Upload 11 files

app.py

@@ -1,10 +1,19 @@
 import subprocess
 import select
-from hf_api import restart_space
+import os
+from mods.btn_reset import restart_space
+
+if not os.path.exists('downloads/causallm_7b.Q5_K_M.gguf'):
+    from huggingface_hub import snapshot_download
+    os.mkdir("downloads")
+    os.mkdir("cache")
+    snapshot_download(repo_id='TheBloke/CausalLM-7B-GGUF', local_dir=r'downloads',
+                      allow_patterns='causallm_7b.Q5_K_M.gguf')
+    snapshot_download(repo_id='Limour/llama-python-streamingllm-cache', repo_type='dataset', local_dir=r'cache')
 
 try:
     # 启动另一个程序，并通过管道捕获其输出
-    process = subprocess.Popen(["python", "sub_app.py"],
+    process = subprocess.Popen(["python", "gradio_streamingllm.py"],
                                stdout=subprocess.PIPE,
                                stderr=subprocess.PIPE,
                                bufsize=1, universal_newlines=True)

gradio_streamingllm.py

@@ -0,0 +1,162 @@
+import gradio as gr
+import threading
+from llama_cpp_python_streamingllm import StreamingLLM
+from mods.read_cfg import cfg
+
+from mods.text_display import init as text_display_init
+
+from mods.btn_rag import init as btn_rag_init
+
+# ========== 按钮中用到的共同的函数 ==========
+from mods.btn_com import init as btn_com_init
+
+# ========== 输出一段旁白 ==========
+from mods.btn_submit import init as btn_submit_init
+
+# ========== 输出一段旁白 ==========
+from mods.btn_vo import init as btn_vo_init
+
+# ========== 给用户提供默认回复的建议 ==========
+from mods.btn_suggest import init as btn_suggest_init
+
+# ========== 重置按钮 ==========
+from mods.btn_reset import init as btn_reset_init
+
+# ========== 聊天的模版 默认 chatml ==========
+from chat_template import ChatTemplate
+
+#  ========== 全局锁，确保只能进行一个会话 ==========
+cfg['session_lock'] = threading.Lock()
+cfg['session_active'] = False
+
+#  ========== 温度、采样之类的设置 ==========
+with gr.Blocks() as setting:
+    with gr.Row():
+        cfg['setting_path'] = gr.Textbox(label="模型路径", max_lines=1, scale=2, **cfg['setting_path'])
+        cfg['setting_cache_path'] = gr.Textbox(label="缓存路径", max_lines=1, scale=2, **cfg['setting_cache_path'])
+        cfg['setting_seed'] = gr.Number(label="随机种子", scale=1, **cfg['setting_seed'])
+        cfg['setting_n_gpu_layers'] = gr.Number(label="n_gpu_layers", scale=1, **cfg['setting_n_gpu_layers'])
+    with gr.Row():
+        cfg['setting_ctx'] = gr.Number(label="上下文大小（Tokens）", **cfg['setting_ctx'])
+        cfg['setting_max_tokens'] = gr.Number(label="最大响应长度（Tokens）", interactive=True,
+                                              **cfg['setting_max_tokens'])
+        cfg['setting_n_keep'] = gr.Number(value=10, label="n_keep", interactive=False)
+        cfg['setting_n_discard'] = gr.Number(label="n_discard", interactive=True, **cfg['setting_n_discard'])
+    with gr.Row():
+        cfg['setting_temperature'] = gr.Number(label="温度", interactive=True, **cfg['setting_temperature'])
+        cfg['setting_repeat_penalty'] = gr.Number(label="重复惩罚", interactive=True, **cfg['setting_repeat_penalty'])
+        cfg['setting_frequency_penalty'] = gr.Number(label="频率惩罚", interactive=True,
+                                                     **cfg['setting_frequency_penalty'])
+        cfg['setting_presence_penalty'] = gr.Number(label="存在惩罚", interactive=True,
+                                                    **cfg['setting_presence_penalty'])
+        cfg['setting_repeat_last_n'] = gr.Number(label="惩罚范围", interactive=True, **cfg['setting_repeat_last_n'])
+    with gr.Row():
+        cfg['setting_top_k'] = gr.Number(label="Top-K", interactive=True, **cfg['setting_top_k'])
+        cfg['setting_top_p'] = gr.Number(label="Top P", interactive=True, **cfg['setting_top_p'])
+        cfg['setting_min_p'] = gr.Number(label="Min P", interactive=True, **cfg['setting_min_p'])
+        cfg['setting_typical_p'] = gr.Number(label="Typical", interactive=True, **cfg['setting_typical_p'])
+        cfg['setting_tfs_z'] = gr.Number(label="TFS", interactive=True, **cfg['setting_tfs_z'])
+    with gr.Row():
+        cfg['setting_mirostat_mode'] = gr.Number(label="Mirostat 模式", **cfg['setting_mirostat_mode'])
+        cfg['setting_mirostat_eta'] = gr.Number(label="Mirostat 学习率", interactive=True,
+                                                **cfg['setting_mirostat_eta'])
+        cfg['setting_mirostat_tau'] = gr.Number(label="Mirostat 目标熵", interactive=True,
+                                                **cfg['setting_mirostat_tau'])
+
+    #  ========== 加载模型 ==========
+    cfg['model'] = StreamingLLM(model_path=cfg['setting_path'].value,
+                                seed=cfg['setting_seed'].value,
+                                n_gpu_layers=cfg['setting_n_gpu_layers'].value,
+                                n_ctx=cfg['setting_ctx'].value)
+    cfg['chat_template'] = ChatTemplate(cfg['model'])
+    cfg['setting_ctx'].value = cfg['model'].n_ctx()
+
+# ========== 展示角色卡 ==========
+with gr.Blocks() as role:
+    with gr.Row():
+        cfg['role_usr'] = gr.Textbox(label="用户名称", max_lines=1, interactive=False, **cfg['role_usr'])
+        cfg['role_char'] = gr.Textbox(label="角色名称", max_lines=1, interactive=False, **cfg['role_char'])
+
+    cfg['role_char_d'] = gr.Textbox(lines=10, label="故事描述", **cfg['role_char_d'])
+    cfg['role_chat_style'] = gr.Textbox(lines=10, label="回复示例", **cfg['role_chat_style'])
+
+    # ========== 加载角色卡-缓存 ==========
+    from mods.load_cache import init as load_cache_init
+
+    text_display_init(cfg)
+    load_cache_init(cfg)
+
+# ========== 聊天页面 ==========
+with gr.Blocks() as chatting:
+    with gr.Row(equal_height=True):
+        cfg['chatbot'] = gr.Chatbot(height='60vh', scale=2, value=cfg['chatbot'],
+                                    avatar_images=(r'assets/user.png', r'assets/chatbot.webp'))
+        with gr.Column(scale=1, elem_id="area"):
+            cfg['rag'] = gr.Textbox(label='RAG', show_copy_button=True, elem_id="RAG-area")
+            cfg['vo'] = gr.Textbox(label='VO', show_copy_button=True, elem_id="VO-area")
+            cfg['s_info'] = gr.Textbox(value=cfg['model'].venv_info, max_lines=1, label='info', interactive=False)
+    cfg['msg'] = gr.Textbox(label='Prompt', lines=2, max_lines=2, elem_id='prompt', autofocus=True, **cfg['msg'])
+    with gr.Row():
+        cfg['btn_vo'] = gr.Button("旁白")
+        cfg['btn_rag'] = gr.Button("RAG")
+        cfg['btn_retry'] = gr.Button("Retry")
+        cfg['btn_com1'] = gr.Button("自定义1")
+        cfg['btn_reset'] = gr.Button("Reset")
+        cfg['btn_debug'] = gr.Button("Debug")
+        cfg['btn_submit'] = gr.Button("Submit")
+        cfg['btn_suggest'] = gr.Button("建议")
+
+    cfg['gr'] = gr
+    btn_com_init(cfg)
+
+    btn_rag_init(cfg)
+
+    btn_submit_init(cfg)
+
+    btn_vo_init(cfg)
+
+    btn_suggest_init(cfg)
+
+    # ========== 用于调试 ==========
+    btn_reset_init(cfg)
+
+#  ========== 让聊天界面的文本框等高 ==========
+custom_css = r'''
+#area > div {
+    height: 100%;
+}
+#RAG-area {
+    flex-grow: 1;
+}
+#RAG-area > label {
+    height: 100%;
+    display: flex;
+    flex-direction: column;
+}
+#RAG-area > label > textarea {
+    flex-grow: 1;
+    max-height: 20vh;
+}
+#VO-area {
+    flex-grow: 1;
+}
+#VO-area > label {
+    height: 100%;
+    display: flex;
+    flex-direction: column;
+}
+#VO-area > label > textarea {
+    flex-grow: 1;
+    max-height: 20vh;
+}
+#prompt > label > textarea {
+    max-height: 63px;
+}
+'''
+
+# ========== 开始运行 ==========
+demo = gr.TabbedInterface([chatting, setting, role],
+                          ["聊天", "设置", '角色'],
+                          css=custom_css)
+gr.close_all()
+demo.queue(api_open=False, max_size=1).launch(share=False)

