Files
VoxCPM/src/voxcpm/model/utils.py
T
oumnya 38d61cdf03 fix(mps): force float32 on Apple Silicon to avoid bf16 quality loss
VoxCPM checkpoints default to bfloat16. Following commit e4e0496 which
added MPS device routing, running with `device=mps` selects bf16 on
Apple Silicon. On Metal, bf16 introduces enough numerical drift in the
diffusion AR loop that the synthesized audio is glitched and trips the
model's badcase detector, which retries until the per-call retry budget
is exhausted. Effectively MPS support is unusable in the default config.

This patch adds a single helper, `pick_runtime_dtype(device, dtype)`,
that promotes any low-precision dtype to float32 when the resolved
device is `mps`. CUDA and CPU paths are untouched. An opt-out env var
`VOXCPM_MPS_DTYPE` lets users force a specific dtype on MPS once future
PyTorch / macOS releases improve bf16 stability.

Both VoxCPMModel and VoxCPM2Model adopt the helper in their __init__,
replacing what would otherwise be duplicated inline checks.

Verified locally on Apple M5 Max, PyTorch 2.11, macOS 15:
- VoxCPM2 (2B): clean output, RTF ~0.78 steady state
- VoxCPM 0.5B: clean output, RTF ~0.92
- No badcase retries fired in any test
- VOXCPM_MPS_DTYPE=bfloat16 round-trips and reproduces the original
  glitched output, confirming the override path.
2026-04-15 12:22:56 +08:00

