lizhilun/AI_NovelGenerator
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

完成对生成逻辑的优化，但也发现新的问题

Browse Source 
前后章节的衔接有问题，这里想着应该要把前一个章节内容发送作为参考，不然中断感很重
This commit is contained in:
YILING0013
2025-02-05 22:43:09 +08:00
parent dd78666071
commit f388403b08
3 changed files with 76 additions and 57 deletions
Download Patch File Download Diff File Expand all files Collapse all files
chapter_directory_parser.py
+15 -9
View File
@@ -21,13 +21,19 @@ def parse_chapter_blueprint(blueprint_text: str):
chunks = re.split(r'\n\s*\n', blueprint_text.strip())
results = []
chapter_number_pattern = re.compile(r'^第\s*(\d+)\s*章\s*-\s*\[(.*?)\]') # 捕获章号与标题
role_pattern = re.compile(r'^本章定位：\s*(.*)$')
purpose_pattern = re.compile(r'^核心作用：\s*(.*)$')
suspense_pattern = re.compile(r'^悬念密度：\s*(.*)$')
foreshadow_pattern = re.compile(r'^伏笔操作：\s*(.*)$')
twist_pattern = re.compile(r'^认知颠覆：\s*(.*)$')
summary_pattern = re.compile(r'^本章简述：\s*\[(.*)\]$')
# 兼容是否使用方括号包裹章节标题
# 例如：
# 第1章 - 紫极光下的预兆
# 或
# 第1章 - [紫极光下的预兆]
chapter_number_pattern = re.compile(r'^第\s*(\d+)\s*章\s*-\s*\[?(.*?)\]?$')
role_pattern = re.compile(r'^本章定位：\s*\[?(.*)\]?$')
purpose_pattern = re.compile(r'^核心作用：\s*\[?(.*)\]?$')
suspense_pattern = re.compile(r'^悬念密度：\s*\[?(.*)\]?$')
foreshadow_pattern = re.compile(r'^伏笔操作：\s*\[?(.*)\]?$')
twist_pattern = re.compile(r'^认知颠覆：\s*\[?(.*)\]?$')
summary_pattern = re.compile(r'^本章简述：\s*\[?(.*)\]?$')
for chunk in chunks:
lines = chunk.strip().splitlines()
@@ -44,9 +50,9 @@ def parse_chapter_blueprint(blueprint_text: str):
chapter_summary = ""
# 先匹配第一行（或前几行），找到章号和标题
header_match = chapter_number_pattern.match(lines[0].strip()) if lines else None
header_match = chapter_number_pattern.match(lines[0].strip())
if not header_match:
# 不符合格式，跳过
# 不符合“第X章 - 标题”的格式，跳过
continue
chapter_number = int(header_match.group(1))
novel_generator.py
+34 -39
View File
@@ -431,8 +431,6 @@ def Chapter_blueprint_generate(
return
# 从内容中尽量提取 number_of_chapters
# 如果之前已经存储了 number_of_chapters，可以在外面传入，这里做简化：
# 这里用正则或者其他逻辑提取，但演示时直接写 10 也可
match_chaps = re.search(r'约(\d+)章', architecture_text)
if match_chaps:
number_of_chapters = int(match_chaps.group(1))
@@ -440,10 +438,7 @@ def Chapter_blueprint_generate(
number_of_chapters = 10 # fallback
# 提取三幕式文本
# 在写入时，我们将 4) 三幕式情节架构 作为传给 prompt 的核心
# 这里做一个简易匹配
plot_arch_text = ""
# 假设 "#=== 4) 三幕式情节架构 ===" 是分隔点
pat_plot = r'#=== 4\) 三幕式情节架构 ===\n([\s\S]+)$'
m = re.search(pat_plot, architecture_text)
if m:
@@ -556,7 +551,6 @@ def update_plot_arcs(
# ========== 3) 生成章节草稿 ==========
def generate_chapter_draft(
api_key: str,
base_url: str,
@@ -570,6 +564,10 @@ def generate_chapter_draft(
key_items: str,
scene_location: str,
time_constraint: str,
embedding_api_key: str,
embedding_url: str,
embedding_interface_format: str,
embedding_model_name: str,
embedding_retrieval_k: int = 2
) -> str:
"""
@@ -577,7 +575,7 @@ def generate_chapter_draft(
- novel_architecture 取自 Novel_architecture.txt
- blueprint 取自 Novel_directory.txt
- global_summary, character_state 分别取自全局摘要、角色状态文件
- 向量库检索上下文
- 从向量库检索上下文（embedding_*参数）
- 用户还可以额外提供四个可选元素：核心人物、关键道具、空间坐标、时间压力
"""
@@ -609,25 +607,21 @@ def generate_chapter_draft(
recent_3_texts = get_last_n_chapters_text(chapters_dir, novel_number, n=3)
merged_query_str = "回顾剧情：\n" + "\n".join(recent_3_texts) + "\n" + user_guidance
# 4) 检索向量库上下文
# 4) 检索向量库上下文 (使用embedding_*参数)
relevant_context = get_relevant_context_from_vector_store(
api_key=api_key,
base_url=base_url,
api_key=embedding_api_key,
base_url=embedding_url,
