txai-dev/yu-zhi-ran
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Initial commit: yu-zhi-ran platform with automation integration

Browse Source
This commit is contained in:
lt
2026-04-19 14:05:09 +08:00
commit 3cb2df51c8
209 changed files with 80379 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
platform/backend/app/core/__init__.py
+1
View File
@@ -0,0 +1 @@
# core package
platform/backend/app/core/generator.py
+53
View File
@@ -0,0 +1,53 @@
import subprocess
from pathlib import Path
import logging
import os
logger = logging.getLogger(__name__)
# 计算项目根目录（从本文件位置上升4层）
PROJECT_ROOT = Path(__file__).resolve().parents[4]
# 允许环境变量覆盖（适合容器部署）
if os.getenv('PROJECT_ROOT'):
PROJECT_ROOT = Path(os.getenv('PROJECT_ROOT'))
def run_creator(topic_id: str = None):
"""运行内容创作脚本，返回简略结果
Args:
topic_id: 可选，指定要创作的选题ID。不指定则创作优先级最高的选题。
"""
script_path = PROJECT_ROOT / "scripts" / "creator.py"
cmd = ["python3", str(script_path)]
if topic_id:
cmd.extend(["--topic-id", topic_id])
result = subprocess.run(
cmd,
cwd=str(PROJECT_ROOT),
capture_output=True,
text=True,
timeout=300 # 5分钟超时
)
if result.returncode != 0:
logger.error(f"Creator failed: {result.stderr}")
return {"ok": False, "error": result.stderr}
# 解析日志，找出选择了哪个选题
topic_id = None
for line in result.stdout.splitlines():
if "选择了选题:" in line:
# 格式: 2026-04-16 ... INFO - 选择了选题: 标题 (优先级: X)
# 标题可能在行内，但ID不一定有。我们稍后用文件同步。
logger.info(line.strip())
if "选题" in line and "已标记为「待发布」" in line:
# 如: 2026-04-16 ... INFO - 选题 A01 已标记为「待发布」
import re
m = re.search(r'选题\s+([A-Za-z0-9]+)', line)
if m:
topic_id = m.group(1)
return {
"ok": True,
"topic_id": topic_id,
"stdout": result.stdout[-1000:] if len(result.stdout) > 1000 else result.stdout
}
platform/backend/app/core/nvidia_client.py
+118
View File
@@ -0,0 +1,118 @@
"""
NVIDIA 专用 LLM 客户端（fixed configuration）
使用 OpenAI 兼容接口调用 stepfun-ai/step-3.5-flash
"""
import requests
import json
from typing import Optional
class LLMError(Exception):
pass
# 固定配置（你的可用 key）
CONFIG = {
"base_url": "https://integrate.api.nvidia.com/v1",
"api_key": "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",
"model": "stepfun-ai/step-3.5-flash",
}
def call_llm(
prompt: str,
system_prompt: str = "你是一个专业的内容创作助手。",
temperature: float = 0.7,
max_tokens: int = 2000,
stream: bool = False,
) -> str:
"""
调用 NVIDIA LLM 生成文本
"""
endpoint = f"{CONFIG['base_url'].rstrip('/')}/chat/completions"
headers = {
"Authorization": f"Bearer {CONFIG['api_key']}",
"Content-Type": "application/json"
}
payload = {
"model": CONFIG["model"],
"messages": [
{"role": "system", "content": system_prompt},
{"role": "user", "content": prompt}
],
"temperature": temperature,
"max_tokens": max_tokens,
"stream": stream,
}
try:
resp = requests.post(endpoint, json=payload, headers=headers, timeout=120, stream=stream)
if resp.status_code != 200:
raise LLMError(f"HTTP {resp.status_code}: {resp.text[:200]}")
if stream:
full = []
for line in resp.iter_lines():
if not line:
continue
if line.startswith(b'data: '):
data = line[6:]
if data == b'[DONE]':
break
try:
chunk = json.loads(data)
delta = chunk['choices'][0]['delta']
# 支持 reasoning_content 或 reasoning 字段
if 'reasoning_content' in delta and delta['reasoning_content']:
full.append(delta['reasoning_content'])
if 'content' in delta and delta['content']:
full.append(delta['content'])
except Exception:
continue
return "".join(full)
else:
data = resp.json()
msg = data["choices"][0]["message"]
content = msg.get('content') or msg.get('reasoning') or msg.get('reasoning_content')
return content.strip() if content else ''
except requests.RequestException as e:
raise LLMError(f"Request failed: {e}")
def expand_content_with_llm(topic: dict, section_title: str, section_content: str, context: str = "") -> str:
