colab is not working anymore

[eunkoh]

Hi, could you double check mt3, t5x versions in colab notebook? Your colab is not working anymore, so if you could double-check this notebook, https://github.com/magenta/mt3/blob/main/mt3/colab/music_transcription_with_transformers.ipynb, that would be appreciated.

I cannot import the models because mt3, t5x cannot be imported in your cell. Thanks.

[bwgJohn]

I have the same problem

[nic-chua]

I am having the same issue, I cannot import t5 in the Import and Definitions cell, but it was working a couple days ago

[iansimon]

Ugh, I'm not sure the best way to fix this. For now, you can add another pip install command that looks like:

!python3 -m pip install clu==0.0.8 numpy==1.22.4 --ignore-installed

I'll see if I can find a cleaner fix before I update the official notebook...

[Jackl-o-o-l]

Unfortunately, the fix does not work for me. I still get the same issue after adding the command.

[SDSC-christian]

Why not fix the version of the only library that is not fixed: !python3 -m pip install --no-dependencies --upgrade ddsp==3.5.0

[Jackl-o-o-l]

This does not work for me either

[nic-chua]

I've tried multiple fixes but haven't had any luck. Has anyone been able to find a fix for this issue?

[zhanh-he]

After installed all packages of Setup Environment, !pip install numpy -I Restart runtime by this cell, then define "load_gtag" and "log_event" again. https://drive.google.com/file/d/1JZCikujYFBKAa2UfCPXEmSglLsc2JhwZ/view?usp=sharing

[Jackl-o-o-l]

After installed all packages of Setup Environment, !pip install numpy -I Restart runtime by this cell, then define "load_gtag" and "log_event" again. https://drive.google.com/file/d/1JZCikujYFBKAa2UfCPXEmSglLsc2JhwZ/view?usp=sharing

I tried this fix and an error occurred when running "Upload Audio": "AudioIOReadError: initialization of _internal failed without raising an exception". This is probably because numpy (1.25.1) installed in the fix is not compatible with numba, which requires numpy<1.24.

[Jackl-o-o-l]

I fix the AudioIOReadError by changing the line !pip install numpy -I to !pip install numpy==1.23.5 -I

However, now another error occurred when running "Transcribe Audio": "ImportError: fluidsynth() was called but pyfluidsynth is not installed".

[Jackl-o-o-l]

Replacing !apt-get update -qq && apt-get install -qq libfluidsynth2 build-essential libasound2-dev libjack-dev with !apt-get update -qq && apt-get install -qq libfluidsynth3 build-essential libasound2-dev libjack-dev would probably solve the issue with pyfluidsynth

[Jackl-o-o-l]

It worked, finally. I also downgraded bokeh so that midi image could come out normally. https://colab.research.google.com/drive/1JkWyAwFhAC6SJJ1Lvf4VPap6FwtniEuy?usp=sharing

[jsphweid]

@Jackl-o-o-l That notebook doesn't work for me: I run into the same issue as https://github.com/magenta/mt3/issues/136#issue-1823434021

Did you make any other alterations?

[Jackl-o-o-l]

@jsphweid I think if you click "RESTART RUNTIME" after running !pip install numpy==1.23.5 -I, it will work for you. I did not change anything before you ask the question but now I added a line of notice in the notebook.

[sxueck]

It worked, finally. I also downgraded bokeh so that midi image could come out normally. https://colab.research.google.com/drive/1JkWyAwFhAC6SJJ1Lvf4VPap6FwtniEuy?usp=sharing

Thank you very much. You've done me a great favor. ; )

[MohammedMehdiTBER]

It worked, finally. I also downgraded bokeh so that midi image could come out normally. https://colab.research.google.com/drive/1JkWyAwFhAC6SJJ1Lvf4VPap6FwtniEuy?usp=sharing

Worked for me but still It produces absurd midi files as I cannot recognize the original sound of the song also It splits one guitar into several absurd identifiers! Guitar songs are lost tot this AI.

[Jackl-o-o-l]

It worked, finally. I also downgraded bokeh so that midi image could come out normally. https://colab.research.google.com/drive/1JkWyAwFhAC6SJJ1Lvf4VPap6FwtniEuy?usp=sharing

Worked for me but still It produces absurd midi files as I cannot recognize the original sound of the song also It splits one guitar into several absurd identifiers! Guitar songs are lost tot this AI.

Unfortunately, I had a similar experience. For the notes given by the same instrument in the input audio, the model somehow decides to use some different and weird instruments that suddenly pop out and last only for a few bars. When I listen to the output midi, sometimes I feel that I am listening to the soundtrack of a thriller, despite that the input is just a normal pop song without the vocal. I believed that those are problems related with the model itself. I hope that they could improve the model so that the output midi file could be more consistent.

[MohammedMehdiTBER]

The model is not fine-tuned in terms of guitars. Sadly, there is no alternative. It is hard to recognize whether those parts are linked to the guitar or not. It could leave you trying for a day.

A technical, translator, and moderate Sunni Muslim. The antidote of hypocrites, disbelievers and grooming perisher.

