fix: avoid OpenAILikeEmbedding always use zhipu api url when knowledge graph building

[Mini256]

When using OpenAILikeEmbedding, if a provider other than zhipu is configured, an authentication information error will appear when building knowledge graph.

[2025-01-08 03:10:38,501: INFO/ForkPoolWorker-2] HTTP Request: POST https://open.bigmodel.cn/api/paas/v4/embeddings "HTTP/1.1 401 Unauthorized"
[2025-01-08 03:10:38,507: ERROR/ForkPoolWorker-2] app.tasks.build_index.build_kg_index_for_chunk[4489bfc3-0e7c-4606-ad77-b973ef24a3f1]: Failed to build knowledge graph index for chunk #60564d4d-1522-476c-9660-ab1859ee2b1e
Traceback (most recent call last):
  File "/app/app/tasks/build_index.py", line 125, in build_kg_index_for_chunk
    index_service.build_kg_index_for_chunk(index_session, db_chunk)

This bug is due to config.pop()call will remove the api_base param from the config, and using the zhipu api_base by default.

In the method of get_embed_model / get_llm / get_reranker_model, the config should be inmutable.

[vercel]

The latest updates on your projects. Learn more about Vercel for Git ↗︎

Name	Status	Preview	Comments	Updated (UTC)
tidb-ai-preview	✅ Ready (Inspect)	Visit Preview	💬 Add feedback	Jan 8, 2025 5:29am
tidb-ai-storybook	✅ Ready (Inspect)	Visit Preview	💬 Add feedback	Jan 8, 2025 5:29am

[github-actions]

E2E Result Deployment

https://tidb-ai-playwright-c78j74tag-djaggers-projects.vercel.app