"""扩写大纲章节，返回包含 ## 标题的完整 Markdown"""
prompt = f"""你是一个专业的内容创作者。请将以下大纲扩展为完整的文章章节。
# 选题信息
- 标题：{topic.get('title')}
- 领域：{topic.get('field')}
- 核心观点：{topic.get('core_concept', '')}
- 受众痛点：{topic.get('audience_pain', '')}
- 独特视角：{topic.get('unique_angle', '')}
# 当前章节
## {section_title}
{section_content}
# 要求
- 以 `## {section_title}` 作为章节标题开头
- 字数：300-500字
- 风格：客观、专业、易懂
- 使用 Markdown 格式
- 包含具体数据或案例（如果有）
- 保持与整体文章调性一致
直接输出完整的 Markdown 章节（包括 ## 标题和正文段落）。"""
if context:
prompt = f"# 参考资料\n{context}\n\n{prompt}"
try:
result = call_llm(prompt, temperature=0.8, max_tokens=2000)
return result.strip()
except Exception as e:
return f"## {section_title}\n\n（LLM 调用失败：{e}，请手动补充）"
# 测试
if __name__ == "__main__":
try:
print(f"[nvidia_client] 使用模型: {CONFIG['model']}")
resp = call_llm("你好，请用一句话介绍你自己。", max_tokens=50)
print(f"[nvidia_client] 响应: {resp}")
except Exception as e:
print(f"[nvidia_client] 错误: {e}")
platform/backend/app/core/optimizer.py
+47
View File
@@ -0,0 +1,47 @@
import subprocess
from pathlib import Path
import logging
import os
import json
from datetime import datetime
from typing import List
logger = logging.getLogger(__name__)
# 计算项目根目录（从本文件位置上升4层）
PROJECT_ROOT = Path(__file__).resolve().parents[4]
if os.getenv('PROJECT_ROOT'):
PROJECT_ROOT = Path(os.getenv('PROJECT_ROOT'))
def run_optimizer(topic_ids: List[str] = None):
"""运行合规优化脚本，返回报告摘要
Args:
topic_ids: 可选，指定要优化的选题ID列表。不指定则优化所有 draft 文章。
"""
script_path = PROJECT_ROOT / "scripts" / "compliance_optimizer.py"
cmd = ["python3", str(script_path)]
if topic_ids:
cmd.extend(["--topic-ids", ','.join(topic_ids)])
logger.info(f"[DEBUG] Running optimizer with topic_ids={topic_ids}, cmd={' '.join(cmd)}")
result = subprocess.run(
cmd,
cwd=str(PROJECT_ROOT),
capture_output=True,
text=True,
timeout=600 # 10分钟
)
if result.returncode != 0:
logger.error(f"Optimizer failed: {result.stderr}")
return {"ok": False, "error": result.stderr}
# 读取优化报告（优化脚本会在 today 的 drafts 目录生成报告）
report_date = datetime.now().strftime("%Y-%m-%d")
report_path = PROJECT_ROOT / "automation" / "data" / "drafts" / report_date / "optimization_report.json"
if report_path.exists():
report = json.loads(report_path.read_text(encoding='utf-8'))
return {"ok": True, "report": report}
else:
logger.warning(f"Report not found: {report_path}")
return {"ok": True, "report": None, "stdout": result.stdout}
platform/backend/app/core/qnaigc_client.py
+113
View File
@@ -0,0 +1,113 @@
"""
qnaigc 专用 LLM 客户端
模型：arcee-ai/trinity-large-preview
"""
import requests
import json
from typing import Optional
class LLMError(Exception):
pass
CONFIG = {
"base_url": "https://api.qnaigc.com/v1",
"api_key": "sk-2cb9561a18351015d3120ffac4abae0480fa17e0d28469bdce5fc905d1a42e0d",
"model": "arcee-ai/trinity-large-preview",
}
def call_llm(
prompt: str,
system_prompt: str = "你是一个专业的内容创作助手。",
temperature: float = 0.7,
max_tokens: int = 2000,
stream: bool = False,
) -> str:
endpoint = f"{CONFIG['base_url'].rstrip('/')}/chat/completions"
headers = {
"Authorization": f"Bearer {CONFIG['api_key']}",
"Content-Type": "application/json"
}
payload = {
"model": CONFIG["model"],
"messages": [
{"role": "system", "content": system_prompt},
{"role": "user", "content": prompt}
],
"temperature": temperature,
"max_tokens": max_tokens,
"stream": stream,
}
try:
resp = requests.post(endpoint, json=payload, headers=headers, timeout=120, stream=stream)
if resp.status_code != 200:
raise LLMError(f"HTTP {resp.status_code}: {resp.text[:200]}")
if stream:
full = []
for line in resp.iter_lines():
if not line:
continue
if line.startswith(b'data: '):
data = line[6:]
if data == b'[DONE]':
break
try:
chunk = json.loads(data)
delta = chunk['choices'][0]['delta']