On Fri, 18 Aug 2023 at 07:35, Jackl-o-o-l @.***> wrote:

It worked, finally. I also downgraded bokeh so that midi image could come out normally. https://colab.research.google.com/drive/1JkWyAwFhAC6SJJ1Lvf4VPap6FwtniEuy?usp=sharing

Worked for me but still It produces absurd midi files as I cannot recognize the original sound of the song also It splits one guitar into several absurd identifiers! Guitar songs are lost tot this AI.

Unfortunately, I had a similar experience. For the notes given by the same instrument in the input audio, the model somehow decides to use some different and weird instruments that suddenly pop out and last only for a few bars. When I listen to the output midi, sometimes I feel that I am listening to the soundtrack of a thriller, despite that the input is just a normal pop song without the vocal. I believed that those are problems related with the model itself. I hope that they could improve the model so that the output midi file could be more consistent.

— Reply to this email directly, view it on GitHub https://github.com/magenta/mt3/issues/134#issuecomment-1683422909, or unsubscribe https://github.com/notifications/unsubscribe-auth/ASU2BOAJIXY6SZRPEG455S3XV4EM5ANCNFSM6AAAAAA2XNQ2W4 . You are receiving this because you commented.Message ID: @.***>

[PimTournaye]

Unfortunately, I'm also getting crashes, both with the official notebook, and with @Jackl-o-o-l 's notebook. In the cell below 'Load Model":

---------------------------------------------------------------------------
ImportError                               Traceback (most recent call last)
[<ipython-input-9-30d8629039fb>](https://localhost:8080/#) in <cell line: 14>()
     12 import librosa
     13 import note_seq
---> 14 import seqio
     15 import t5
     16 import t5x

5 frames
[/usr/local/lib/python3.10/dist-packages/flax/core/axes_scan.py](https://localhost:8080/#) in <module>
     20 from jax import core
     21 from jax import lax
---> 22 from jax.extend import linear_util as lu
     23 from jax.interpreters import partial_eval as pe
     24 import jax.numpy as jnp

ImportError: cannot import name 'linear_util' from 'jax.extend' (/usr/local/lib/python3.10/dist-packages/jax/extend/__init__.py)

---------------------------------------------------------------------------
NOTE: If your import is failing due to a missing package, you can
manually install dependencies using either !pip or !apt.

To view examples of installing some common dependencies, click the
"Open Examples" button below.
---------------------------------------------------------------------------

[gtr3qq]

Unfortunately, I'm also getting crashes, both with the official notebook, and with @Jackl-o-o-l 's notebook. In the cell below 'Load Model":

---------------------------------------------------------------------------
ImportError                               Traceback (most recent call last)
[<ipython-input-9-30d8629039fb>](https://localhost:8080/#) in <cell line: 14>()
     12 import librosa
     13 import note_seq
---> 14 import seqio
     15 import t5
     16 import t5x

5 frames
[/usr/local/lib/python3.10/dist-packages/flax/core/axes_scan.py](https://localhost:8080/#) in <module>
     20 from jax import core
     21 from jax import lax
---> 22 from jax.extend import linear_util as lu
     23 from jax.interpreters import partial_eval as pe
     24 import jax.numpy as jnp

ImportError: cannot import name 'linear_util' from 'jax.extend' (/usr/local/lib/python3.10/dist-packages/jax/extend/__init__.py)

---------------------------------------------------------------------------
NOTE: If your import is failing due to a missing package, you can
manually install dependencies using either !pip or !apt.

To view examples of installing some common dependencies, click the
"Open Examples" button below.
---------------------------------------------------------------------------

same

[Kaes3kuch3n]

Looks like the latest flax version broke something. It started working again for me after downgrading to flax==0.7.2

[gtr3qq]

Looks like the latest flax version broke something. It started working again for me after downgrading to flax==0.7.2

Yeah! it works now!

[gtr3qq]

It worked, finally. I also downgraded bokeh so that midi image could come out normally. https://colab.research.google.com/drive/1JkWyAwFhAC6SJJ1Lvf4VPap6FwtniEuy?usp=sharing

Worked for me but still It produces absurd midi files as I cannot recognize the original sound of the song also It splits one guitar into several absurd identifiers! Guitar songs are lost tot this AI.

Would you mind to update the flax ==0.7.2 ? plesse?

[0xdevalias]

TypeError: register_with_handler() got an unexpected keyword argument 'for_restore'

See also:

• https://github.com/magenta/mt3/issues/151

[Apple-jun]

@jsphweid I think if you click "RESTART RUNTIME" after running !pip install numpy==1.23.5 -I, it will work for you. I did not change anything before you ask the question but now I added a line of notice in the notebook.

RESTART RUNTIME doesn't work for me: I run into the same issue.

[0xdevalias]

It seems like upstream was broken, as per this issue:

• https://github.com/google-research/t5x/issues/1550

Within that issue, they mention that the commit before the one they mentioned seemed to work; so presumably this could be fixed by forcing that package to the version before the error was introduced (and then potentially any related dependencies/etc)

Originally posted by @0xdevalias in https://github.com/magenta/mt3/issues/151#issuecomment-2110257006

---

!pip install --upgrade orbax-checkpoint==0.5.8 works as a workaround for me.

Originally posted by @garzon in https://github.com/magenta/mt3/issues/151#issuecomment-2123348933