There is no RAG Service for refs/klayout_docs.txt

[charliecaffe]

hi @jusjinuk ,
Thank you for releasing the codes! I have set up the environment and run the scripts successfully with config --ctx-mode none. However, the generated DRC and GDS files are not correct. When I change the --ctx-mode to ic, the total token is over the maximum token length of the model, so I change the --ctx-mode to rag. I find there is no RAG service in the project. Could you release the implementation of the rag service?

[jusjinuk]

Hello, @charliecaffe. Thank you for your interest in our work.

The reference file refs/klayout_docs.txt is around 62K tokens, which should fit within the context window of the models used in our paper: 128K for GPT-OSS models and 262K for Qwen3. Therefore, ic mode should not exceed the context length by itself. Are you using an older model with a shorter context limit? It may also be worth checking whether the max context length is being set lower than what the model actually supports.

Regarding the RAG code, it is legacy code that was tested with an earlier version of the project but has not been validated with the latest version. I uploaded a small zip file containing the related code and README instructions for running the RAG server here:

• Download

Please note that it may not be fully compatible with the current version of the documentation. In particular, the chunking logic may not match because the original RAG code was written for HTML-formatted docs, while the current reference document is in Markdown. Still, I am sharing it for reference.