.Thread(target=worker).start()
-    thread.start()
-
-    # 先返回提示
     yield "⏳ 正在生成中...", log_state
 
-    # 每 0.1 秒轮询结果队列（避免阻塞 UI）
+    while True:
-    while thread.is_alive() or not result_q.empty():
         try:
             result = result_q.get(timeout=0.1)
             yield result, log_state
@@ -112,6 +121,9 @@ def chat(
 with gr.Blocks(title="调试界面") as demo:
     gr.Markdown(f"## 💬 调试界面  \n权重 **{MODEL_PATH.name}**")
 
+    with gr.Row():
+        api_choice = gr.Dropdown(choices=["/generate", "/v1/completion"],
+                                value="/generate", label="选择推理接口")
     # 采样参数控件
     with gr.Row():
         max_new = gr.Slider(32, 32768, 128, label="max_new_tokens")
@@ -132,12 +144,14 @@ with gr.Blocks(title="调试界面") as demo:
     chatbot = gr.ChatInterface(
         fn=chat,
         additional_inputs=[max_new, temp, top_p, top_k,
-                           rep_pen, pres_pen, stop_txt, log_state],
+                        rep_pen, pres_pen, stop_txt,
+                        api_choice, log_state],  # ✅ 加入 dropdown
         additional_outputs=[log_state],
         type="messages"
     )
 
 
+
     # 日志刷新定时器
     timer = gr.Timer(1.0, render=True)
     timer.tick(

meta_ui_old.py

@@ -20,16 +20,26 @@ now = lambda: datetime.datetime.now().strftime("%H:%M:%S")
 
 # ───────────────────── 日志队列 ─────────────────────
 LOG_Q: "queue.Queue[str]" = queue.Queue()
+LOG_TXT = ""  # ✅ 全局日志缓存，避免 chat 焦点阻断 log_box 更新
+
 def log(msg):                 # 写终端 + 推队列
     print(msg, flush=True)
     LOG_Q.put(msg)
 
-def consume_logs(state_txt: str):
+prev_log_value = ""  # 上一帧的日志内容
-    """供 Interval 调用：把队列里所有新行取出拼接到 state"""
+
-    buf = deque(state_txt.splitlines(), maxlen=400)
+def consume_logs(dummy=None):
+    """每秒更新 log_box 内容，避免 chat 阻塞 UI 刷新"""
+    global LOG_TXT, prev_log_value
+    buf = deque(LOG_TXT.splitlines(), maxlen=400)
     while not LOG_Q.empty():
         buf.append(LOG_Q.get())
-    return "\n".join(buf)
+    LOG_TXT = "\n".join(buf)
+    if LOG_TXT != prev_log_value:
+        prev_log_value = LOG_TXT
+        return gr.update(value=LOG_TXT)
+    return gr.update()  # 无更新则不触发前端刷新
+
 
 # ───────────────────── 后端调用 ─────────────────────
 def backend(text, sampling):
@@ -62,6 +72,9 @@ def chat(
     rep_pen, pres_pen, stop_raw,
     log_state
 ):
+    import threading
+    from queue import Queue, Empty
+
     stop = [s.strip() for s in stop_raw.split(",") if s.strip()] or None
     samp = {
         "max_new_tokens": int(max_new),
@@ -72,9 +85,28 @@ def chat(
         "presence_penalty": pres_pen,
         **({"stop": stop} if stop else {})
     }
+
+    result_q = Queue()
+
+    # 后台线程执行 backend 推理
+    def worker():
         out = backend(user, samp)
-    # 返回回答，同时把 log_state 原样带回（不刷新由 Interval 处理）
+        result_q.put(out)
-    return out, log_state
+
+    thread = threading.Thread(target=worker)
+    thread.start()
+
+    # 先返回提示
+    yield "⏳ 正在生成中...", log_state
+
+    # 每 0.1 秒轮询结果队列（避免阻塞 UI）
+    while thread.is_alive() or not result_q.empty():
+        try:
+            result = result_q.get(timeout=0.1)
+            yield result, log_state
+        except Empty:
+            continue
+
 
 # ───────────────────── Gradio UI ─────────────────────
 with gr.Blocks(title="调试界面") as demo:
@@ -82,7 +114,7 @@ with gr.Blocks(title="调试界面") as demo:
 
     # 采样参数控件
     with gr.Row():
-        max_new = gr.Slider(32, 32768, 2048, label="max_new_tokens")
+        max_new = gr.Slider(32, 32768, 128, label="max_new_tokens")
         temp    = gr.Slider(0, 1.5, 0.8, step=0.05, label="temperature")
     with gr.Row():
         top_p   = gr.Slider(0, 1, 0.95, step=0.01, label="top_p")
@@ -92,31 +124,30 @@ with gr.Blocks(title="调试界面") as demo:
         pres_pen= gr.Slider(0, 2, 0.0, step=0.05, label="presence_penalty")
     stop_txt = gr.Textbox("", label="stop 序列（逗号分隔）")
 
-    dbg_chk  = gr.Checkbox(label="📜 显示 Debug 面板", value=True)
+    log_state = gr.State("")  # 状态透传
-    log_box  = gr.Textbox(label="实时日志", lines=20, interactive=False, visible=True)
+    dbg_chk   = gr.Checkbox(label="📜 显示 Debug 面板", value=False)  # ✅ 默认关闭
-    log_state= gr.State("")   # 保存全部日志文本
+    log_box   = gr.Textbox(label="实时日志", lines=20, interactive=False, visible=False)  # ✅ 默认隐藏
 
-    # ────────────── 定时刷新日志 ──────────────
+    # Chat 界面（移到日志之前）
-    timer = gr.Timer(1.0, render=False)       # 每秒触发一次
-    timer.tick(                               # ⬅ 关键是用 .tick
-        fn=consume_logs,                      # 把队列里的新行合并
-        inputs=[log_box],                     # 直接取 Textbox 当前内容
-        outputs=[log_box],                    # 写回同一个 Textbox
-    )   
-
-    # 显示到 log_box
-    log_state.change(lambda txt: gr.update(value=txt), log_state, log_box)
-
-    # Debug 面板可见性切换
-    dbg_chk.change(lambda v: gr.update(visible=v), dbg_chk, log_box)
-
-    # Chatbot
     chatbot = gr.ChatInterface(
         fn=chat,
         additional_inputs=[max_new, temp, top_p, top_k,
                            rep_pen, pres_pen, stop_txt, log_state],
-        additional_outputs=[log_state],      # ★ 必加
+        additional_outputs=[log_state],
         type="messages"
     )
 
+
+    # 日志刷新定时器
+    timer = gr.Timer(1.0, render=True)
+    timer.tick(
+        fn=consume_logs,
+        inputs=[],
+        outputs=[log_box],
+    )
+
+    log_state.change(lambda txt: gr.update(value=txt), log_state, log_box)
+    dbg_chk.change(lambda v: gr.update(visible=v), dbg_chk, log_box)
+
+
 demo.launch(server_name="0.0.0.0", server_port=30001)