232 lines
7.8 KiB
Python

import os
from typing import List, Optional
import torch
from transformers import PreTrainedTokenizer
_LOW_PRECISION_DTYPES = {"bfloat16", "bf16", "float16", "fp16", "half"}
_VALID_DTYPE_OVERRIDES = {
"bfloat16", "bf16",
"float16", "fp16", "half",
"float32", "fp32",
}
# Ref: https://github.com/OpenBMB/VoxCPM/issues/256#issuecomment-4235252732
# Explicitly close partially-consumed generators so inference_mode cleanup
# does not get deferred to Python's GC/finalizer path.
def next_and_close(gen):
try:
return next(gen)
finally:
gen.close()
def mask_multichar_chinese_tokens(tokenizer: PreTrainedTokenizer):
"""Create a tokenizer wrapper that converts multi-character Chinese tokens to single characters.
This function creates a wrapper around the provided tokenizer that automatically
splits multi-character Chinese tokens into individual characters. This is useful
for ensuring consistent tokenization of Chinese text.
Args:
tokenizer: The base tokenizer to wrap
Returns:
A CharTokenizerWrapper instance that handles multi-character Chinese tokens
Example:
>>> from transformers import LlamaTokenizerFast
>>> tokenizer = LlamaTokenizerFast.from_pretrained("path/to/tokenizer")
>>> wrapped_tokenizer = mask_multichar_chinese_tokens(tokenizer)
>>> tokens = wrapped_tokenizer("你好世界")
"""
# Pre-compute multi-character tokens (length >= 2, pure Chinese characters)
multichar_tokens = {
token for token in tokenizer.vocab.keys() if len(token) >= 2 and all("\u4e00" <= c <= "\u9fff" for c in token)
}
class CharTokenizerWrapper:
"""Wrapper class for tokenizers that handles multi-character Chinese tokens.
This wrapper automatically splits multi-character Chinese tokens into
individual characters while preserving the original tokenizer's interface.
"""
def __init__(self, base_tokenizer: PreTrainedTokenizer) -> None:
"""Initialize the wrapper with a base tokenizer.
Args:
base_tokenizer: The tokenizer to wrap
"""
self.tokenizer = base_tokenizer
self.multichar_tokens = multichar_tokens
def tokenize(self, text: str, **kwargs) -> List[str]:
"""Tokenize text and split multi-character Chinese tokens into single characters.
Args:
text: Input text to tokenize
**kwargs: Additional arguments passed to the base tokenizer
Returns:
List of processed tokens with multi-character Chinese tokens split
Example:
>>> wrapper = CharTokenizerWrapper(tokenizer)
>>> tokens = wrapper.tokenize("你好世界")
>>> # Returns ["", "", "", ""] instead of ["你好", "世界"]
"""
if not isinstance(text, str):
raise TypeError(f"Expected string input, got {type(text)}")
tokens = self.tokenizer.tokenize(text, **kwargs)
processed = []
for token in tokens:
# Remove possible subword prefix
clean_token = token.replace("", "")
if clean_token in self.multichar_tokens:
# Split multi-character token into single characters
chars = list(clean_token)
processed.extend(chars)
else:
processed.append(token)
return processed
def __call__(self, text: str, **kwargs) -> List[int]:
"""Call the tokenizer and return token IDs.
This method provides the same interface as the original tokenizer
but with multi-character Chinese token handling.
Args:
text: Input text to tokenize
**kwargs: Additional arguments passed to the base tokenizer
Returns:
List of token IDs
Raises:
TypeError: If input is not a string
ValueError: If tokenization fails
"""
try:
tokens = self.tokenize(text, **kwargs)
result = self.tokenizer.convert_tokens_to_ids(tokens)
return result
except Exception as e:
raise ValueError(f"Tokenization failed: {str(e)}") from e
return CharTokenizerWrapper(tokenizer)
def get_dtype(dtype: str):
if dtype == "bfloat16":
return torch.bfloat16
elif dtype == "bf16":
return torch.bfloat16
elif dtype == "float16":
return torch.float16
elif dtype == "fp16":
return torch.float16
elif dtype == "float32":
return torch.float32
elif dtype == "fp32":
return torch.float32
else:
raise ValueError(f"Unsupported dtype: {dtype}")
def _has_mps() -> bool:
return hasattr(torch.backends, "mps") and torch.backends.mps.is_available()
def pick_runtime_dtype(device: str, configured_dtype: str) -> str:
"""Pick a safe runtime dtype for the resolved device.
On Apple Silicon (MPS), bfloat16/float16 produce enough numerical drift
in the diffusion AR loop that the output is glitched and the model's
badcase detector triggers infinite retries. float32 is the only stable
option today. CUDA and CPU keep whatever the checkpoint was trained with.
Users can override with ``VOXCPM_MPS_DTYPE`` (e.g. ``bfloat16``) when
they want to test future MPS improvements.
"""
if device != "mps":
return configured_dtype
override = os.environ.get("VOXCPM_MPS_DTYPE", "").strip().lower()
if override:
if override not in _VALID_DTYPE_OVERRIDES:
raise ValueError(
f"VOXCPM_MPS_DTYPE='{override}' is not one of "
f"{sorted(_VALID_DTYPE_OVERRIDES)}"
)
return override
if (configured_dtype or "").lower() in _LOW_PRECISION_DTYPES:
return "float32"
return configured_dtype
def auto_select_device(preferred_device: Optional[str] = "cuda") -> str:
"""
Choose a runtime device automatically.
Preference order:
- if the preferred device is available, use it
- otherwise fall back to CUDA -> MPS -> CPU
"""
preferred = (preferred_device or "cuda").strip().lower()
if preferred.startswith("cuda") and torch.cuda.is_available():
return preferred
if preferred == "mps" and _has_mps():
return "mps"
if preferred == "cpu":
return "cpu"
if torch.cuda.is_available():
return "cuda"
if _has_mps():
return "mps"
return "cpu"
def resolve_runtime_device(device: Optional[str], configured_device: str = "cuda") -> str:
"""
Resolve the actual runtime device.
Semantics:
- ``device`` is ``None`` or ``"auto"``: use automatic fallback selection
- otherwise: treat it as an explicit user choice and validate availability
"""
explicit = None if device is None else device.strip().lower()
if explicit is None or explicit == "auto":
return auto_select_device(configured_device)
if explicit.startswith("cuda"):
if not torch.cuda.is_available():
raise ValueError(
f"Requested device '{device}', but CUDA is not available. "
"Use device='auto' for automatic fallback."
)
return explicit
if explicit == "mps":
if not _has_mps():
raise ValueError(
"Requested device 'mps', but MPS is not available. "
"Use device='auto' for automatic fallback."
)
return "mps"
if explicit == "cpu":
return "cpu"
raise ValueError(
f"Unsupported device '{device}'. Supported values are 'auto', 'cpu', 'mps', "
"'cuda', or indexed CUDA devices like 'cuda:0'."
)