mods/btn_com.py

@@ -0,0 +1,102 @@
+def init(cfg):
+    chat_template = cfg['chat_template']
+    model = cfg['model']
+    gr = cfg['gr']
+    lock = cfg['session_lock']
+
+    # ========== 流式输出函数 ==========
+    def btn_com(_n_keep, _n_discard,
+                _temperature, _repeat_penalty, _frequency_penalty,
+                _presence_penalty, _repeat_last_n, _top_k,
+                _top_p, _min_p, _typical_p,
+                _tfs_z, _mirostat_mode, _mirostat_eta,
+                _mirostat_tau, _role, _max_tokens):
+        # ========== 初始化输出模版 ==========
+        t_bot = chat_template(_role)
+        completion_tokens = []  # 有可能多个 tokens 才能构成一个 utf-8 编码的文字
+        history = ''
+        # ========== 流式输出 ==========
+        for token in model.generate_t(
+                tokens=t_bot,
+                n_keep=_n_keep,
+                n_discard=_n_discard,
+                im_start=chat_template.im_start_token,
+                top_k=_top_k,
+                top_p=_top_p,
+                min_p=_min_p,
+                typical_p=_typical_p,
+                temp=_temperature,
+                repeat_penalty=_repeat_penalty,
+                repeat_last_n=_repeat_last_n,
+                frequency_penalty=_frequency_penalty,
+                presence_penalty=_presence_penalty,
+                tfs_z=_tfs_z,
+                mirostat_mode=_mirostat_mode,
+                mirostat_tau=_mirostat_tau,
+                mirostat_eta=_mirostat_eta,
+        ):
+            if token in chat_template.eos or token == chat_template.nlnl:
+                t_bot.extend(completion_tokens)
+                print('token in eos', token)
+                break
+            completion_tokens.append(token)
+            all_text = model.str_detokenize(completion_tokens)
+            if not all_text:
+                continue
+            t_bot.extend(completion_tokens)
+            history += all_text
+            yield history
+            if token in chat_template.onenl:
+                # ========== 移除末尾的换行符 ==========
+                if t_bot[-2] in chat_template.onenl:
+                    model.venv_pop_token()
+                    break
+                if t_bot[-2] in chat_template.onerl and t_bot[-3] in chat_template.onenl:
+                    model.venv_pop_token()
+                    break
+            if history[-2:] == '\n\n':  # 各种 'x\n\n' 的token，比如'。\n\n'
+                print('t_bot[-4:]', t_bot[-4:], repr(model.str_detokenize(t_bot[-4:])),
+                      repr(model.str_detokenize(t_bot[-1:])))
+                break
+            if len(t_bot) > _max_tokens:
+                break
+            completion_tokens = []
+        # ========== 查看末尾的换行符 ==========
+        print('history', repr(history))
+        # ========== 给 kv_cache 加上输出结束符 ==========
+        model.eval_t(chat_template.im_end_nl, _n_keep, _n_discard)
+        t_bot.extend(chat_template.im_end_nl)
+
+    cfg['btn_com'] = btn_com
+
+    def btn_start_or_finish(finish):
+        tmp = gr.update(interactive=finish)
+
+        def _inner():
+            with lock:
+                if cfg['session_active'] != finish:
+                    raise RuntimeError
+                cfg['session_active'] = not cfg['session_active']
+            return tmp, tmp, tmp
+
+        return _inner
+
+    btn_start_or_finish_outputs = [cfg['btn_submit'], cfg['btn_vo'], cfg['btn_suggest']]
+
+    cfg['btn_start'] = {
+        'fn': btn_start_or_finish(False),
+        'outputs': btn_start_or_finish_outputs
+    }
+
+    cfg['btn_finish'] = {
+        'fn': btn_start_or_finish(True),
+        'outputs': btn_start_or_finish_outputs
+    }
+
+    cfg['setting'] = [cfg[x] for x in ('setting_n_keep', 'setting_n_discard',
+                                       'setting_temperature', 'setting_repeat_penalty', 'setting_frequency_penalty',
+                                       'setting_presence_penalty', 'setting_repeat_last_n', 'setting_top_k',
+                                       'setting_top_p', 'setting_min_p', 'setting_typical_p',
+                                       'setting_tfs_z', 'setting_mirostat_mode', 'setting_mirostat_eta',
+                                       'setting_mirostat_tau', 'role_usr', 'role_char',
+                                       'rag', 'setting_max_tokens')]

