调整根目录结构，kb/logs/media/nltk_data 移动到专用数据目录（可配置，默认 data）。注意知识库文件要做相应移动

.gitignore

@@ -3,9 +3,10 @@
 *.bak
 logs
 /media/
-/knowledge_base/*
-!/knowledge_base/samples
-/knowledge_base/samples/vector_store
+/data/*
+!/data/knowledge_base/samples
+/data/knowledge_base/samples/vector_store
+!/data/nltk_data
 
 /configs/*.py
 /configs/loom.yaml

.gitmodules

@@ -1,3 +1,3 @@
 [submodule "knowledge_base/samples/content/wiki"]
-	path = knowledge_base/samples/content/wiki
+	path = data/knowledge_base/samples/content/wiki
 	url = https://github.com/chatchat-space/Langchain-Chatchat.wiki.git

openai_plugins/imitater/app.py

@@ -1,94 +0,0 @@
-import multiprocessing as mp
-from multiprocessing import Process
-from typing import List
-
-from loom_core.constants import LOOM_LOG_BACKUP_COUNT, LOOM_LOG_MAX_BYTES
-from loom_core.openai_plugins.core.adapter import ProcessesInfo
-from loom_core.openai_plugins.core.application import ApplicationAdapter
-
-import os
-import sys
-import logging
-
-from loom_core.openai_plugins.deploy.utils import get_timestamp_ms, get_config_dict, get_log_file
-from omegaconf import OmegaConf
-
-logger = logging.getLogger(__name__)
-# 为了能使用插件中的函数，需要将当前目录加入到sys.path中
-root_dir = os.path.dirname(os.path.abspath(__file__))
-sys.path.append(root_dir)
-import imitater_process_dict
-from imitater_config import ImitaterCfg
-from imitater_wrapper import run_imitater
-
-
-class ImitaterApplicationAdapter(ApplicationAdapter):
-    model_worker_started: mp.Event = None
-
-    def __init__(self, cfg=None, state_dict: dict = None):
-        self.processesInfo = None
-        self._cfg = ImitaterCfg(cfg=cfg)
-        super().__init__(state_dict=state_dict)
-
-    def class_name(self) -> str:
-        """Get class name."""
-        return self.__name__
-
-    @classmethod
-    def from_config(cls, cfg=None):
-        _state_dict = {
-            "application_name": "Imitate",
-            "application_version": "0.0.1",
-            "application_description": "Imitate application",
-            "application_author": "Imitate"
-        }
-        state_dict = cfg.get("state_dict", {})
-        if state_dict is not None and _state_dict is not None:
-            _state_dict = {**state_dict, **_state_dict}
-        else:
-            # 处理其中一个或两者都为 None 的情况
-            _state_dict = state_dict or _state_dict or {}
-        return cls(cfg=cfg, state_dict=_state_dict)
-
-    def init_processes(self, processesInfo: ProcessesInfo):
-
-        self.processesInfo = processesInfo
-
-        logging_conf = get_config_dict(
-            processesInfo.log_level,
-            get_log_file(log_path=self._cfg.get_cfg().get("logdir"), sub_dir=f"local_{get_timestamp_ms()}"),
-            LOOM_LOG_BACKUP_COUNT,
-            LOOM_LOG_MAX_BYTES,
-        )
-        logging.config.dictConfig(logging_conf)  # type: ignore
-        worker_name = self._cfg.get_cfg().get("worker_name", [])
-        imitater_process_dict.mp_manager = mp.Manager()
-
-        # prevent re-init cuda error.
-        mp.set_start_method(method="spawn", force=True)
-
-        self.model_worker_started = imitater_process_dict.mp_manager.Event()
-
-        process = Process(
-            target=run_imitater,
-            name=f"model_worker - {worker_name}",
-            kwargs=dict(cfg=self._cfg,
-                        worker_name=worker_name,
-                        started_event=self.model_worker_started,
-                        logging_conf=logging_conf),
-            daemon=True,
-        )
-        imitater_process_dict.processes[worker_name] = process
-
-    def start(self):
-
-        for n, p in imitater_process_dict.processes.items():
-            p.start()
-            p.name = f"{p.name} ({p.pid})"
-
-        # 等待 model_worker启动完成
-        # self.model_worker_started.wait()
-
-    def stop(self):
-        imitater_process_dict.stop()
-

