2noise/ChatTTS · error · ValueError
Total number of attention heads ({total_num_attention_heads}
Error message
Total number of attention heads ({total_num_attention_heads}) must be divisible by tensor parallel size ({tensor_parallel_size}). What it means
Error "Total number of attention heads ({total_num_attention_heads}) must be divisible by tensor parallel size ({tensor_parallel_size})." thrown in 2noise/ChatTTS.
Source
Thrown at ChatTTS/model/velocity/configs.py:210
"optimized yet. The speed can be slower than "
"non-quantized models."
)
def _verify_cuda_graph(self) -> None:
if self.max_context_len_to_capture is None:
self.max_context_len_to_capture = self.max_model_len
self.max_context_len_to_capture = min(
self.max_context_len_to_capture, self.max_model_len
)
def verify_with_parallel_config(
self,
parallel_config: "ParallelConfig",
) -> None:
total_num_attention_heads = self.hf_config.num_attention_heads
tensor_parallel_size = parallel_config.tensor_parallel_size
if total_num_attention_heads % tensor_parallel_size != 0:
raise ValueError(
f"Total number of attention heads ({total_num_attention_heads})"
" must be divisible by tensor parallel size "
f"({tensor_parallel_size})."
)
total_num_hidden_layers = self.hf_config.num_hidden_layers
pipeline_parallel_size = parallel_config.pipeline_parallel_size
if total_num_hidden_layers % pipeline_parallel_size != 0:
raise ValueError(
f"Total number of hidden layers ({total_num_hidden_layers}) "
"must be divisible by pipeline parallel size "
f"({pipeline_parallel_size})."
)
def get_sliding_window(self) -> Optional[int]:
return getattr(self.hf_config, "sliding_window", None)
def get_vocab_size(self) -> int:View on GitHub (pinned to 77b89ee281)
When it happens
Trigger: Thrown at ChatTTS/model/velocity/configs.py:210 when the library encounters an invalid state.
Common situations: See trigger scenarios.
AI-assisted analysis of 2noise/ChatTTS@77b89ee281 (2026-08-26).
Data as JSON: /api/errors/cf1689d862840d3e.
Report an issue: GitHub.