Math Domain Error in Pretraining tutorial.
thedragoncoder opened this issue · 1 comments
Describe the bug
I have been getting math domain error. When I tried to fine tune the existing model using the sb.Brain class, if my modules has any classifiers or models with parameters, I have noticed that this error comes up.
I have been referring the pretraining tutorial provided by speechbrain to fine tune my model. When I tried running the same tutorial, the exact error pops up.
The only change I have created is
as the name of the hparams file was wrong.
Expected behaviour
I think, the error pops up because of exceptions that are not handled properly.
To Reproduce
This is the pretraining notebook
https://colab.research.google.com/drive/1PK3S1_hgKH0H7Eio1HUxa6XLykUfAlnV?usp=sharing#scrollTo=bf1rDbYBeh-7
To make it easier to locate the error, I have created a notebook which only consists of cells that I have executed.
https://colab.research.google.com/drive/1h5kd64eV4b0qnL6hHR35pW71INT5s0Bq?usp=sharing
Environment Details
No response
Relevant Log Output
---------------------------------------------------------------------------
ValueError Traceback (most recent call last)
<ipython-input-84-fe2297a0da3e> in <cell line: 1>()
----> 1 brain = EncDecFineTune(modules, hparams=hparams)
2 # brain.tokenizer = language_id.tokenizer
2 frames
/usr/local/lib/python3.10/dist-packages/speechbrain/utils/logger.py in format_order_of_magnitude(number, abbreviate)
137 style = ORDERS_ABBREV if abbreviate else ORDERS_WORDS
138 precision = "{num:3.1f}"
--> 139 order = 3 * math.floor(math.log(math.fabs(number), 1000))
140 # Fallback for very large numbers:
141 while order not in style and order != 0:
ValueError: math domain error
Additional Context
No response
Hello, I fixed this issue with PR #2537. Thanks a lot for reporting this bug!