Error on running `sample_text_to_3d.ipynb`

[nourheshamshaheen]

I receive this error when running on python=3.10, pytorch=1.11.0+cu113, torchvision=0.12.0+cu113

TypeError                                 Traceback (most recent call last)
Cell In[8], line 4
      1 batch_size = 4
      2 guidance_scale = 15.0
----> 4 latents = sample_latents(
      5     batch_size=batch_size,
      6     model=model,
      7     diffusion=diffusion,
      8     guidance_scale=guidance_scale,
      9     model_kwargs=dict(texts=[prompt] * batch_size),
     10     progress=True,
     11     clip_denoised=True,
     12     use_fp16=True,
     13     use_karras=True,
     14     karras_steps=64,
     15     sigma_min=1e-3,
     16     sigma_max=160,
     17     s_churn=0,
     18 )

File ~/.conda/envs/cv2/lib/python3.10/site-packages/shap_e/diffusion/sample.py:62, in sample_latents(batch_size, model, diffusion, model_kwargs, guidance_scale, clip_denoised, use_fp16, use_karras, karras_steps, sigma_min, sigma_max, s_churn, device, progress)
     60 with torch.autocast(device_type=device.type, enabled=use_fp16):
     61     if use_karras:
---> 62         samples = karras_sample(
     63             diffusion=diffusion,
     64             model=model,
     65             shape=sample_shape,
     66             steps=karras_steps,
     67             clip_denoised=clip_denoised,
     68             model_kwargs=model_kwargs,
     69             device=device,
     70             sigma_min=sigma_min,
     71             sigma_max=sigma_max,
     72             s_churn=s_churn,
     73             guidance_scale=guidance_scale,
     74             progress=progress,
     75         )
     76     else:
     77         internal_batch_size = batch_size

File ~/.conda/envs/cv2/lib/python3.10/site-packages/shap_e/diffusion/k_diffusion.py:113, in karras_sample(*args, **kwargs)
    111 def karras_sample(*args, **kwargs):
    112     last = None
--> 113     for x in karras_sample_progressive(*args, **kwargs):
    114         last = x["x"]
    115     return last

File ~/.conda/envs/cv2/lib/python3.10/site-packages/shap_e/diffusion/k_diffusion.py:181, in karras_sample_progressive(diffusion, model, shape, steps, clip_denoised, progress, model_kwargs, device, sigma_min, sigma_max, rho, sampler, s_churn, s_tmin, s_tmax, s_noise, guidance_scale)
    178 else:
    179     guided_denoiser = denoiser
--> 181 for obj in sample_fn(
    182     guided_denoiser,
    183     x_T,
    184     sigmas,
    185     progress=progress,
    186     **sampler_args,
    187 ):
    188     if isinstance(diffusion, GaussianDiffusion):
    189         yield diffusion.unscale_out_dict(obj)

File ~/.conda/envs/cv2/lib/python3.10/site-packages/torch/autograd/grad_mode.py:43, in _DecoratorContextManager._wrap_generator.<locals>.generator_context(*args, **kwargs)
     40 try:
     41     # Issuing `None` to a generator fires it up
     42     with self.clone():
---> 43         response = gen.send(None)
     45     while True:
     46         try:
     47             # Forward the response to our caller and get its next request

File ~/.conda/envs/cv2/lib/python3.10/site-packages/shap_e/diffusion/k_diffusion.py:265, in sample_heun(denoiser, x, sigmas, progress, s_churn, s_tmin, s_tmax, s_noise)
    263 if gamma > 0:
    264     x = x + eps * (sigma_hat**2 - sigmas[i] ** 2) ** 0.5
--> 265 denoised = denoiser(x, sigma_hat * s_in)
    266 d = to_d(x, sigma_hat, denoised)
    267 yield {"x": x, "i": i, "sigma": sigmas[i], "sigma_hat": sigma_hat, "pred_xstart": denoised}