openai_plugins/imitater/controller.py

@@ -1,80 +0,0 @@
-import time
-from multiprocessing import Process
-
-from loom_core.openai_plugins.core.control import ControlAdapter
-
-import os
-import sys
-import logging
-
-logger = logging.getLogger(__name__)
-# 为了能使用插件中的函数，需要将当前目录加入到sys.path中
-root_dir = os.path.dirname(os.path.abspath(__file__))
-sys.path.append(root_dir)
-
-import imitater_process_dict
-
-from imitater_wrapper import run_imitater
-from imitater_config import ImitaterCfg
-
-
-class ImitaterControlAdapter(ControlAdapter):
-
-    def __init__(self, cfg=None, state_dict: dict = None):
-        self._cfg = ImitaterCfg(cfg=cfg)
-        super().__init__(state_dict=state_dict)
-
-    def class_name(self) -> str:
-        """Get class name."""
-        return self.__name__
-
-    def start_model(self, new_model_name):
-
-        imitater_process_dict.stop()
-
-        logger.info(f"准备启动新进程：{new_model_name}")
-        e = imitater_process_dict.mp_manager.Event()
-        process = Process(
-            target=run_imitater,
-            name=f"model_worker - {new_model_name}",
-            kwargs=dict(cfg=self._cfg,
-                        worker_name=new_model_name,
-                        started_event=e),
-            daemon=True,
-        )
-        process.start()
-        process.name = f"{process.name} ({process.pid})"
-        imitater_process_dict.processes[new_model_name] = process
-        # e.wait()
-        logger.info(f"成功启动新进程：{new_model_name}")
-
-    def stop_model(self, model_name: str):
-
-        if model_name in imitater_process_dict.processes:
-            process = imitater_process_dict.processes.pop(model_name)
-            time.sleep(1)
-            process.kill()
-            logger.info(f"停止进程：{model_name}")
-        else:
-            logger.error(f"未找到进程：{model_name}")
-            raise Exception(f"未找到进程：{model_name}")
-
-    def replace_model(self, model_name: str, new_model_name: str):
-        pass
-
-    @classmethod
-    def from_config(cls, cfg=None):
-        _state_dict = {
-            "controller_name": "Imitate",
-            "controller_version": "0.0.1",
-            "controller_description": "Imitate controller",
-            "controller_author": "Imitate"
-        }
-        state_dict = cfg.get("state_dict", {})
-        if state_dict is not None and _state_dict is not None:
-            _state_dict = {**state_dict, **_state_dict}
-        else:
-            # 处理其中一个或两者都为 None 的情况
-            _state_dict = state_dict or _state_dict or {}
-
-        return cls(cfg=cfg, state_dict=_state_dict)

openai_plugins/imitater/generation_config/internlm2/generation_config.json

@@ -1,8 +0,0 @@
-{
-  "bos_token_id": 1,
-  "eos_token_id": [2, 92542],
-  "pad_token_id": 2,
-  "max_new_tokens": 1024,
-  "do_sample": true,
-  "transformers_version": "4.33.2"
-}

openai_plugins/imitater/generation_config/qwen/generation_config.json

@@ -1,9 +0,0 @@
-{
-  "eos_token_id": [151643, 151645],
-  "pad_token_id": 151643,
-  "max_new_tokens": 1024,
-  "do_sample": true,
-  "top_k": 0,
-  "top_p": 0.8,
-  "transformers_version": "4.34.0"
-}