query=merged_query_str,
embedding_model_name=model_name,
interface_format=embedding_interface_format,
embedding_model_name=embedding_model_name,
filepath=filepath,
k=embedding_retrieval_k
)
if not relevant_context.strip():
relevant_context = "（无检索到的上下文）"
# 5) 构造prompt，调用 scene_dynamics_prompt
# 在这里，我们拆分架构文本，以便给模型提供：
# - “世界观”与“小说设定”可以从 arch_file 中的相应片段读取
# 这里为了简化，直接把 novel_architecture_text 整体塞入 novel_setting
# 也可更精细地拆分 "#=== 3) 世界观 ===" 片段给 world_building
# 下方仅作示例。
# 5) 构造prompt
# 拆分 world_building_text, novel_architecture_text 等等
world_building_text = ""
match_world = re.search(r'#=== 3\) 世界观 ===\n([\s\S]+?)\n#===', novel_architecture_text)
if match_world:
@@ -658,9 +652,8 @@ def generate_chapter_draft(
character_state=character_state_text
)
# 因为我们还想让模型了解向量库检索到的上下文，可以合并到最后
# 合并检索到的上下文和用户指导
prompt_text += f"\n\n【检索到的上下文】\n{relevant_context}"
# 也可合并用户指导
prompt_text += f"\n\n【用户指导】\n{user_guidance}\n"
model = ChatOpenAI(
@@ -674,8 +667,7 @@ def generate_chapter_draft(
if not chapter_content.strip():
logging.warning("Generated chapter draft is empty.")
# 6) 写入 chapters 目录
chapters_dir = os.path.join(filepath, "chapters")
# 6) 写入 chapters
os.makedirs(chapters_dir, exist_ok=True)
chapter_file = os.path.join(chapters_dir, f"chapter_{novel_number}.txt")
@@ -695,6 +687,9 @@ def finalize_chapter(
model_name: str,
temperature: float,
filepath: str,
embedding_api_key: str,
embedding_url: str,
embedding_interface_format: str,
embedding_model_name: str
):
"""
@@ -707,7 +702,7 @@ def finalize_chapter(
logging.warning(f"Chapter {novel_number} is empty, cannot finalize.")
return
# 如果长度比目标少很多，可考虑在此扩写
# 若篇幅过短，可尝试扩写
if len(chapter_text) < 0.6 * word_number:
chapter_text = enrich_chapter_text(chapter_text, word_number, api_key, base_url, model_name, temperature)
clear_file_content(chapter_file)
@@ -750,12 +745,13 @@ def finalize_chapter(
clear_file_content(character_state_file)
save_string_to_txt(new_char_state, character_state_file)
# 3) 更新向量库
# 3) 更新向量库 (embedding相关)
update_vector_store(
api_key=api_key,
base_url=base_url,
api_key=embedding_api_key,
base_url=embedding_url,
new_chapter=chapter_text,
model_name=embedding_model_name, # 用于embedding
interface_format=embedding_interface_format,
embedding_model_name=embedding_model_name,
filepath=filepath
)
@@ -827,15 +823,14 @@ def advanced_split_content(content: str,
return final_segments
def import_knowledge_file(
api_key: str,
base_url: str,
interface_format: str,
embedding_api_key: str,
embedding_url: str,
embedding_interface_format: str,
embedding_model_name: str,
file_path: str,
embedding_base_url: str,
filepath: str
):
logging.info(f"开始导入知识库文件: {file_path}, 接口格式: {interface_format}, 模型: {embedding_model_name}")
logging.info(f"开始导入知识库文件: {file_path}, 接口格式: {embedding_interface_format}, 模型: {embedding_model_name}")
if not os.path.exists(file_path):
logging.warning(f"知识库文件不存在: {file_path}")
return
@@ -847,20 +842,20 @@ def import_knowledge_file(
paragraphs = advanced_split_content(content)
# 若向量库不存在则初始化，否则追加
# 尝试加载已有的向量库
store = load_vector_store(
api_key=api_key,
base_url=base_url if base_url else "http://localhost:11434/v1",
interface_format=interface_format,
api_key=embedding_api_key,
base_url=embedding_url if embedding_url else "http://localhost:11434/v1",
interface_format=embedding_interface_format,
embedding_model_name=embedding_model_name,
filepath=filepath
)
if not store:
logging.info("Vector store does not exist. Initializing a new one for knowledge import...")
