优化Token溢出时的处理

config.py

@@ -23,6 +23,8 @@ else:
 
 
 # [step 3]>> 以下配置可以优化体验，但大部分场合下并不需要修改
+# 对话窗的高度
+CHATBOT_HEIGHT = 1117
 
 # 发送请求到OpenAI后，等待多久判定为超时
 TIMEOUT_SECONDS = 25

crazy_functions/test_project/cpp/longcode/prod_cons.h

@@ -0,0 +1,433 @@
+#pragma once
+
+#include <atomic>
+#include <utility>
+#include <cstring>
+#include <type_traits>
+#include <cstdint>
+
+#include "libipc/def.h"
+
+#include "libipc/platform/detail.h"
+#include "libipc/circ/elem_def.h"
+#include "libipc/utility/log.h"
+#include "libipc/utility/utility.h"
+
+namespace ipc {
+
+////////////////////////////////////////////////////////////////
+/// producer-consumer implementation
+////////////////////////////////////////////////////////////////
+
+template <typename Flag>
+struct prod_cons_impl;
+
+template <>
+struct prod_cons_impl<wr<relat::single, relat::single, trans::unicast>> {
+
+    template <std::size_t DataSize, std::size_t AlignSize>
+    struct elem_t {
+        std::aligned_storage_t<DataSize, AlignSize> data_ {};
+    };
+
+    alignas(cache_line_size) std::atomic<circ::u2_t> rd_; // read index
+    alignas(cache_line_size) std::atomic<circ::u2_t> wt_; // write index
+
+    constexpr circ::u2_t cursor() const noexcept {
+        return 0;
+    }
+
+    template <typename W, typename F, typename E>
+    bool push(W* /*wrapper*/, F&& f, E* elems) {
+        auto cur_wt = circ::index_of(wt_.load(std::memory_order_relaxed));
+        if (cur_wt == circ::index_of(rd_.load(std::memory_order_acquire) - 1)) {
+            return false; // full
+        }
+        std::forward<F>(f)(&(elems[cur_wt].data_));
+        wt_.fetch_add(1, std::memory_order_release);
+        return true;
+    }
+
+    /**
+     * In single-single-unicast, 'force_push' means 'no reader' or 'the only one reader is dead'.
+     * So we could just disconnect all connections of receiver, and return false.
+    */
+    template <typename W, typename F, typename E>
+    bool force_push(W* wrapper, F&&, E*) {
+        wrapper->elems()->disconnect_receiver(~static_cast<circ::cc_t>(0u));
+        return false;
+    }
+
+    template <typename W, typename F, typename R, typename E>
+    bool pop(W* /*wrapper*/, circ::u2_t& /*cur*/, F&& f, R&& out, E* elems) {
+        auto cur_rd = circ::index_of(rd_.load(std::memory_order_relaxed));
+        if (cur_rd == circ::index_of(wt_.load(std::memory_order_acquire))) {
+            return false; // empty
+        }
+        std::forward<F>(f)(&(elems[cur_rd].data_));
+        std::forward<R>(out)(true);
+        rd_.fetch_add(1, std::memory_order_release);
+        return true;
+    }
+};
+
+template <>
+struct prod_cons_impl<wr<relat::single, relat::multi , trans::unicast>>
+     : prod_cons_impl<wr<relat::single, relat::single, trans::unicast>> {
+
+    template <typename W, typename F, typename E>
+    bool force_push(W* wrapper, F&&, E*) {
+        wrapper->elems()->disconnect_receiver(1);
+        return false;
+    }
+
+    template <typename W, typename F, typename R, 
+              template <std::size_t, std::size_t> class E, std::size_t DS, std::size_t AS>
+    bool pop(W* /*wrapper*/, circ::u2_t& /*cur*/, F&& f, R&& out, E<DS, AS>* elems) {
+        byte_t buff[DS];
+        for (unsigned k = 0;;) {
+            auto cur_rd = rd_.load(std::memory_order_relaxed);
+            if (circ::index_of(cur_rd) ==
+                circ::index_of(wt_.load(std::memory_order_acquire))) {
+                return false; // empty
+            }
+            std::memcpy(buff, &(elems[circ::index_of(cur_rd)].data_), sizeof(buff));
+            if (rd_.compare_exchange_weak(cur_rd, cur_rd + 1, std::memory_order_release)) {
+                std::forward<F>(f)(buff);
+                std::forward<R>(out)(true);
+                return true;
+            }
+            ipc::yield(k);
+        }
+    }
+};
+
+template <>
+struct prod_cons_impl<wr<relat::multi , relat::multi, trans::unicast>>
+     : prod_cons_impl<wr<relat::single, relat::multi, trans::unicast>> {
+
+    using flag_t = std::uint64_t;
+
+    template <std::size_t DataSize, std::size_t AlignSize>
+    struct elem_t {
+        std::aligned_storage_t<DataSize, AlignSize> data_ {};
+        std::atomic<flag_t> f_ct_ { 0 }; // commit flag
+    };
+
+    alignas(cache_line_size) std::atomic<circ::u2_t> ct_; // commit index