mods/btn_rag.py

@@ -0,0 +1,8 @@
+def init(cfg):
+    # ========== 待实现 ==========
+    def btn_rag_(_rag, _msg):
+        retn = ''
+        return retn
+
+    cfg['btn_rag'].click(fn=btn_rag_, outputs=cfg['rag'],
+                         inputs=[cfg['rag'], cfg['msg']])

mods/btn_reset.py

@@ -0,0 +1,40 @@
+import os
+
+from huggingface_hub import HfApi
+
+API = HfApi(token=os.environ.get("HF_TOKEN"))
+REPO_ID = "Limour/llama-python-streamingllm"
+
+
+def restart_space():
+    API.restart_space(repo_id=REPO_ID, token=os.environ.get("HF_TOKEN"))
+
+
+def init(cfg):
+    # ========== 共同 ==========
+    model = cfg['model']
+    s_info = cfg['s_info']
+
+    def btn_reset(_cache_path):
+        try:
+            with cfg['session_lock']:
+                _tmp = model.load_session(_cache_path)
+                print(f'load cache from {_cache_path} {_tmp}')
+                cfg['session_active'] = False
+                return model.venv_info
+        except Exception as e:
+            restart_space()
+            raise e
+
+    cfg['btn_reset'].click(
+        fn=btn_reset,
+        inputs=cfg['setting_cache_path'],
+        outputs=s_info
+    ).success(
+        **cfg['btn_finish']
+    )
+
+    cfg['btn_debug'].click(
+        fn=lambda: model.str_detokenize(model._input_ids),
+        outputs=cfg['vo']
+    )

