Enhanced Colab notebook

[SMarioMan]

Added support for priming the model.
Data now persists in Google Drive.
Added the ability to resume from checkpoints.

[thisIsTheFoxe]

just as a though: since Downloading from gce (Downloading the model, I assume..?) takes quite a long time, especially for the 5b -models, it might make sense to split those two. So you can edit the Hyperparams without reloading the model each time :)

[MysteryPancake]

Sorry to bother, but I was wondering if there is a way to use v3 labels with this Colab?

[xandramax]

Which labels are used is tied to which model is used: 5b uses v2 labels and 1b uses v3 labels.

[ApexJohn]

Is it possible to upscale audio made with 1b with this while keeping artist data and such intact? I keep getting an error with it not recognizing the artist as it seems to try and use 5b and v2 artists, instead of v3.

[Yonben]

Hi,
In the notebook while using '5b' and not '5b_lyrics', the level_2 generation returns the following error: jukebox Range is [2646000.0,26460000.0), got tensor([[2204928.],[2204928.],[2204928.]], device='cuda:0') I tried factory resetting etc... and it seems to be consistently reproducable.

[liamwazherealso]

Facing the same issue as @Yonben.

Using primed 5b.
Only modifcation is.

metas = [dict(artist = "Yello",
            genre = "Electronic",
            total_length = hps.sample_length,
            offset = 0,
            lyrics=" "*40),
          ] * hps.n_samples```
  

Full Error Output

Sampling level 2
Sampling 8192 tokens for [0,8192]. Conditioning on 3445 tokens
Primed sampling 1 samples with temp=0.98, top_k=0, top_p=0.0

---------------------------------------------------------------------------

AssertionError                            Traceback (most recent call last)

<ipython-input-49-988948e1e679> in <module>()
     16   x = load_prompts(audio_files, duration, hps)
     17   zs = top_prior.encode(x, start_level=0, end_level=len(priors), bs_chunks=x.shape[0])
---> 18   zs = _sample(zs, labels, sampling_kwargs, [None, None, top_prior], [2], hps)
     19 else:
     20   raise ValueError(f'Unknown sample mode {sample_hps.mode}.')

8 frames

/usr/local/lib/python3.6/dist-packages/jukebox/prior/conditioners.py in forward(self, pos_start, pos_end)
     89         # Check if [pos_start,pos_end] in [pos_min, pos_max)
     90         assert len(pos_start.shape) == 2, f"Expected shape with 2 dims, got {pos_start.shape}"
---> 91         assert (self.pos_min <= pos_start).all() and (pos_start < self.pos_max).all(), f"Range is [{self.pos_min},{self.pos_max}), got {pos_start}"
     92         pos_start = pos_start.float()
     93         if pos_end is not None:

AssertionError: Range is [2646000.0,26460000.0), got tensor([[2204928.]], device='cuda:0')
          ] * hps.n_sample