init_vector_store(
api_key=api_key,
base_url=base_url if base_url else "http://localhost:11434/v1",
interface_format=interface_format,
api_key=embedding_api_key,
base_url=embedding_url if embedding_url else "http://localhost:11434/v1",
interface_format=embedding_interface_format,
embedding_model_name=embedding_model_name,
texts=paragraphs,
filepath=filepath
ui.py
+27 -9
View File
@@ -597,11 +597,13 @@ class NovelGeneratorGUI:
def task():
self.disable_button_safe(self.btn_generate_chapter)
try:
# LLM相关
api_key = self.api_key_var.get().strip()
base_url = self.base_url_var.get().strip()
model_name = self.model_name_var.get().strip()
temperature = self.temperature_var.get()
# 章节信息
chap_num = self.safe_get_int(self.chapter_num_var, 1)
word_number = self.safe_get_int(self.word_number_var, 3000)
user_guidance = self.user_guide_text.get("0.0", "end").strip()
@@ -612,6 +614,10 @@ class NovelGeneratorGUI:
scene_loc = self.scene_location_var.get().strip()
time_constr = self.time_constraint_var.get().strip()
# Embedding相关
embedding_api_key = self.embedding_api_key_var.get().strip()
embedding_url = self.embedding_url_var.get().strip()
embedding_interface_format = self.embedding_interface_format_var.get().strip()
embedding_model_name = self.embedding_model_name_var.get().strip()
embedding_k = self.safe_get_int(self.embedding_retrieval_k_var, 4)
@@ -629,6 +635,10 @@ class NovelGeneratorGUI:
key_items=key_items,
scene_location=scene_loc,
time_constraint=time_constr,
embedding_api_key=embedding_api_key,
embedding_url=embedding_url,
embedding_interface_format=embedding_interface_format,
embedding_model_name=embedding_model_name,
embedding_retrieval_k=embedding_k
)
if draft_text:
@@ -659,13 +669,19 @@ class NovelGeneratorGUI:
def task():
self.disable_button_safe(self.btn_finalize_chapter)
try:
# LLM相关
api_key = self.api_key_var.get().strip()
base_url = self.base_url_var.get().strip()
model_name = self.model_name_var.get().strip()
temperature = self.temperature_var.get()
# Embedding相关
embedding_api_key = self.embedding_api_key_var.get().strip()
embedding_url = self.embedding_url_var.get().strip()
embedding_interface_format = self.embedding_interface_format_var.get().strip()
embedding_model_name = self.embedding_model_name_var.get().strip()
# 章节参数
chap_num = self.safe_get_int(self.chapter_num_var, 1)
word_number = self.safe_get_int(self.word_number_var, 3000)
@@ -686,6 +702,9 @@ class NovelGeneratorGUI:
model_name=model_name,
temperature=temperature,
filepath=filepath,
embedding_api_key=embedding_api_key,
embedding_url=embedding_url,
embedding_interface_format=embedding_interface_format,
embedding_model_name=embedding_model_name
)
self.safe_log(f"✅ 第{chap_num}章定稿完成（已更新全局摘要、角色状态、向量库）。")
@@ -754,19 +773,18 @@ class NovelGeneratorGUI:
def task():
self.disable_button_safe(self.btn_import_knowledge)
try:
api_key = self.embedding_api_key_var.get().strip()
base_url = self.embedding_url_var.get().strip()
interface_format = self.embedding_interface_format_var.get().strip()
embedding_model_name = self.embedding_model_name_var.get().strip()
emb_api_key = self.embedding_api_key_var.get().strip()
emb_url = self.embedding_url_var.get().strip()
emb_format = self.embedding_interface_format_var.get().strip()
emb_model = self.embedding_model_name_var.get().strip()
self.safe_log(f"开始导入知识库文件: {selected_file}")
import_knowledge_file(
api_key=api_key,
base_url=base_url,
interface_format=interface_format,
embedding_model_name=embedding_model_name,
embedding_api_key=emb_api_key,
embedding_url=emb_url,
embedding_interface_format=emb_format,
embedding_model_name=emb_model,
file_path=selected_file,
embedding_base_url=base_url,
filepath=self.filepath_var.get().strip()
)
self.safe_log("✅ 知识库文件导入完成。")