mods/btn_submit.py

@@ -0,0 +1,76 @@
+def init(cfg):
+    # ========== 共同 ==========
+    model = cfg['model']
+    btn_com = cfg['btn_com']
+    s_info = cfg['s_info']
+    lock = cfg['session_lock']
+
+    # ========== 特殊 ==========
+    chat_template = cfg['chat_template']
+    msg = cfg['msg']
+    chatbot = cfg['chatbot']
+    chat_display_format = cfg['chat_display_format']
+
+    # ========== 显示用户消息 ==========
+    def btn_submit_usr(message: str, history):
+        # print('btn_submit_usr', message, history)
+        if history is None:
+            history = []
+        return "", history + [[message.strip(), '']]
+
+    # ========== 模型流式响应 ==========
+    def btn_submit_bot(history, _n_keep, _n_discard,
+                       _temperature, _repeat_penalty, _frequency_penalty,
+                       _presence_penalty, _repeat_last_n, _top_k,
+                       _top_p, _min_p, _typical_p,
+                       _tfs_z, _mirostat_mode, _mirostat_eta,
+                       _mirostat_tau, _usr, _char,
+                       _rag, _max_tokens):
+        with lock:
+            if not cfg['session_active']:
+                raise RuntimeError
+            # ========== 需要临时注入的内容 ==========
+            if len(_rag) > 0:
+                model.venv_create('rag')  # 记录 venv_idx
+                t_rag = chat_template('system', _rag)
+                model.eval_t(t_rag, _n_keep, _n_discard)
+            # ========== 释放不再需要的环境 ==========
+            model.venv_disband({'usr', 'char'})
+            print('venv_disband char', model.venv_info)
+            # ========== 用户输入 ==========
+            model.venv_create('usr')
+            t_msg = history[-1][0]
+            t_msg = chat_template(_usr, t_msg)
+            model.eval_t(t_msg, _n_keep, _n_discard)
+            yield history, model.venv_info
+            # ========== 模型输出 ==========
+            model.venv_create('char')
+            _tmp = btn_com(_n_keep, _n_discard,
+                           _temperature, _repeat_penalty, _frequency_penalty,
+                           _presence_penalty, _repeat_last_n, _top_k,
+                           _top_p, _min_p, _typical_p,
+                           _tfs_z, _mirostat_mode, _mirostat_eta,
+                           _mirostat_tau, _char, _max_tokens)
+            for _h in _tmp:
+                history[-1][1] = _h
+                yield history, model.venv_info
+            # ========== 输出完毕后格式化输出 ==========
+            history[-1][1] = chat_display_format(history[-1][1])
+            yield history, model.venv_info
+            # ========== 响应完毕后清除注入的内容 ==========
+            model.venv_remove('rag')  # 销毁对应的 venv
+            yield history, model.venv_info
+
+    cfg['btn_submit'].click(
+        **cfg['btn_start']
+    ).success(
+        fn=btn_submit_usr, api_name="submit",
+        inputs=[msg, chatbot],
+        outputs=[msg, chatbot]
+    ).success(
+        fn=btn_submit_bot,
+        inputs=[chatbot]+cfg['setting'],
+        outputs=[chatbot, s_info]
+    ).success(
+        **cfg['btn_finish']
+    )