File ~/.conda/envs/cv2/lib/python3.10/site-packages/shap_e/diffusion/k_diffusion.py:173, in karras_sample_progressive.<locals>.guided_denoiser(x_t, sigma)
    171 x_t = th.cat([x_t, x_t], dim=0)
    172 sigma = th.cat([sigma, sigma], dim=0)
--> 173 x_0 = denoiser(x_t, sigma)
    174 cond_x_0, uncond_x_0 = th.split(x_0, len(x_0) // 2, dim=0)
    175 x_0 = uncond_x_0 + guidance_scale * (cond_x_0 - uncond_x_0)

File ~/.conda/envs/cv2/lib/python3.10/site-packages/shap_e/diffusion/k_diffusion.py:160, in karras_sample_progressive.<locals>.denoiser(x_t, sigma)
    159 def denoiser(x_t, sigma):
--> 160     _, denoised = model.denoise(
    161         x_t, sigma, clip_denoised=clip_denoised, model_kwargs=model_kwargs
    162     )
    163     return denoised

File ~/.conda/envs/cv2/lib/python3.10/site-packages/shap_e/diffusion/k_diffusion.py:99, in GaussianToKarrasDenoiser.denoise(self, x_t, sigmas, clip_denoised, model_kwargs)
     98 def denoise(self, x_t, sigmas, clip_denoised=True, model_kwargs=None):
---> 99     t = th.tensor(
    100         [self.sigma_to_t(sigma) for sigma in sigmas.cpu().numpy()],
    101         dtype=th.long,
    102         device=sigmas.device,
    103     )
    104     c_in = append_dims(1.0 / (sigmas**2 + 1) ** 0.5, x_t.ndim)
    105     out = self.diffusion.p_mean_variance(
    106         self.model, x_t * c_in, t, clip_denoised=clip_denoised, model_kwargs=model_kwargs
    107     )

TypeError: 'float' object cannot be interpreted as an integer

[azigler]

I have this same issue, running the notebook in JupyterLab.

Python implementation : CPython
Python version        : 3.10.9
IPython version       : 8.10.0
pytorch version       : 1.12.1
torchvision version   : 0.13.1

Compiler    : Clang 14.0.6 
OS          : Darwin
Release     : 22.5.0
Machine     : arm64
Processor   : arm
CPU cores   : 10
Architecture: 64bit

[xiaohulihutu]

same here

[nebomuk]

for me everything works with pytorch 2.x instead of 1.x

[azigler]

@nebomuk Thanks for that input! I install the latest pytorch (conda install pytorch torchvision -c pytorch), reset the repository, and rebuilt shap-e. I'm running it now in Jupyter Notebook (was having issue with the widgets in Lab) and no longer getting the error.

I'm running this on an M1 for the last 15 minutes but still haven't seen any progress on this step:

I see the kernel is busy and is taking 100% of CPU but it's still 0/64 on progress. How long should this notebook take to successfully run? Is there a configuration I can make in Jupyter Notebook to increase the allotted resources?

[nebomuk]

I also had some weird progress reporting. Sometimes the progress would start delayed, results appear delayed or i had to click it twice. But at least there doesn't seem to be a bug in shap-e because in the end everything worked.

[azigler]

In my case, because I'm on M1 and don't have access to cuda it's running on CPU and it's just unusably slow. I'll have to try on a different machine's GPU.

[MrAntonio7]

I'm work with pytorch version 1.13.1 and i have the same error (TypeError: 'float' object cannot be interpreted as an integer).

I solve the error converting in integer with int(), in the line 106. Here are the solution:

t = th.tensor( [int(self.sigma_to_t(sigma)) for sigma in sigmas.cpu().numpy()], dtype=th.long, device=sigmas.device, )

[eliecer696]

there is an addon that simplified the use of the model within a 3d modeling software check it out https://devbud.gumroad.com/l/Shap-e