sacrefizz
Forum Replies Created
-
Hi Tim,
I replicated your fix by hooking into http_request_args and re-encoding the batch JSON with JSON_UNESCAPED_SLASHES before it reaches the OpenAI API. The
<\/p> artifacts are gone.The \n issue also seems resolved in my tests — likely because the LLM no longer tries to “match” the escaped style it receives in the input.
Looking forward to 2.8.2. Thanks for investigating and fixing this!
Thanks Tim. The <\/p> is actually valid JSON escaping — OpenAI’s response is correct JSON. The issue is that Loco embeds the raw API JSON string
into its own AJAX response without decoding it first, causing double-encoding. A json_decode() before re-serializing would fix this. I’ve written a
client-side workaround (jQuery dataFilter) that cleans the response before Loco parses it, which confirms the issue is in the decode/re-encode
pipeline, not the model output.What is the issue ?
Forum: Plugins
In reply to: [ProfileGrid – User Profiles, Groups and Communities] Icon-pictures @ Groupssame