mods/btn_suggest.py

@@ -0,0 +1,41 @@
+def init(cfg):
+    # ========== 共同 ==========
+    model = cfg['model']
+    btn_com = cfg['btn_com']
+    s_info = cfg['s_info']
+    lock = cfg['session_lock']
+
+    # ========== 给用户提供默认回复的建议 ==========
+    def btn_suggest(_n_keep, _n_discard,
+                    _temperature, _repeat_penalty, _frequency_penalty,
+                    _presence_penalty, _repeat_last_n, _top_k,
+                    _top_p, _min_p, _typical_p,
+                    _tfs_z, _mirostat_mode, _mirostat_eta,
+                    _mirostat_tau, _usr, _char,
+                    _rag, _max_tokens):
+        with lock:
+            if not cfg['session_active']:
+                raise RuntimeError
+            # ========== 模型输出建议 ==========
+            model.venv_create('suggest')  # 创建隔离环境
+            _tmp = btn_com(_n_keep, _n_discard,
+                           _temperature, _repeat_penalty, _frequency_penalty,
+                           _presence_penalty, _repeat_last_n, _top_k,
+                           _top_p, _min_p, _typical_p,
+                           _tfs_z, _mirostat_mode, _mirostat_eta,
+                           _mirostat_tau, _usr, _max_tokens)
+            _h = ''
+            for _h in _tmp:
+                yield _h, model.venv_info
+            model.venv_remove('suggest')  # 销毁隔离环境
+            yield _h, model.venv_info
+
+    cfg['btn_suggest'].click(
+        **cfg['btn_start']
+    ).success(
+        fn=btn_suggest,
+        inputs=cfg['setting'],
+        outputs=[cfg['msg'], s_info]
+    ).success(
+        **cfg['btn_finish']
+    )

mods/btn_vo.py

@@ -0,0 +1,41 @@
+def init(cfg):
+    # ========== 共同 ==========
+    model = cfg['model']
+    btn_com = cfg['btn_com']
+    s_info = cfg['s_info']
+    lock = cfg['session_lock']
+
+    # ========== 输出一段旁白 ==========
+    def btn_vo(_n_keep, _n_discard,
+               _temperature, _repeat_penalty, _frequency_penalty,
+               _presence_penalty, _repeat_last_n, _top_k,
+               _top_p, _min_p, _typical_p,
+               _tfs_z, _mirostat_mode, _mirostat_eta,
+               _mirostat_tau, _usr, _char,
+               _rag, _max_tokens):
+        with lock:
+            if not cfg['session_active']:
+                raise RuntimeError
+            # ========== 及时清理上一次生成的旁白 ==========
+            model.venv_remove('vo')
+            print('清理旁白', model.venv_info)
+            # ========== 模型输出旁白 ==========
+            model.venv_create('vo')  # 创建隔离环境
+            _tmp = btn_com(_n_keep, _n_discard,
+                           _temperature, _repeat_penalty, _frequency_penalty,
+                           _presence_penalty, _repeat_last_n, _top_k,
+                           _top_p, _min_p, _typical_p,
+                           _tfs_z, _mirostat_mode, _mirostat_eta,
+                           _mirostat_tau, '旁白', _max_tokens)
+            for _h in _tmp:
+                yield _h, model.venv_info
+
+    cfg['btn_vo'].click(
+        **cfg['btn_start']
+    ).success(
+        fn=btn_vo,
+        inputs=cfg['setting'],
+        outputs=[cfg['vo'], s_info]
+    ).success(
+        **cfg['btn_finish']
+    )

