Merge remote-tracking branch 'upstream/r-dev' into sync/pr-1564-upstream-20260331
# Conflicts: # src/chat/brain_chat/PFC/conversation.py # src/chat/brain_chat/PFC/pfc_KnowledgeFetcher.py # src/chat/knowledge/lpmm_ops.py
This commit is contained in:
A-Dawn
@@ -139,14 +139,14 @@ class EmbeddingStore:
|
||||
asyncio.set_event_loop(loop)
|
||||
|
||||
try:
|
||||
# 创建新的LLMRequest实例
|
||||
from src.llm_models.utils_model import LLMRequest
|
||||
from src.config.config import model_config
|
||||
# 创建新的服务层实例
|
||||
from src.services.llm_service import LLMServiceClient
|
||||
|
||||
llm = LLMRequest(model_set=model_config.model_task_config.embedding, request_type="embedding")
|
||||
llm = LLMServiceClient(task_name="embedding", request_type="embedding")
|
||||
|
||||
# 使用新的事件循环运行异步方法
|
||||
embedding, _ = loop.run_until_complete(llm.get_embedding(s))
|
||||
embedding_result = loop.run_until_complete(llm.embed_text(s))
|
||||
embedding = embedding_result.embedding
|
||||
|
||||
if embedding and len(embedding) > 0:
|
||||
return embedding
|
||||
@@ -195,13 +195,12 @@ class EmbeddingStore:
|
||||
start_idx, chunk_strs = chunk_data
|
||||
chunk_results = []
|
||||
|
||||
# 为每个线程创建独立的LLMRequest实例
|
||||
from src.llm_models.utils_model import LLMRequest
|
||||
from src.config.config import model_config
|
||||
# 为每个线程创建独立的服务层实例
|
||||
from src.services.llm_service import LLMServiceClient
|
||||
|
||||
try:
|
||||
# 创建线程专用的LLM实例
|
||||
llm = LLMRequest(model_set=model_config.model_task_config.embedding, request_type="embedding")
|
||||
# 创建线程专用的服务层实例
|
||||
llm = LLMServiceClient(task_name="embedding", request_type="embedding")
|
||||
|
||||
for i, s in enumerate(chunk_strs):
|
||||
try:
|
||||
@@ -209,7 +208,8 @@ class EmbeddingStore:
|
||||
loop = asyncio.new_event_loop()
|
||||
asyncio.set_event_loop(loop)
|
||||
try:
|
||||
embedding = loop.run_until_complete(llm.get_embedding(s))
|
||||
embedding_result = loop.run_until_complete(llm.embed_text(s))
|
||||
embedding = embedding_result.embedding
|
||||
finally:
|
||||
loop.close()
|
||||
|
||||
|
||||
@@ -1,18 +1,27 @@
|
||||
import asyncio
|
||||
import json
|
||||
import time
|
||||
from typing import List, Union
|
||||
from typing import Dict, List, Tuple, Union
|
||||
|
||||
from .global_logger import logger
|
||||
from . import prompt_template
|
||||
from . import INVALID_ENTITY
|
||||
from src.llm_models.utils_model import LLMRequest
|
||||
from json_repair import repair_json
|
||||
|
||||
from src.services.llm_service import LLMServiceClient
|
||||
|
||||
def _extract_json_from_text(text: str):
|
||||
from . import INVALID_ENTITY
|
||||
from . import prompt_template
|
||||
from .global_logger import logger
|
||||
|
||||
|
||||
def _extract_json_from_text(text: str) -> List[str] | List[List[str]] | Dict[str, object]:
|
||||
# sourcery skip: assign-if-exp, extract-method
|
||||
"""从文本中提取JSON数据的高容错方法"""
|
||||
"""从文本中提取 JSON 数据。
|
||||
|
||||
Args:
|
||||
text: 原始模型输出文本。
|
||||
|
||||
Returns:
|
||||
List[str] | List[List[str]] | Dict[str, object]: 修复并解析后的 JSON 结果。
|
||||
"""
|
||||
if text is None:
|
||||
logger.error("输入文本为None")
|
||||
return []
|
||||
@@ -46,20 +55,30 @@ def _extract_json_from_text(text: str):
|
||||
return []
|
||||
|
||||
|
||||
def _entity_extract(llm_req: LLMRequest, paragraph: str) -> List[str]:
|
||||
def _entity_extract(llm_req: LLMServiceClient, paragraph: str) -> List[str]:
|
||||
# sourcery skip: reintroduce-else, swap-if-else-branches, use-named-expression
|
||||
"""对段落进行实体提取，返回提取出的实体列表（JSON格式）"""
|
||||
"""对单段文本执行实体提取。
|
||||
|
||||
Args:
|
||||
llm_req: LLM 服务门面实例。
|
||||
paragraph: 待提取实体的原始段落文本。
|
||||
|
||||
Returns:
|
||||
List[str]: 提取出的实体列表。
|
||||
"""
|
||||
entity_extract_context = prompt_template.build_entity_extract_context(paragraph)
|
||||
|
||||
# 使用 asyncio.run 来运行异步方法
|
||||
try:
|
||||
# 如果当前已有事件循环在运行，使用它
|
||||
loop = asyncio.get_running_loop()
|
||||
future = asyncio.run_coroutine_threadsafe(llm_req.generate_response_async(entity_extract_context), loop)
|
||||
response, _ = future.result()
|
||||
future = asyncio.run_coroutine_threadsafe(llm_req.generate_response(entity_extract_context), loop)
