Error on generation: ValueError: Value too wide (>4 bytes) & other issues #291
Labels
No Label
bug
duplicate
enhancement
help wanted
insufficient info
invalid
news
not a bug
question
wontfix
No Milestone
No project
No Assignees
2 Participants
Notifications
Due Date
No due date set.
Dependencies
No dependencies set.
Reference: mrq/ai-voice-cloning#291
Loading…
Reference in New Issue
Block a user
No description provided.
Delete Branch "%!s()"
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?
I trained a model, and I'm having issues with all the saved pth files form that training. Notice that I trained and used models before with no issues.
I have 3 pth files under training/MyVoice/finetune/models:
25_gpt.pth
50_gpt.pth
75_gpt.pth
The first two (25 and 50), when I run it give the error:
The last model, when I use it gives a different error:
I'm using Windows 11 Pro, an RTX 3070, CUDA V12.1.105
Let me know any other information you need
In regards to this last error, I added a breakpoint to the utils.py file, just to see the size of audio_cache[name]['settings']. I figured that, if it's saying that the value is too wide, I would at least see the size of what is being dumped to metadata['lyrics']. I added:
import sys
sys.getsizeof(audio_cache[name]['settings'])
and the value was 1176
Thanks for the help, I hope to have given you all the info you need.
I'm not too sure how there would be a discrepancy between the models, but the first error seems to be that you don't actually have any voice samples to feed into TorToiSe (or it's unable to actually load them, for whatever reason).
The second error I'm not really too sure, but you can bandaid the issue, if I remember right, by going into Settings and unchecking
Embed Output Metadata
(I should probably default disable this, since no one really uses this feature to my knowledge).Many thanks! All my problems seem to be fixed, I'll let you know if I find any others. Thank you for your work!