+
+    template <typename W, typename F, typename E>
+    bool push(W* /*wrapper*/, F&& f, E* elems) {
+        circ::u2_t cur_ct, nxt_ct;
+        for (unsigned k = 0;;) {
+            cur_ct = ct_.load(std::memory_order_relaxed);
+            if (circ::index_of(nxt_ct = cur_ct + 1) ==
+                circ::index_of(rd_.load(std::memory_order_acquire))) {
+                return false; // full
+            }
+            if (ct_.compare_exchange_weak(cur_ct, nxt_ct, std::memory_order_acq_rel)) {
+                break;
+            }
+            ipc::yield(k);
+        }
+        auto* el = elems + circ::index_of(cur_ct);
+        std::forward<F>(f)(&(el->data_));
+        // set flag & try update wt
+        el->f_ct_.store(~static_cast<flag_t>(cur_ct), std::memory_order_release);
+        while (1) {
+            auto cac_ct = el->f_ct_.load(std::memory_order_acquire);
+            if (cur_ct != wt_.load(std::memory_order_relaxed)) {
+                return true;
+            }
+            if ((~cac_ct) != cur_ct) {
+                return true;
+            }
+            if (!el->f_ct_.compare_exchange_strong(cac_ct, 0, std::memory_order_relaxed)) {
+                return true;
+            }
+            wt_.store(nxt_ct, std::memory_order_release);
+            cur_ct = nxt_ct;
+            nxt_ct = cur_ct + 1;
+            el = elems + circ::index_of(cur_ct);
+        }
+        return true;
+    }
+
+    template <typename W, typename F, typename E>
+    bool force_push(W* wrapper, F&&, E*) {
+        wrapper->elems()->disconnect_receiver(1);
+        return false;
+    }
+
+    template <typename W, typename F, typename R, 
+              template <std::size_t, std::size_t> class E, std::size_t DS, std::size_t AS>
+    bool pop(W* /*wrapper*/, circ::u2_t& /*cur*/, F&& f, R&& out, E<DS, AS>* elems) {
+        byte_t buff[DS];
+        for (unsigned k = 0;;) {
+            auto cur_rd = rd_.load(std::memory_order_relaxed);
+            auto cur_wt = wt_.load(std::memory_order_acquire);
+            auto id_rd  = circ::index_of(cur_rd);
+            auto id_wt  = circ::index_of(cur_wt);
+            if (id_rd == id_wt) {
+                auto* el = elems + id_wt;
+                auto cac_ct = el->f_ct_.load(std::memory_order_acquire);
+                if ((~cac_ct) != cur_wt) {
+                    return false; // empty
+                }
+                if (el->f_ct_.compare_exchange_weak(cac_ct, 0, std::memory_order_relaxed)) {
+                    wt_.store(cur_wt + 1, std::memory_order_release);
+                }
+                k = 0;
+            }
+            else {
+                std::memcpy(buff, &(elems[circ::index_of(cur_rd)].data_), sizeof(buff));
+                if (rd_.compare_exchange_weak(cur_rd, cur_rd + 1, std::memory_order_release)) {
+                    std::forward<F>(f)(buff);
+                    std::forward<R>(out)(true);
+                    return true;
+                }
+                ipc::yield(k);
+            }
+        }
+    }
+};
+
+template <>
+struct prod_cons_impl<wr<relat::single, relat::multi, trans::broadcast>> {
+
+    using rc_t = std::uint64_t;
+
+    enum : rc_t {
+        ep_mask = 0x00000000ffffffffull,
+        ep_incr = 0x0000000100000000ull
+    };
+
+    template <std::size_t DataSize, std::size_t AlignSize>
+    struct elem_t {
+        std::aligned_storage_t<DataSize, AlignSize> data_ {};
+        std::atomic<rc_t> rc_ { 0 }; // read-counter
+    };
+
+    alignas(cache_line_size) std::atomic<circ::u2_t> wt_;   // write index
+    alignas(cache_line_size) rc_t epoch_ { 0 };             // only one writer
+
+    circ::u2_t cursor() const noexcept {
+        return wt_.load(std::memory_order_acquire);
+    }
+
+    template <typename W, typename F, typename E>
+    bool push(W* wrapper, F&& f, E* elems) {
+        E* el;
+        for (unsigned k = 0;;) {
+            circ::cc_t cc = wrapper->elems()->connections(std::memory_order_relaxed);
+            if (cc == 0) return false; // no reader
+            el = elems + circ::index_of(wt_.load(std::memory_order_relaxed));
+            // check all consumers have finished reading this element
+            auto cur_rc = el->rc_.load(std::memory_order_acquire);
+            circ::cc_t rem_cc = cur_rc & ep_mask;
