forked from Ikalus1988/MisakaNet
-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathlesson-chroma-建库无-checkpoint-进程一死全部丢失.json
More file actions
12 lines (12 loc) · 1.25 KB
/
Copy pathlesson-chroma-建库无-checkpoint-进程一死全部丢失.json
File metadata and controls
12 lines (12 loc) · 1.25 KB
1
2
3
4
5
6
7
8
9
10
11
12
{
"task_id": "lesson-chroma-建库无-checkpoint-进程一死全部丢失",
"title": "Chroma 建库无 Checkpoint 进程一死全部丢失",
"domain": "general",
"tags": [
"contrib"
],
"problem": "Agent 内跑 BGE-large-zh 建库(约30分钟),Gateway 重启后整个数据库为空,0条向量入库。",
"solution": "改造 `build_edoc_chroma.py` 为小批次写入:\n\n```python\n# 改造后:每 5000 条立即写入\ndef embed_and_build(chunks, model, collection, batch_size=5000):\n for i in range(0, len(chunks), batch_size):\n batch = chunks[i:i+batch_size]\n embeddings = model.encode([c[\"text\"] for c in batch])\n collection.add(\n ids=[c[\"id\"] for c in batch],\n embeddings=embeddings.tolist(),\n documents=[c[\"text\"] for c in batch]\n )\n print(f\"[Checkpoint] Written {i+batch_size}/{len(chunks)}\")\n```\n\n**启动方式改为后台独立运行**(不在 agent 里跑):\n\n```bash\ncd ~//hf-mirror.com \\\n~/.hermes/hermes-agent/.venv/bin/python3 ~/.hermes/scripts/build_edoc_chroma.py \\\n2>&1 | tee /mnt/d/Eric/知识库/chroma_db_v3/build.log &\n```",
"source": "lessons/contrib/chroma-建库无-checkpoint-进程一死全部丢失.md",
"test_cmd": ""
}