|
||||
generation_result = future.result()
|
||||
response = generation_result.response
|
||||
except RuntimeError:
|
||||
# 如果没有运行中的事件循环，直接使用 asyncio.run
|
||||
response, _ = asyncio.run(llm_req.generate_response_async(entity_extract_context))
|
||||
generation_result = asyncio.run(llm_req.generate_response(entity_extract_context))
|
||||
response = generation_result.response
|
||||
|
||||
# 添加调试日志
|
||||
logger.debug(f"LLM返回的原始响应: {response}")
|
||||
@@ -92,8 +111,21 @@ def _entity_extract(llm_req: LLMRequest, paragraph: str) -> List[str]:
|
||||
return entity_extract_result
|
||||
|
||||
|
||||
def _rdf_triple_extract(llm_req: LLMRequest, paragraph: str, entities: list) -> List[List[str]]:
|
||||
"""对段落进行实体提取，返回提取出的实体列表（JSON格式）"""
|
||||
def _rdf_triple_extract(
|
||||
llm_req: LLMServiceClient,
|
||||
paragraph: str,
|
||||
entities: List[str],
|
||||
) -> List[List[str]]:
|
||||
"""对单段文本执行 RDF 三元组提取。
|
||||
|
||||
Args:
|
||||
llm_req: LLM 服务门面实例。
|
||||
paragraph: 待提取的原始段落文本。
|
||||
entities: 已识别出的实体列表。
|
||||
|
||||
Returns:
|
||||
List[List[str]]: 提取出的三元组列表。
|
||||
"""
|
||||
rdf_extract_context = prompt_template.build_rdf_triple_extract_context(
|
||||
paragraph, entities=json.dumps(entities, ensure_ascii=False)
|
||||
)
|
||||
@@ -102,11 +134,13 @@ def _rdf_triple_extract(llm_req: LLMRequest, paragraph: str, entities: list) ->
|
||||
try:
|
||||
# 如果当前已有事件循环在运行，使用它
|
||||
loop = asyncio.get_running_loop()
|
||||
future = asyncio.run_coroutine_threadsafe(llm_req.generate_response_async(rdf_extract_context), loop)
|
||||
response, _ = future.result()
|
||||
future = asyncio.run_coroutine_threadsafe(llm_req.generate_response(rdf_extract_context), loop)
|
||||
generation_result = future.result()
|
||||
response = generation_result.response
|
||||
except RuntimeError:
|
||||
# 如果没有运行中的事件循环，直接使用 asyncio.run
|
||||
response, _ = asyncio.run(llm_req.generate_response_async(rdf_extract_context))
|
||||
generation_result = asyncio.run(llm_req.generate_response(rdf_extract_context))
|
||||
response = generation_result.response
|
||||
|
||||
# 添加调试日志
|
||||
logger.debug(f"RDF LLM返回的原始响应: {response}")
|
||||
@@ -140,8 +174,21 @@ def _rdf_triple_extract(llm_req: LLMRequest, paragraph: str, entities: list) ->
|
||||
|
||||
|
||||
def info_extract_from_str(
|
||||
llm_client_for_ner: LLMRequest, llm_client_for_rdf: LLMRequest, paragraph: str
|
||||
) -> Union[tuple[None, None], tuple[list[str], list[list[str]]]]:
|
||||
llm_client_for_ner: LLMServiceClient,
|
||||
llm_client_for_rdf: LLMServiceClient,
|
||||
paragraph: str,
|
||||
) -> Union[Tuple[None, None], Tuple[List[str], List[List[str]]]]:
|
||||
"""从文本中提取实体与三元组信息。
|
||||
|
||||
Args:
|
||||
llm_client_for_ner: 实体提取使用的 LLM 服务门面。
|
||||
llm_client_for_rdf: RDF 三元组提取使用的 LLM 服务门面。
|
||||
paragraph: 原始段落文本。
|
||||
|
||||
Returns:
|
||||
Union[Tuple[None, None], Tuple[List[str], List[List[str]]]]: 成功时返回
|
||||
``(实体列表, 三元组列表)``，失败时返回 ``(None, None)``。
|
||||
"""
|
||||
try_count = 0
|
||||
while True:
|
||||
try:
|
||||
@@ -176,17 +223,30 @@ def info_extract_from_str(
|
||||
|
||||
|
||||
class IEProcess:
|
||||
"""
|
||||
信息抽取处理器类，提供更方便的批次处理接口。
|
||||
"""
|
||||
"""信息抽取处理器。"""
|
||||
|
||||
def __init__(self, llm_ner: LLMRequest, llm_rdf: LLMRequest = None):
|
||||
def __init__(
|
||||
self,
|
||||
llm_ner: LLMServiceClient,
|
||||
llm_rdf: LLMServiceClient | None = None,
|
||||
) -> None:
|
||||
"""初始化信息抽取处理器。
|
||||
|
||||
Args:
|
||||
llm_ner: 实体提取使用的 LLM 服务门面。
|
||||
llm_rdf: RDF 三元组提取使用的 LLM 服务门面；为空时复用 `llm_ner`。
|
||||
"""
|
||||
self.llm_ner = llm_ner
|
||||
self.llm_rdf = llm_rdf or llm_ner
|
||||
|
||||
async def process_paragraphs(self, paragraphs: List[str]) -> List[dict]:
|
||||
"""
|
||||
异步处理多个段落。
|
||||
async def process_paragraphs(self, paragraphs: List[str]) -> List[Dict[str, object]]:
|
||||
"""异步处理多个段落。
|
||||
|
||||
Args:
|
||||
paragraphs: 待处理的段落列表。
|
||||
|
||||
Returns:
|
||||
List[Dict[str, object]]: 每个成功段落对应的抽取结果。
|
||||
"""
|
||||
from .utils.hash import get_sha256
|
||||
|
||||
|
||||
Reference in New Issue
Block a user