+            if ((cc & rem_cc) && ((cur_rc & ~ep_mask) == epoch_)) {
+                return false; // has not finished yet
+            }
+            // consider rem_cc to be 0 here
+            if (el->rc_.compare_exchange_weak(
+                        cur_rc, epoch_ | static_cast<rc_t>(cc), std::memory_order_release)) {
+                break;
+            }
+            ipc::yield(k);
+        }
+        std::forward<F>(f)(&(el->data_));
+        wt_.fetch_add(1, std::memory_order_release);
+        return true;
+    }
+
+    template <typename W, typename F, typename E>
+    bool force_push(W* wrapper, F&& f, E* elems) {
+        E* el;
+        epoch_ += ep_incr;
+        for (unsigned k = 0;;) {
+            circ::cc_t cc = wrapper->elems()->connections(std::memory_order_relaxed);
+            if (cc == 0) return false; // no reader
+            el = elems + circ::index_of(wt_.load(std::memory_order_relaxed));
+            // check all consumers have finished reading this element
+            auto cur_rc = el->rc_.load(std::memory_order_acquire);
+            circ::cc_t rem_cc = cur_rc & ep_mask;
+            if (cc & rem_cc) {
+                ipc::log("force_push: k = %u, cc = %u, rem_cc = %u\n", k, cc, rem_cc);
+                cc = wrapper->elems()->disconnect_receiver(rem_cc); // disconnect all invalid readers
+                if (cc == 0) return false; // no reader
+            }
+            // just compare & exchange
+            if (el->rc_.compare_exchange_weak(
+                        cur_rc, epoch_ | static_cast<rc_t>(cc), std::memory_order_release)) {
+                break;
+            }
+            ipc::yield(k);
+        }
+        std::forward<F>(f)(&(el->data_));
+        wt_.fetch_add(1, std::memory_order_release);
+        return true;
+    }
+
+    template <typename W, typename F, typename R, typename E>
+    bool pop(W* wrapper, circ::u2_t& cur, F&& f, R&& out, E* elems) {
+        if (cur == cursor()) return false; // acquire
+        auto* el = elems + circ::index_of(cur++);
+        std::forward<F>(f)(&(el->data_));
+        for (unsigned k = 0;;) {
+            auto cur_rc = el->rc_.load(std::memory_order_acquire);
+            if ((cur_rc & ep_mask) == 0) {
+                std::forward<R>(out)(true);
+                return true;
+            }
+            auto nxt_rc = cur_rc & ~static_cast<rc_t>(wrapper->connected_id());
+            if (el->rc_.compare_exchange_weak(cur_rc, nxt_rc, std::memory_order_release)) {
+                std::forward<R>(out)((nxt_rc & ep_mask) == 0);
+                return true;
+            }
+            ipc::yield(k);
+        }
+    }
+};
+
+template <>
+struct prod_cons_impl<wr<relat::multi, relat::multi, trans::broadcast>> {
+
+    using rc_t   = std::uint64_t;
+    using flag_t = std::uint64_t;
+
+    enum : rc_t {
+        rc_mask = 0x00000000ffffffffull,
+        ep_mask = 0x00ffffffffffffffull,
+        ep_incr = 0x0100000000000000ull,
+        ic_mask = 0xff000000ffffffffull,
+        ic_incr = 0x0000000100000000ull
+    };
+
+    template <std::size_t DataSize, std::size_t AlignSize>
+    struct elem_t {
+        std::aligned_storage_t<DataSize, AlignSize> data_ {};
+        std::atomic<rc_t  > rc_   { 0 }; // read-counter
+        std::atomic<flag_t> f_ct_ { 0 }; // commit flag
+    };
+
+    alignas(cache_line_size) std::atomic<circ::u2_t> ct_;   // commit index
+    alignas(cache_line_size) std::atomic<rc_t> epoch_ { 0 };
+
+    circ::u2_t cursor() const noexcept {
+        return ct_.load(std::memory_order_acquire);
+    }
+
+    constexpr static rc_t inc_rc(rc_t rc) noexcept {
+        return (rc & ic_mask) | ((rc + ic_incr) & ~ic_mask);
+    }
+
+    constexpr static rc_t inc_mask(rc_t rc) noexcept {
+        return inc_rc(rc) & ~rc_mask;
+    }
+
+    template <typename W, typename F, typename E>
+    bool push(W* wrapper, F&& f, E* elems) {
+        E* el;
+        circ::u2_t cur_ct;
+        rc_t epoch = epoch_.load(std::memory_order_acquire);
+        for (unsigned k = 0;;) {
+            circ::cc_t cc = wrapper->elems()->connections(std::memory_order_relaxed);
+            if (cc == 0) return false; // no reader
+            el = elems + circ::index_of(cur_ct = ct_.load(std::memory_order_relaxed));
+            // check all consumers have finished reading this element