openai_plugins/imitater/imitater_config.py

@@ -1,80 +0,0 @@
-# httpx 请求默认超时时间（秒）。如果加载模型或对话较慢，出现超时错误，可以适当加大该值。
-from typing import List, TypedDict
-
-HTTPX_DEFAULT_TIMEOUT = 300.0
-log_verbose = True
-
-
-class ImitaterModel(TypedDict):
-    name: str
-    chat_model_path: str
-    chat_model_device: str
-    chat_template_path: str
-    generation_config_path: str
-    agent_type: str
-
-
-class ImitaterEmbedding(TypedDict):
-    name: str
-    embed_model_path: str
-    embed_model_device: str
-    embed_batch_size: int
-
-
-class ImitaterWorker(TypedDict):
-    name: str
-    model: ImitaterModel
-    embedding: ImitaterEmbedding
-
-
-class ImitaterCfg:
-    def __init__(self, cfg: dict = None):
-        if cfg is None:
-            raise RuntimeError("ImitaterCfg cfg is None.")
-        self._cfg = cfg
-
-    def get_cfg(self):
-        return self._cfg
-
-    def get_run_openai_api_cfg(self):
-        return self._cfg.get("run_openai_api", {})
-
-    def get_imitate_model_workers_by_name(self, name: str) -> ImitaterWorker:
-
-        imitate_model_workers_cfg = self._cfg.get("imitate_model_workers", None)
-        if imitate_model_workers_cfg is None:
-            raise RuntimeError("imitate_model_workers_cfg is None.")
-
-        get = lambda model_name: imitate_model_workers_cfg[
-            self.get_imitate_model_workers_index_by_name(model_name)
-        ].get(model_name, {})
-        imitate = get(name)
-        # 初始化imitate为ImitaterWorker
-        worker_cfg = ImitaterWorker(name=name,
-                                    model=ImitaterModel(**imitate.get("model", {})),
-                                    embedding=ImitaterEmbedding(**imitate.get("embedding", {}))
-                                    )
-        return worker_cfg
-
-    def get_imitate_model_workers_names(self) -> List[str]:
-
-        imitate_model_workers_cfg = self._cfg.get("imitate_model_workers", None)
-        if imitate_model_workers_cfg is None:
-            raise RuntimeError("imitate_model_workers_cfg is None.")
-        worker_name_cfg = []
-        for cfg in imitate_model_workers_cfg:
-            for key, imitate_model_workers in cfg.items():
-                worker_name_cfg.append(key)
-        return worker_name_cfg
-
-    def get_imitate_model_workers_index_by_name(self, name) -> int:
-
-        imitate_model_workers_cfg = self._cfg.get("imitate_model_workers", None)
-        if imitate_model_workers_cfg is None:
-            raise RuntimeError("imitate_model_workers_cfg is None.")
-
-        for cfg in imitate_model_workers_cfg:
-            for key, imitate_model_workers in cfg.items():
-                if key == name:
-                    return imitate_model_workers_cfg.index(cfg)
-        return -1

openai_plugins/imitater/imitater_process_dict.py

@@ -1,22 +0,0 @@
-from multiprocessing import Process
-from typing import Dict
-import logging
-
-logger = logging.getLogger(__name__)
-mp_manager = None
-processes: Dict[str, Process] = {}
-
-
-def stop():
-    for n, process in processes.items():
-        logger.warning("Sending SIGKILL to %s", p)
-        try:
-
-            process.kill()
-        except Exception as e:
-            logger.info("Failed to kill process %s", p, exc_info=True)
-
-    for n, p in processes.items():
-        logger.info("Process status: %s", p)
-
-    del processes

openai_plugins/imitater/imitater_wrapper.py

@@ -1,60 +0,0 @@
-import multiprocessing
-from typing import List, Optional, Dict
-
-from fastapi import FastAPI
-import sys
-
-import multiprocessing as mp
-import uvicorn
-import os
-import logging
-import asyncio
-import signal
-import inspect
-
-logger = logging.getLogger(__name__)
-# 为了能使用fastchat_wrapper.py中的函数，需要将当前目录加入到sys.path中
-root_dir = os.path.dirname(os.path.abspath(__file__))
-sys.path.append(root_dir)
-
-from imitater_config import ImitaterCfg
-
-"""
-防止Can't pickle Function
-"""
-
-
-def _start_imitater(
-        started_event: mp.Event = None
-):
-    from imitater.service.app import launch_app
-    # 跳过键盘中断，
-    signal.signal(signal.SIGINT, lambda *_: None)
-    launch_app()
-
-
-def run_imitater(
-        cfg: ImitaterCfg,
-        worker_name: str,
-        started_event: mp.Event = None,
-        logging_conf: Optional[dict] = None):
-    # 跳过键盘中断，
-    signal.signal(signal.SIGINT, lambda *_: None)
-
-    logging.config.dictConfig(logging_conf)  # type: ignore
-    import os
-    worker_cfg = cfg.get_imitate_model_workers_by_name(worker_name)
-
-    os.environ["AGENT_TYPE"] = worker_cfg.get("model").get("agent_type")
-    os.environ["CHAT_MODEL_PATH"] = worker_cfg.get("model").get("chat_model_path")
-    os.environ["CHAT_MODEL_DEVICE"] = worker_cfg.get("model").get("chat_model_device")
-    os.environ["CHAT_TEMPLATE_PATH"] = worker_cfg.get("model").get("chat_template_path")
-    os.environ["GENERATION_CONFIG_PATH"] = worker_cfg.get("model").get("generation_config_path")
-
-    os.environ["EMBED_MODEL_PATH"] = worker_cfg.get("embedding").get("embed_model_path")
-    os.environ["EMBED_MODEL_DEVICE"] = worker_cfg.get("embedding").get("embed_model_device")
-    os.environ["EMBED_BATCH_SIZE"] = str(worker_cfg.get("embedding").get("embed_batch_size"))
-
-    os.environ["SERVICE_PORT"] = str(cfg.get_run_openai_api_cfg().get("port", 30000))
-
-    _start_imitater(started_event=started_event)