Inference issues with Dia-1.6B: very short audio, extremely long noisy audio, or no audio output

import soundfile as sf
from dia.model import Dia

model = Dia.from_pretrained("nari-labs/Dia-1.6B-0626")

text = "[S1] Dia is an open weights text to dialogue model. [S2] You get full control over scripts and voices. [S1] Wow. Amazing. (laughs) [S2] Try it now on GitHub or Hugging Face."

output = model.generate(text)

sf.write("simple.mp3", output, 44100)

---------------------------------------------------------------------------
RuntimeError                              Traceback (most recent call last)
Cell In[7], line 32
     28 try:
     29     # Generate audio
     30     start_time = time.time()
---> 32     output_audio = model.generate(
     33         text=text,
     34         # max_tokens=MAX_TOKENS,
     35         # cfg_scale=CFG_SCALE,
     36         # temperature=TEMPERATURE,
     37         # top_p=TOP_P,
     38     )
     40     print("✓ Audio generation complete!")
     41     print(f"Audio shape: {output_audio.shape}")

File ~/miniconda3/envs/dia-tts/lib/python3.12/site-packages/torch/utils/_contextlib.py:116, in context_decorator.<locals>.decorate_context(*args, **kwargs)
    113 @functools.wraps(func)
    114 def decorate_context(*args, **kwargs):
    115     with ctx_factory():
--> 116         return func(*args, **kwargs)

File /workspace/dia-finetuning/dia/model.py:464, in Dia.generate(self, text, max_tokens, cfg_scale, temperature, top_p, use_cfg_filter, use_torch_compile, cfg_filter_top_k, audio_prompt_path)
    460 output_codes = generated_BxTxC[:, prompt_len_inc_bos : step + 1, :]
    462 generated_codes = output_codes[0]
--> 464 audio = codebook_to_audio(
    465     generated_codes.transpose(1, 0), self.dac_model, delay_pattern, B=1, T=max_tokens, C=num_channels
    466 )
    467 return audio.squeeze().cpu().numpy()

File /workspace/dia-finetuning/dia/audio.py:281, in codebook_to_audio(generated_codes, model, delay_pattern, B, T, C)
    279 # Set invalid values to 0 (modify the tensor in-place)
    280 codebook[invalid_mask] = 0
--> 281 audio_array = decode(model, codebook)
    283 return audio_array

File ~/miniconda3/envs/dia-tts/lib/python3.12/site-packages/torch/utils/_contextlib.py:116, in context_decorator.<locals>.decorate_context(*args, **kwargs)
    113 @functools.wraps(func)
    114 def decorate_context(*args, **kwargs):
    115     with ctx_factory():
--> 116         return func(*args, **kwargs)

File ~/miniconda3/envs/dia-tts/lib/python3.12/site-packages/torch/utils/_contextlib.py:116, in context_decorator.<locals>.decorate_context(*args, **kwargs)
    113 @functools.wraps(func)
    114 def decorate_context(*args, **kwargs):
    115     with ctx_factory():
--> 116         return func(*args, **kwargs)

File /workspace/dia-finetuning/dia/audio.py:236, in decode(model, audio_codes)
    231     raise ValueError(f"Expected one frame, got {len(audio_codes)}")
    233 try:
    234     # print(audio_codes.shape)
    235     # print(audio_codes)
--> 236     audio_values = model.quantizer.from_codes(audio_codes)
    237     audio_values = model.decode(audio_values[0])
    239     return audio_values

File ~/miniconda3/envs/dia-tts/lib/python3.12/site-packages/dac/nn/quantize.py:218, in ResidualVectorQuantize.from_codes(self, codes)
    215     z_p_i = self.quantizers[i].decode_code(codes[:, i, :])
    216     z_p.append(z_p_i)
--> 218     z_q_i = self.quantizers[i].out_proj(z_p_i)
    219     z_q = z_q + z_q_i
    220 return z_q, torch.cat(z_p, dim=1), codes

File ~/miniconda3/envs/dia-tts/lib/python3.12/site-packages/torch/nn/modules/module.py:1739, in Module._wrapped_call_impl(self, *args, **kwargs)
   1737     return self._compiled_call_impl(*args, **kwargs)  # type: ignore[misc]
   1738 else:
-> 1739     return self._call_impl(*args, **kwargs)

File ~/miniconda3/envs/dia-tts/lib/python3.12/site-packages/torch/nn/modules/module.py:1845, in Module._call_impl(self, *args, **kwargs)
   1842     return inner()
   1844 try:
-> 1845     return inner()
   1846 except Exception:
   1847     # run always called hooks if they have not already been run
   1848     # For now only forward hooks have the always_call option but perhaps
   1849     # this functionality should be added to full backward hooks as well.
   1850     for hook_id, hook in _global_forward_hooks.items():

File ~/miniconda3/envs/dia-tts/lib/python3.12/site-packages/torch/nn/modules/module.py:1793, in Module._call_impl.<locals>.inner()
   1790     bw_hook = BackwardHook(self, full_backward_hooks, backward_pre_hooks)
   1791     args = bw_hook.setup_input_hook(args)
-> 1793 result = forward_call(*args, **kwargs)
   1794 if _global_forward_hooks or self._forward_hooks:
   1795     for hook_id, hook in (
   1796         *_global_forward_hooks.items(),
   1797         *self._forward_hooks.items(),
   1798     ):
   1799         # mark that always called hook is run

File ~/miniconda3/envs/dia-tts/lib/python3.12/site-packages/torch/nn/modules/conv.py:375, in Conv1d.forward(self, input)
    374 def forward(self, input: Tensor) -> Tensor:
--> 375     return self._conv_forward(input, self.weight, self.bias)

File ~/miniconda3/envs/dia-tts/lib/python3.12/site-packages/torch/nn/modules/conv.py:370, in Conv1d._conv_forward(self, input, weight, bias)
    358 if self.padding_mode != "zeros":
    359     return F.conv1d(
    360         F.pad(
    361             input, self._reversed_padding_repeated_twice, mode=self.padding_mode
   (...)    368         self.groups,
    369     )
--> 370 return F.conv1d(
    371     input, weight, bias, self.stride, self.padding, self.dilation, self.groups
    372 )

RuntimeError: Calculated padded input size per channel: (0). Kernel size: (1). Kernel size can't be greater than actual input size