+            auto cur_rc = el->rc_.load(std::memory_order_relaxed);
+            circ::cc_t rem_cc = cur_rc & rc_mask;
+            if ((cc & rem_cc) && ((cur_rc & ~ep_mask) == epoch)) {
+                return false; // has not finished yet
+            }
+            else if (!rem_cc) {
+                auto cur_fl = el->f_ct_.load(std::memory_order_acquire);
+                if ((cur_fl != cur_ct) && cur_fl) {
+                    return false; // full
+                }
+            }
+            // consider rem_cc to be 0 here
+            if (el->rc_.compare_exchange_weak(
+                        cur_rc, inc_mask(epoch | (cur_rc & ep_mask)) | static_cast<rc_t>(cc), std::memory_order_relaxed) &&
+                epoch_.compare_exchange_weak(epoch, epoch, std::memory_order_acq_rel)) {
+                break;
+            }
+            ipc::yield(k);
+        }
+        // only one thread/process would touch here at one time
+        ct_.store(cur_ct + 1, std::memory_order_release);
+        std::forward<F>(f)(&(el->data_));
+        // set flag & try update wt
+        el->f_ct_.store(~static_cast<flag_t>(cur_ct), std::memory_order_release);
+        return true;
+    }
+
+    template <typename W, typename F, typename E>
+    bool force_push(W* wrapper, F&& f, E* elems) {
+        E* el;
+        circ::u2_t cur_ct;
+        rc_t epoch = epoch_.fetch_add(ep_incr, std::memory_order_release) + ep_incr;
+        for (unsigned k = 0;;) {
+            circ::cc_t cc = wrapper->elems()->connections(std::memory_order_relaxed);
+            if (cc == 0) return false; // no reader
+            el = elems + circ::index_of(cur_ct = ct_.load(std::memory_order_relaxed));
+            // check all consumers have finished reading this element
+            auto cur_rc = el->rc_.load(std::memory_order_acquire);
+            circ::cc_t rem_cc = cur_rc & rc_mask;
+            if (cc & rem_cc) {
+                ipc::log("force_push: k = %u, cc = %u, rem_cc = %u\n", k, cc, rem_cc);
+                cc = wrapper->elems()->disconnect_receiver(rem_cc); // disconnect all invalid readers
+                if (cc == 0) return false; // no reader
+            }
+            // just compare & exchange
+            if (el->rc_.compare_exchange_weak(
+                        cur_rc, inc_mask(epoch | (cur_rc & ep_mask)) | static_cast<rc_t>(cc), std::memory_order_relaxed)) {
+                if (epoch == epoch_.load(std::memory_order_acquire)) {
+                    break;
+                }
+                else if (push(wrapper, std::forward<F>(f), elems)) {
+                    return true;
+                }
+                epoch = epoch_.fetch_add(ep_incr, std::memory_order_release) + ep_incr;
+            }
+            ipc::yield(k);
+        }
+        // only one thread/process would touch here at one time
+        ct_.store(cur_ct + 1, std::memory_order_release);
+        std::forward<F>(f)(&(el->data_));
+        // set flag & try update wt
+        el->f_ct_.store(~static_cast<flag_t>(cur_ct), std::memory_order_release);
+        return true;
+    }
+
+    template <typename W, typename F, typename R, typename E, std::size_t N>
+    bool pop(W* wrapper, circ::u2_t& cur, F&& f, R&& out, E(& elems)[N]) {
+        auto* el = elems + circ::index_of(cur);
+        auto cur_fl = el->f_ct_.load(std::memory_order_acquire);
+        if (cur_fl != ~static_cast<flag_t>(cur)) {
+            return false; // empty
+        }
+        ++cur;
+        std::forward<F>(f)(&(el->data_));
+        for (unsigned k = 0;;) {
+            auto cur_rc = el->rc_.load(std::memory_order_acquire);
+            if ((cur_rc & rc_mask) == 0) {
+                std::forward<R>(out)(true);
+                el->f_ct_.store(cur + N - 1, std::memory_order_release);
+                return true;
+            }
+            auto nxt_rc = inc_rc(cur_rc) & ~static_cast<rc_t>(wrapper->connected_id());
+            bool last_one = false;
+            if ((last_one = (nxt_rc & rc_mask) == 0)) {
+                el->f_ct_.store(cur + N - 1, std::memory_order_release);
+            }
+            if (el->rc_.compare_exchange_weak(cur_rc, nxt_rc, std::memory_order_release)) {
+                std::forward<R>(out)(last_one);
+                return true;
+            }
+            ipc::yield(k);
+        }
+    }
+};
+
+} // namespace ipc

