fix(honcho): embedding 升 Qwen3-8B 并 force dimensions=1536 兼容旧 schema

- vm61 数据库 public.documents.embedding dim 是 1536(旧 OpenAI 部署遗留)
- Qwen3-Embedding-8B 默认 4096 维,撞 schema 不匹配导致 honcho-api 启动失败
- 通过 SiliconFlow OpenAI 兼容 API 的 dimensions 参数强制输出 1536
- env.cfg + stack.yml + readme 全部同步更新
- docker compose config 校验通过,vm61 honcho-api /health HTTP 200
This commit is contained in:
2026-06-19 18:26:45 +08:00
parent b4561f1d9e
commit 80d485605b
3 changed files with 15 additions and 12 deletions
+6 -4
View File
@@ -42,9 +42,10 @@ services:
- NAMESPACE=${HONCHO_NAMESPACE:-honcho}
- EMBED_MESSAGES=${HONCHO_EMBED_MESSAGES:-true}
- EMBEDDING_MODEL_CONFIG__TRANSPORT=openai
- EMBEDDING_MODEL_CONFIG__MODEL=${HONCHO_EMBEDDING_MODEL:-Qwen/Qwen3-Embedding-0.6B}
- EMBEDDING_MODEL_CONFIG__MODEL=${HONCHO_EMBEDDING_MODEL:-Qwen/Qwen3-Embedding-8B}
- EMBEDDING_MODEL_CONFIG__OVERRIDES__BASE_URL=${HONCHO_EMBEDDING_BASE_URL:-https://api.siliconflow.cn/v1}
- EMBEDDING_VECTOR_DIMENSIONS=1024
- EMBEDDING_MODEL_CONFIG__OVERRIDES__DIMENSIONS=1536
- EMBEDDING_VECTOR_DIMENSIONS=1536
- EMBEDDING_API_KEY=${HONCHO_EMBEDDING_API_KEY:-}
- DERIVER_MODEL_CONFIG__TRANSPORT=openai
- DERIVER_MODEL_CONFIG__MODEL=${HONCHO_DERIVER_MODEL:-gpt-4o-mini}
@@ -95,9 +96,10 @@ services:
- DERIVER_MODEL_CONFIG__MODEL=${HONCHO_DERIVER_MODEL:-gpt-5.4-mini}
- DERIVER_WORKERS=${HONCHO_DERIVER_WORKERS:-1}
- EMBEDDING_MODEL_CONFIG__TRANSPORT=openai
- EMBEDDING_MODEL_CONFIG__MODEL=${HONCHO_EMBEDDING_MODEL:-Qwen/Qwen3-Embedding-0.6B}
- EMBEDDING_MODEL_CONFIG__MODEL=${HONCHO_EMBEDDING_MODEL:-Qwen/Qwen3-Embedding-8B}
- EMBEDDING_MODEL_CONFIG__OVERRIDES__BASE_URL=${HONCHO_EMBEDDING_BASE_URL:-https://api.siliconflow.cn/v1}
- EMBEDDING_VECTOR_DIMENSIONS=1024
- EMBEDDING_MODEL_CONFIG__OVERRIDES__DIMENSIONS=1536
- EMBEDDING_VECTOR_DIMENSIONS=1536
- LLM_OPENAI_API_KEY=${HONCHO_LLM_API_KEY:?set HONCHO_LLM_API_KEY}
- EMBEDDING_API_KEY=${HONCHO_EMBEDDING_API_KEY:-}
- MODEL_CONFIG__OVERRIDES__BASE_URL=${HONCHO_LLM_BASE_URL:-https://api.openai.com/v1}