mods/load_cache.py

@@ -0,0 +1,70 @@
+import os
+
+
+def init(cfg):
+    if os.path.exists(cfg['setting_cache_path'].value):
+        # ========== 加载角色卡-缓存 ==========
+        tmp = cfg['model'].load_session(cfg['setting_cache_path'].value)
+        print(f"load cache from {cfg['setting_cache_path'].value} {tmp}")
+        tmp = cfg['chat_template']('system',
+                                   cfg['text_format'](cfg['role_char_d'].value,
+                                                      char=cfg['role_char'].value,
+                                                      user=cfg['role_usr'].value))
+        cfg['setting_n_keep'].value = len(tmp)
+        tmp = cfg['chat_template'](cfg['role_char'].value,
+                                   cfg['text_format'](cfg['role_chat_style'].value,
+                                                      char=cfg['role_char'].value,
+                                                      user=cfg['role_usr'].value))
+        cfg['setting_n_keep'].value += len(tmp)
+        # ========== 加载角色卡-第一条消息 ==========
+        cfg['chatbot'] = []
+        for one in cfg["role_char_first"]:
+            one['name'] = cfg['text_format'](one['name'],
+                                             char=cfg['role_char'].value,
+                                             user=cfg['role_usr'].value)
+            one['value'] = cfg['text_format'](one['value'],
+                                              char=cfg['role_char'].value,
+                                              user=cfg['role_usr'].value)
+            if one['name'] == cfg['role_char'].value:
+                cfg['chatbot'].append((None, cfg['chat_display_format'](one['value'])))
+            print(one)
+    else:
+        # ========== 加载角色卡-角色描述 ==========
+        tmp = cfg['chat_template']('system',
+                                   cfg['text_format'](cfg['role_char_d'].value,
+                                                      char=cfg['role_char'].value,
+                                                      user=cfg['role_usr'].value))
+        cfg['setting_n_keep'].value = cfg['model'].eval_t(tmp)  # 此内容永久存在
+
+        # ========== 加载角色卡-回复示例 ==========
+        tmp = cfg['chat_template'](cfg['role_char'].value,
+                                   cfg['text_format'](cfg['role_chat_style'].value,
+                                                      char=cfg['role_char'].value,
+                                                      user=cfg['role_usr'].value))
+        cfg['setting_n_keep'].value += cfg['model'].eval_t(tmp)  # 此内容永久存在
+
+        # ========== 加载角色卡-第一条消息 ==========
+        cfg['chatbot'] = []
+        for one in cfg["role_char_first"]:
+            one['name'] = cfg['text_format'](one['name'],
+                                             char=cfg['role_char'].value,
+                                             user=cfg['role_usr'].value)
+            one['value'] = cfg['text_format'](one['value'],
+                                              char=cfg['role_char'].value,
+                                              user=cfg['role_usr'].value)
+            if one['name'] == cfg['role_char'].value:
+                cfg['chatbot'].append((None, cfg['chat_display_format'](one['value'])))
+            print(one)
+            tmp = cfg['chat_template'](one['name'], one['value'])
+            cfg['model'].eval_t(tmp)  # 此内容随上下文增加将被丢弃
+
+        # ========== 保存角色卡-缓存 ==========
+        with open(cfg['setting_cache_path'].value, 'wb') as f:
+            pass
+        tmp = cfg['model'].save_session(cfg['setting_cache_path'].value)
+        print(f'save cache {tmp}')
+        # ========== 上传缓存 ==========
+        if os.environ.get("HF_TOKEN"):
+            from huggingface_hub import login, CommitScheduler
+            login(token=os.environ.get("HF_TOKEN"), write_permission=True)
+            CommitScheduler(repo_id='Limour/llama-python-streamingllm-cache', repo_type='dataset', folder_path='cache')

mods/read_cfg.py

@@ -0,0 +1,16 @@
+import hashlib
+import json
+
+
+# ========== 哈希函数 ==========
+def x_hash(x: str):
+    return hashlib.sha1(x.encode('utf-8')).hexdigest()
+
+
+# ========== 读取配置文件 ==========
+with open('rp_config.json', encoding='utf-8') as f:
+    tmp = f.read()
+with open('rp_sample_config.json', encoding='utf-8') as f:
+    cfg = json.load(f)
+cfg['setting_cache_path']['value'] += x_hash(tmp)
+cfg.update(json.loads(tmp))

mods/text_display.py

@@ -0,0 +1,24 @@
+import re
+
+
+#  ========== 适配 SillyTavern 的模版 ==========
+def text_format(text: str, _env=None, **env):
+    if _env is not None:
+        for k, v in _env.items():
+            text = text.replace(r'{{' + k + r'}}', v)
+    for k, v in env.items():
+        text = text.replace(r'{{' + k + r'}}', v)
+    return text
+
+
+#  ========== 给引号加粗 ==========
+reg_q = re.compile(r'“(.+?)”')
+
+
+def chat_display_format(text: str):
+    return reg_q.sub(r' **\g<0>** ', text)
+
+
+def init(cfg):
+    cfg['text_format'] = text_format
+    cfg['chat_display_format'] = chat_display_format