crazy_functions/高级功能函数模板.py

@@ -11,7 +11,7 @@ def 高阶功能模板函数(txt, top_p, temperature, chatbot, history, systemPr
     for i in range(5):
         currentMonth = (datetime.date.today() + datetime.timedelta(days=i)).month
         currentDay = (datetime.date.today() + datetime.timedelta(days=i)).day
-        i_say = f'历史中哪些事件发生在{currentMonth}月{currentDay}日？列举两条并发送相关图片。发送图片时，请使用Markdown，将Unsplash API中的PUT_YOUR_QUERY_HERE替换成描述改事件的三个最重要的单词。'
+        i_say = f'历史中哪些事件发生在{currentMonth}月{currentDay}日？列举两条并发送相关图片。发送图片时，请使用Markdown，将Unsplash API中的PUT_YOUR_QUERY_HERE替换成描述该事件的一个最重要的单词。'
         chatbot.append((i_say, "[Local Message] waiting gpt response."))
         yield chatbot, history, '正常'  # 由于请求gpt需要一段时间，我们先及时地做一次状态显示
 

functional_crazy.py

@@ -19,10 +19,10 @@ def get_crazy_functionals():
 
     function_plugins = {
         "请解析并解构此项目本身": {
-            # HotReload 的意思是热更新，修改函数插件后，不需要重启程序，代码直接生效
+            "AsButton": False,  # 加入下拉菜单中
             "Function": 解析项目本身
         },
-        "解析整个py项目": {
+        "解析整个Py项目": {
             "Color": "stop",    # 按钮颜色
             "Function": 解析一个Python项目
         },
@@ -32,9 +32,10 @@ def get_crazy_functionals():
         },
         "解析整个C++项目": {
             "Color": "stop",    # 按钮颜色
+            "AsButton": False,  # 加入下拉菜单中
             "Function": 解析一个C项目
         },
-        "读tex论文写摘要": {
+        "读Tex论文写摘要": {
             "Color": "stop",    # 按钮颜色
             "Function": 读文章写摘要
         },
@@ -52,7 +53,7 @@ def get_crazy_functionals():
         },
     }
 
-    # VisibleLevel=1 经过测试，但功能未达到理想状态
+    # VisibleLevel=1 经过测试，但功能上距离达到完美状态还差一点点
     if UserVisibleLevel >= 1:
         from crazy_functions.批量总结PDF文档 import 批量总结PDF文档
         from crazy_functions.批量总结PDF文档pdfminer import 批量总结PDF文档pdfminer
@@ -60,11 +61,11 @@ def get_crazy_functionals():
         function_plugins.update({
             "[仅供开发调试] 批量总结PDF文档": {
                 "Color": "stop",
-                # HotReload 的意思是热更新，修改函数插件代码后，不需要重启程序，代码直接生效
-                "Function": HotReload(批量总结PDF文档)
+                "Function": HotReload(批量总结PDF文档) # HotReload 的意思是热更新，修改函数插件代码后，不需要重启程序，代码直接生效
             },
             "[仅供开发调试] 批量总结PDF文档pdfminer": {
                 "Color": "stop",
+                "AsButton": False,  # 加入下拉菜单中
                 "Function": HotReload(批量总结PDF文档pdfminer)
             },
             "[仅供开发调试] 批量总结Word文档": {

main.py

@@ -4,9 +4,8 @@ from predict import predict
 from toolbox import format_io, find_free_port, on_file_uploaded, on_report_generated, get_conf
 
 # 建议您复制一个config_private.py放自己的秘密, 如API和代理网址, 避免不小心传github被别人看到
-proxies, WEB_PORT, LLM_MODEL, CONCURRENT_COUNT, AUTHENTICATION = \
-    get_conf('proxies', 'WEB_PORT', 'LLM_MODEL', 'CONCURRENT_COUNT', 'AUTHENTICATION')
-
+proxies, WEB_PORT, LLM_MODEL, CONCURRENT_COUNT, AUTHENTICATION, CHATBOT_HEIGHT = \
+    get_conf('proxies', 'WEB_PORT', 'LLM_MODEL', 'CONCURRENT_COUNT', 'AUTHENTICATION', 'CHATBOT_HEIGHT')
 
 # 如果WEB_PORT是-1, 则随机选取WEB端口
 PORT = find_free_port() if WEB_PORT <= 0 else WEB_PORT
@@ -17,18 +16,18 @@ title_html = """<h1 align="center">ChatGPT 学术优化</h1>"""
 
 # 问询记录, python 版本建议3.9+（越新越好）
 import logging
-os.makedirs('gpt_log', exist_ok=True)
-try:logging.basicConfig(filename='gpt_log/chat_secrets.log', level=logging.INFO, encoding='utf-8')
-except:logging.basicConfig(filename='gpt_log/chat_secrets.log', level=logging.INFO)
-print('所有问询记录将自动保存在本地目录./gpt_log/chat_secrets.log, 请注意自我隐私保护哦！')
+os.makedirs("gpt_log", exist_ok=True)
+try:logging.basicConfig(filename="gpt_log/chat_secrets.log", level=logging.INFO, encoding="utf-8")
+except:logging.basicConfig(filename="gpt_log/chat_secrets.log", level=logging.INFO)
+print("所有问询记录将自动保存在本地目录./gpt_log/chat_secrets.log, 请注意自我隐私保护哦！")
 
