Summary
Memory chunks are repeatedly skipped after changing from a 384-dimension embedding model to a 768-dimension model. Rebuilding memories does not resolve the mismatch, leaving Memory Recall effectively disabled.
Expected behavior
Rebuilding memories should re-vectorize existing memory chunks with the current embedding model so Memory Recall can continue working.
Actual behavior
The console repeatedly reports that memory chunk insertion is skipped because the embedding dimension changed from 384 to 768. The error persists after using Rebuild Memories.
Steps to reproduce
- Create or retain memories using a 384-dimension embedding model.
- Change to a 768-dimension embedding model.
- Run Rebuild Memories.
- Use Memory Recall and observe the recurring embedding-dimension warning in the console.
Environment
- Marinara Engine version: v2.4.2, build fc58ee4, Staging branch
- Install type: Not provided
- OS + version: Windows 11
- Browser or app shell: Firefox
Logs, screenshots, or video
Console log: "Skipping memory chunk insert for chat y0hbHxrR2epzzyoTqo6ID because embedding dimension changed from 384 to 768. Rebuild memories before mixing embedding models." Attached screenshot: https://cdn.discordapp.com/attachments/1536183121387593869/1536183121706221578/image.png?ex=6a7a7985&is=6a792805&hm=1581913d6815887ec36fc4a26a0d477a467377e6e8e8c0d14c85f3be26ca1fb6&
Additional context
The embedding model was changed some time ago. The user requests a "Re-Vectorize All" function for Memory Recall similar to the existing Lorebooks function, or guidance for clearing the stale embedding data.
Discord thread: FR - Re-Vectorize Memory Chunks - Like in Lorebooks
Discord thread link: https://discord.com/channels/1417099416812392641/1536183121387593869
Requested by: Marinara (marinara_spaghetti, Discord ID 169561011928432653)
Discord thread transcript (1 message)
[2026-08-10T01:23:17.041Z] ylliselmanii: Release: v2.4.2
Build: fc58ee4
Branch: Staging
Installation: Windows 11 accessed through Firefox browser
I am being plagued by a single, ever reoccuring message in my ME console log. I shiver just thinking about it... :pain:
Behold, the terror that escaped from my nightmares into the waking world!
> Skipping memory chunk insert for chat y0hbHxrR2epzzyoTqo6ID because embedding dimension changed from 384 to 768. Rebuild memories before mixing embedding models."
Uh, anyway, I can't seem to get rid of it after changing embedding models ages ago, effectively disabling the Memory Recall function for me completely. Rebuilding Memories did not do the trick either... don't even really know what it did.
Anyway, I wanted to ask if the function to revectorize entries in Lorebooks could be copy-pasted into the Memory Recall window, because I don't know what else could be the problem. You know, "Re-Vectorize All".
I'd really, really appreciate it.
Or if not, maybe someone could just explain how to fix that?
Thank you.
[attached image 1: https://cdn.discordapp.com/attachments/1536183121387593869/1536183121706221578/image.png?ex=6a7a7985&is=6a792805&hm=1581913d6815887ec36fc4a26a0d477a467377e6e8e8c0d14c85f3be26ca1fb6&]
[Discord message: https://discord.com/channels/1417099416812392641/1536183121387593869/1536183121387593869]
Template check
Please uncheck (untick) the box below before submitting so we know you read the template. It is intentionally pre-checked:
Summary
Memory chunks are repeatedly skipped after changing from a 384-dimension embedding model to a 768-dimension model. Rebuilding memories does not resolve the mismatch, leaving Memory Recall effectively disabled.
Expected behavior
Rebuilding memories should re-vectorize existing memory chunks with the current embedding model so Memory Recall can continue working.
Actual behavior
The console repeatedly reports that memory chunk insertion is skipped because the embedding dimension changed from 384 to 768. The error persists after using Rebuild Memories.
Steps to reproduce
Environment
Logs, screenshots, or video
Console log: "Skipping memory chunk insert for chat y0hbHxrR2epzzyoTqo6ID because embedding dimension changed from 384 to 768. Rebuild memories before mixing embedding models." Attached screenshot: https://cdn.discordapp.com/attachments/1536183121387593869/1536183121706221578/image.png?ex=6a7a7985&is=6a792805&hm=1581913d6815887ec36fc4a26a0d477a467377e6e8e8c0d14c85f3be26ca1fb6&
Additional context
The embedding model was changed some time ago. The user requests a "Re-Vectorize All" function for Memory Recall similar to the existing Lorebooks function, or guidance for clearing the stale embedding data.
Discord thread: FR - Re-Vectorize Memory Chunks - Like in Lorebooks
Discord thread link: https://discord.com/channels/1417099416812392641/1536183121387593869
Requested by: Marinara (marinara_spaghetti, Discord ID 169561011928432653)
Discord thread transcript (1 message)
Template check
Please uncheck (untick) the box below before submitting so we know you read the template. It is intentionally pre-checked: