Problem about interpret using glm4.7 from chatglm

[mugpeng]

Hi,

I really like the new feature that integrates LLM into the enrichment analysis.

However, I found that when running the following:

> interpret(enrichment$up$KEGG, task = "interpretation", model = "glm-4.7")
Interpreting cluster: Default
## Interpretation Result

### Cluster: Default

### 1. Overview

Warning message:
In value[[3L]](cond) :
  Failed to parse JSON response from LLM. Returning raw text. Error: parse error: premature EOF
                                       
                     (right here) ------^

Here, enrichment$up$KEGG is an enrichResult object, and the previous version (glm-4) works fine:

So I guess the issue might be related to changes in the output format of the new glm version.

In addition, I noticed that we have to deploy the “fanyi” API to run the function:

> interpret(result$up$KEGG, task = "interpretation")
Interpreting cluster: Default
Error in value[[3L]](cond) : 
  Failed to call fanyi::chat_request. Error: API key for deepseek is missing.

Even though I don’t need translation, it’s a bit tedious to check the API documentation and manually call set_translate_option.
It might be better to add an option to disable fanyi when not needed.

Thanks again for the great work!

Best,
Peng

[GuangchuangYu]

Please use DeepSeek, as it is currently better tested for this functionality.

Additionally, we plan to transition the backend to rely on the aisdk package, https://github.com/YuLab-SMU/aisdk, in a future update.