 # 一些普通功能模块
 from functional import get_functionals
 functional = get_functionals()
 
-# 对一些丧心病狂的实验性功能模块进行测试
+# 高级函数插件
 from functional_crazy import get_crazy_functionals
-crazy_functional = get_crazy_functionals()
+crazy_fns = get_crazy_functionals()
 
 # 处理markdown文本格式的转变
 gr.Chatbot.postprocess = format_io
@@ -40,11 +39,10 @@ set_theme = adjust_theme()
 cancel_handles = []
 with gr.Blocks(theme=set_theme, analytics_enabled=False) as demo:
     gr.HTML(title_html)
-    with gr.Row():
+    with gr.Row().style(equal_height=True):
         with gr.Column(scale=2):
             chatbot = gr.Chatbot()
-            chatbot.style(height=1150)
-            chatbot.style()
+            chatbot.style(height=CHATBOT_HEIGHT)
             history = gr.State([])
         with gr.Column(scale=1):
             with gr.Row():
@@ -66,49 +64,70 @@ with gr.Blocks(theme=set_theme, analytics_enabled=False) as demo:
                 with gr.Row():
                     gr.Markdown("注意：以下“红颜色”标识的函数插件需从input区读取路径作为参数.")
                 with gr.Row():
-                    for k in crazy_functional:
-                        variant = crazy_functional[k]["Color"] if "Color" in crazy_functional[k] else "secondary"
-                        crazy_functional[k]["Button"] = gr.Button(k, variant=variant)
+                    for k in crazy_fns:
+                        if not crazy_fns[k].get("AsButton", True): continue
+                        variant = crazy_fns[k]["Color"] if "Color" in crazy_fns[k] else "secondary"
+                        crazy_fns[k]["Button"] = gr.Button(k, variant=variant)
                 with gr.Row():
-                    with gr.Accordion("展开“文件上传区”。上传本地文件供“红颜色”的函数插件调用。", open=False):
-                        file_upload = gr.Files(label='任何文件, 但推荐上传压缩文件(zip, tar)', file_count="multiple")
+                    with gr.Accordion("更多函数插件", open=True):
+                        dropdown_fn_list = [k for k in crazy_fns.keys() if not crazy_fns[k].get("AsButton", True)]
+                        with gr.Column(scale=1):
+                            dropdown = gr.Dropdown(dropdown_fn_list, value=r"打开插件列表", label="").style(container=False)
+                        with gr.Column(scale=1):
+                            switchy_bt = gr.Button(r"请先从插件列表中选择", variant="secondary")
+                with gr.Row():
+                    with gr.Accordion("点击展开“文件上传区”。上传本地文件可供红色函数插件调用。", open=False) as area_file_up:
+                        file_upload = gr.Files(label="任何文件, 但推荐上传压缩文件(zip, tar)", file_count="multiple")
             with gr.Accordion("展开SysPrompt & GPT参数 & 交互界面布局", open=False):
                 system_prompt = gr.Textbox(show_label=True, placeholder=f"System Prompt", label="System prompt", value=initial_prompt)
                 top_p = gr.Slider(minimum=-0, maximum=1.0, value=1.0, step=0.01,interactive=True, label="Top-p (nucleus sampling)",)
                 temperature = gr.Slider(minimum=-0, maximum=2.0, value=1.0, step=0.01, interactive=True, label="Temperature",)
-                checkboxes = gr.CheckboxGroup(["基础功能区", "函数插件区"], 
-                                        value=["基础功能区", "函数插件区"], label="显示哪些功能区")
+                checkboxes = gr.CheckboxGroup(["基础功能区", "函数插件区"], value=["基础功能区", "函数插件区"], label="显示/隐藏功能区")
 