if 'reasoning_content' in delta and delta['reasoning_content']:
full.append(delta['reasoning_content'])
if 'content' in delta and delta['content']:
full.append(delta['content'])
except Exception:
continue
return "".join(full)
else:
data = resp.json()
msg = data["choices"][0]["message"]
content = msg.get('content') or msg.get('reasoning') or msg.get('reasoning_content')
return content.strip() if content else ''
except requests.RequestException as e:
raise LLMError(f"Request failed: {e}")
def expand_content_with_llm(topic: dict, section_title: str, section_content: str, context: str = "") -> str:
"""扩写大纲章节，返回包含 ## 标题的完整 Markdown"""
prompt = f"""你是一个专业的内容创作者。请将以下大纲扩展为完整的文章章节。
# 选题信息
- 标题：{topic.get('title')}
- 领域：{topic.get('field')}
- 核心观点：{topic.get('core_concept', '')}
- 受众痛点：{topic.get('audience_pain', '')}
- 独特视角：{topic.get('unique_angle', '')}
# 当前章节
## {section_title}
{section_content}
# 要求
- 以 `## {section_title}` 作为章节标题开头
- 字数：300-500字
- 风格：客观、专业、易懂
- 使用 Markdown 格式
- 包含具体数据或案例（如果有）
- 保持与整体文章调性一致
- 所有数据和时间必须基于2025年及以后，避免引用2024年以前的具体事件或统计数据。如果信息不足，请使用'近期'、'最新'等模糊表述，不要编造旧数据。
直接输出完整的 Markdown 章节（包括 ## 标题和正文段落）。"""
if context:
prompt = f"# 参考资料\n{context}\n\n{prompt}"
try:
result = call_llm(prompt, temperature=0.8, max_tokens=2000)
return result.strip()
except Exception as e:
return f"## {section_title}\n\n（LLM 调用失败：{e}，请手动补充）"
# 测试
if __name__ == "__main__":
try:
print(f"[qnaigc_client] 使用模型: {CONFIG['model']}")
resp = call_llm("你好，请用一句话介绍你自己。", max_tokens=50)
print(f"[qnaigc_client] 响应: {resp}")
except Exception as e:
print(f"[qnaigc_client] 错误: {e}")
platform/backend/app/core/sync.py
+65
View File
@@ -0,0 +1,65 @@
import json
from datetime import datetime, date
from pathlib import Path
from sqlalchemy.orm import Session
from ..database import SessionLocal
from ..models import Topic
import os
# 计算项目根目录（从本文件位置上升4层）
PROJECT_ROOT = Path(__file__).resolve().parents[4]
if os.getenv('PROJECT_ROOT'):
PROJECT_ROOT = Path(os.getenv('PROJECT_ROOT'))
TOPICS_FILE = PROJECT_ROOT / "automation" / "data" / "sustainability_topics.json"
def sync_topic_to_db(topic_id: str, db: Session = None) -> Topic:
topics = json.loads(TOPICS_FILE.read_text(encoding='utf-8'))
topic_data = next((t for t in topics if t['id'] == topic_id), None)
if not topic_data:
raise ValueError(f"Topic {topic_id} not found in file")
close_db = False
if db is None:
db = SessionLocal()
close_db = True
try:
db_topic = db.query(Topic).filter(Topic.id == topic_id).first()
if db_topic is None:
db_topic = Topic(
id=topic_data['id'],
title=topic_data['title'],
field=topic_data['field'],
format=topic_data.get('format'),
core_concept=topic_data.get('core_concept'),
audience_pain=topic_data.get('audience_pain'),
unique_angle=topic_data.get('unique_angle'),
priority=topic_data.get('priority'),
priority_score=topic_data.get('priority_score', 0),
total_score=topic_data.get('total_score')
)
db.add(db_topic)
db_topic.status = topic_data.get('status', db_topic.status)
db_topic.ready_at = datetime.strptime(topic_data['ready_at'], '%Y-%m-%d').date() if topic_data.get('ready_at') else None
db_topic.published_at = datetime.strptime(topic_data['published_at'], '%Y-%m-%d').date() if topic_data.get('published_at') else None
db_topic.compliance_score = topic_data.get('compliance_score', db_topic.compliance_score)
db_topic.platform_urls = topic_data.get('platform_urls', {})
db_topic.updated_at = datetime.now()
db.commit()
db.refresh(db_topic)
return db_topic
finally:
if close_db:
db.close()
def sync_all_topics():
db = SessionLocal()
try:
topics = json.loads(TOPICS_FILE.read_text(encoding='utf-8'))
for t in topics:
sync_topic_to_db(t['id'], db)
print(f"✅ 同步 {len(topics)} 个选题到数据库")
finally:
db.close()
if __name__ == "__main__":
sync_all_topics()