RuntimeError: probability tensor contains either inf, nan or element < 0 when using ST or switching characters

[quarterturn]

Not sure this warrants a bug report, but when:
MiquMaid-v1-70B-4bpw-exl2
LOADER: "ExLlamav2_HF"
TRUNCATION LENGTH: 32764
INSTRUCTION TEMPLATE: "Alpaca"

Changing characters, or trying to use SillyTavern at all, results in:

  File "/home/derp/Documents/text-generation-webui/installer_files/env/lib/python3.11/site-packages/transformers/generation/utils.py", line 2658, in sample
    next_tokens = torch.multinomial(probs, num_samples=1).squeeze(1)
                  ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
RuntimeError: probability tensor contains either `inf`, `nan` or element < 0
Output generated in 3.39 seconds (0.00 tokens/s, 0 tokens, context 2259, seed 1398472213)

I'm on:

commit 0f134bf744acef2715edd7d39e76f865d8d83a19 (HEAD -> main, origin/main, origin/HEAD)
Merge: a329db06 775902c1
Author: oobabooga <[email protected]>
Date:   Tue Feb 6 11:50:21 2024 -0300

Presumably, something is feeding in a token the model doesn't like, but I'm not sure what.