-    def what_is_this(a):
+    # 功能区显示开关与功能区的互动
+    def fn_area_visibility(a):
         ret = {}
-        # if area_basic_fn.visible != ("基础功能区" in a): 
-        ret.update({area_basic_fn: gr.update(visible=("基础功能区" in a))}) 
-        # if area_crazy_fn.visible != ("函数插件区" in a): 
-        ret.update({area_crazy_fn: gr.update(visible=("函数插件区" in a))}) 
+        ret.update({area_basic_fn: gr.update(visible=("基础功能区" in a))})
+        ret.update({area_crazy_fn: gr.update(visible=("函数插件区" in a))})
         return ret
-
-    checkboxes.select(what_is_this, [checkboxes], [area_basic_fn, area_crazy_fn] )
-
-    predict_args = dict(fn=predict, inputs=[txt, top_p, temperature, chatbot, history, system_prompt], outputs=[chatbot, history, statusDisplay], show_progress=True)
+    checkboxes.select(fn_area_visibility, [checkboxes], [area_basic_fn, area_crazy_fn] )
+    # 整理反复出现的控件句柄组合
+    input_combo = [txt, top_p, temperature, chatbot, history, system_prompt]
+    output_combo = [chatbot, history, statusDisplay]
+    predict_args = dict(fn=predict, inputs=input_combo, outputs=output_combo, show_progress=True)
     empty_txt_args = dict(fn=lambda: "", inputs=[], outputs=[txt]) # 用于在提交后清空输入栏
-
-    cancel_handles.append(txt.submit(**predict_args))
-    # txt.submit(**empty_txt_args) 在提交后清空输入栏
-    cancel_handles.append(submitBtn.click(**predict_args))
-    # submitBtn.click(**empty_txt_args) 在提交后清空输入栏
-    resetBtn.click(lambda: ([], [], "已重置"), None, [chatbot, history, statusDisplay])
+    # 提交按钮、重置按钮
+    cancel_handles.append(txt.submit(**predict_args)) #; txt.submit(**empty_txt_args) 在提交后清空输入栏
+    cancel_handles.append(submitBtn.click(**predict_args)) #; submitBtn.click(**empty_txt_args) 在提交后清空输入栏
+    resetBtn.click(lambda: ([], [], "已重置"), None, output_combo)
+    # 基础功能区的回调函数注册
     for k in functional:
-        click_handle = functional[k]["Button"].click(predict,
-            [txt, top_p, temperature, chatbot, history, system_prompt, gr.State(True), gr.State(k)], [chatbot, history, statusDisplay], show_progress=True)
+        click_handle = functional[k]["Button"].click(predict, [*input_combo, gr.State(True), gr.State(k)], output_combo, show_progress=True)
         cancel_handles.append(click_handle)
+    # 文件上传区，接收文件后与chatbot的互动
     file_upload.upload(on_file_uploaded, [file_upload, chatbot, txt], [chatbot, txt])
-    for k in crazy_functional:
-        click_handle = crazy_functional[k]["Button"].click(crazy_functional[k]["Function"],
-            [txt, top_p, temperature, chatbot, history, system_prompt, gr.State(PORT)], [chatbot, history, statusDisplay]
-        )
-        try: click_handle.then(on_report_generated, [file_upload, chatbot], [file_upload, chatbot])
-        except: pass
+    # 函数插件-固定按钮区
+    for k in crazy_fns:
+        if not crazy_fns[k].get("AsButton", True): continue
+        click_handle = crazy_fns[k]["Button"].click(crazy_fns[k]["Function"], [*input_combo, gr.State(PORT)], output_combo)
+        click_handle.then(on_report_generated, [file_upload, chatbot], [file_upload, chatbot])
         cancel_handles.append(click_handle)
+    # 函数插件-下拉菜单与随变按钮的互动
+    def on_dropdown_changed(k):
+        variant = crazy_fns[k]["Color"] if "Color" in crazy_fns[k] else "secondary"
+        return {switchy_bt: gr.update(value=k, variant=variant)}
+    dropdown.select(on_dropdown_changed, [dropdown], [switchy_bt] )
+    # 随变按钮的回调函数注册
+    def route(k, *args, **kwargs):
+        if k in [r"打开插件列表", r"先从插件列表中选择"]: return 
+        yield from crazy_fns[k]["Function"](*args, **kwargs)
+    click_handle = switchy_bt.click(route,[switchy_bt, *input_combo, gr.State(PORT)], output_combo)
+    click_handle.then(on_report_generated, [file_upload, chatbot], [file_upload, chatbot])
+    def expand_file_area(file_upload, area_file_up):
+        if len(file_upload)>0: return {area_file_up: gr.update(open=True)}
+    click_handle.then(expand_file_area, [file_upload, area_file_up], [area_file_up])
+    cancel_handles.append(click_handle)
+    # 终止按钮的回调函数注册
     stopBtn.click(fn=None, inputs=None, outputs=None, cancels=cancel_handles)
 
 # gradio的inbrowser触发不太稳定，回滚代码到原始的浏览器打开函数
@@ -117,7 +136,7 @@ def auto_opentab_delay():
     print(f"如果浏览器没有自动打开，请复制并转到以下URL: http://localhost:{PORT}")
     def open(): 
         time.sleep(2)
-        webbrowser.open_new_tab(f'http://localhost:{PORT}')
+        webbrowser.open_new_tab(f"http://localhost:{PORT}")
     threading.Thread(target=open, name="open-browser", daemon=True).start()
 
 auto_opentab_delay()

predict.py

@@ -96,13 +96,19 @@ def predict_no_ui_long_connection(inputs, top_p, temperature, history=[], sys_pr
         except StopIteration: break
         if len(chunk)==0: continue
         if not chunk.startswith('data:'): 
-            chunk = get_full_error(chunk.encode('utf8'), stream_response)
-            raise ConnectionAbortedError("OpenAI拒绝了请求:" + chunk.decode())
-        delta = json.loads(chunk.lstrip('data:'))['choices'][0]["delta"]
+            error_msg = get_full_error(chunk.encode('utf8'), stream_response).decode()
+            if "reduce the length" in error_msg:
+                raise ConnectionAbortedError("OpenAI拒绝了请求:" + error_msg)
+            else:
+                raise RuntimeError("OpenAI拒绝了请求：" + error_msg)
+        json_data = json.loads(chunk.lstrip('data:'))['choices'][0]
+        delta = json_data["delta"]
         if len(delta) == 0: break
         if "role" in delta: continue
         if "content" in delta: result += delta["content"]; print(delta["content"], end='')
         else: raise RuntimeError("意外Json结构："+delta)
+    if json_data['finish_reason'] == 'length':
+        raise ConnectionAbortedError("正常结束，但显示Token不足。")
     return result
 
 

toolbox.py

@@ -2,21 +2,21 @@ import markdown, mdtex2html, threading, importlib, traceback, importlib, inspect
 from show_math import convert as convert_math
 from functools import wraps
 
-def get_reduce_token_percent(e):
+def get_reduce_token_percent(text):
     try:
         # text = "maximum context length is 4097 tokens. However, your messages resulted in 4870 tokens"
         pattern = r"(\d+)\s+tokens\b"
         match = re.findall(pattern, text)
-        eps = 50 # 稍微留一点余地, 确保下次别再超过token
-        max_limit = float(match[0]) - eps
+        EXCEED_ALLO = 500 # 稍微留一点余地，否则在回复时会因余量太少出问题
+        max_limit = float(match[0]) - EXCEED_ALLO
         current_tokens = float(match[1])
         ratio = max_limit/current_tokens
         assert ratio > 0 and ratio < 1
-        return ratio
+        return ratio, str(int(current_tokens-max_limit))
     except:
-        return 0.5
+        return 0.5, '不详'
 
-def predict_no_ui_but_counting_down(i_say, i_say_show_user, chatbot, top_p, temperature, history=[], sys_prompt='', long_connection=False):
+def predict_no_ui_but_counting_down(i_say, i_say_show_user, chatbot, top_p, temperature, history=[], sys_prompt='', long_connection=True):
     """
         调用简单的predict_no_ui接口，但是依然保留了些许界面心跳功能，当对话太长时，会自动采用二分法截断
         i_say: 当前输入
@@ -45,19 +45,18 @@ def predict_no_ui_but_counting_down(i_say, i_say_show_user, chatbot, top_p, temp
                 break
             except ConnectionAbortedError as token_exceeded_error:
                 # 尝试计算比例，尽可能多地保留文本
-                p_ratio = get_reduce_token_percent(str(token_exceeded_error))
+                p_ratio, n_exceed = get_reduce_token_percent(str(token_exceeded_error))
                 if len(history) > 0:
                     history = [his[     int(len(his)    *p_ratio):      ] for his in history if his is not None]
-                    mutable[1] = 'Warning! History conversation is too long, cut into half. '
                 else:
                     i_say = i_say[:     int(len(i_say)  *p_ratio)     ]
-                    mutable[1] = 'Warning! Input file is too long, cut into half. '
+                mutable[1] = f'警告，文本过长将进行截断，Token溢出数：{n_exceed}，截断比例：{(1-p_ratio):.0%}。'
             except TimeoutError as e:
-                mutable[0] = '[Local Message] Failed with timeout.'
+                mutable[0] = '[Local Message] 请求超时。'
                 raise TimeoutError
             except Exception as e:
-                mutable[0] = f'[Local Message] Failed with {str(e)}.'
-                raise RuntimeError(f'[Local Message] Failed with {str(e)}.')
+                mutable[0] = f'[Local Message] 异常：{str(e)}.'
+                raise RuntimeError(f'[Local Message] 异常：{str(e)}.')
     # 创建新线程发出http请求
     thread_name = threading.Thread(target=mt, args=(i_say, history)); thread_name.start()
     # 原来的线程则负责持续更新UI，实现一个超时倒计时，并等待新线程的任务完成
@@ -286,7 +285,7 @@ def on_report_generated(files, chatbot):
     report_files = find_recent_files('gpt_log')
     if len(report_files) == 0: return report_files, chatbot
     # files.extend(report_files)
-    chatbot.append(['汇总报告如何远程获取？', '汇总报告已经添加到右侧文件上传区，请查收。'])
+    chatbot.append(['汇总报告如何远程获取？', '汇总报告已经添加到右侧“文件上传区”（可能处于折叠状态），请查收。'])
     return report_files, chatbot
 
 def get_conf(*args):