MCPcopy Create free account

hub / github.com/boson-ai/higgs-audio / functions

Functions316 in github.com/boson-ai/higgs-audio

Functionsp_size
(self)
boson_multimodal/model/higgs_audio/utils.py:451
Functionspell_out_number
(text: str, inflect_parser)
boson_multimodal/serve/utils.py:160
Functionsplit_interleaved_delayed_audios
( audio_data: Union[list[list[int]], torch.Tensor], audio_tokenizer: HiggsAudioTokenizer, audio_st
boson_multimodal/serve/utils.py:212
Functionsplit_paragraph
(text: str, tokenize, lang="zh", token_max_n=80, token_min_n=60, merge_len=20, comma_split=False)
boson_multimodal/serve/utils.py:98
Functionsupport_deepspeed_ulysses
A decorator around Pytorch module. It is needed for the module that needs access to sequence parallel info.
boson_multimodal/model/higgs_audio/utils.py:444
Methodsymbolic
(graph, input_, group)
boson_multimodal/model/higgs_audio/utils.py:561
Methodsymbolic
(graph, input_, dim, group, grad_scale)
boson_multimodal/model/higgs_audio/utils.py:650
Methodsymbolic
(graph, input_, dim, group, grad_scale)
boson_multimodal/model/higgs_audio/utils.py:672
Functionsync_buffer
Sync grad for buffers. If average is False, broadcast instead of averaging.
boson_multimodal/audio_processing/quantization/distrib.py:73
Functionsync_grad
Simpler alternative to DistributedDataParallel, that doesn't rely on any black magic. For simple models it can also be as fast. Just call
boson_multimodal/audio_processing/quantization/distrib.py:93
Methodto_unsplit
(self)
boson_multimodal/model/higgs_audio/custom_modules.py:81
Methodto_unsplit
(self)
boson_multimodal/model/higgs_audio/custom_modules.py:141
Methodtps
(self)
boson_multimodal/audio_processing/higgs_audio_tokenizer.py:138
Functionuniform_init
(*shape: int)
boson_multimodal/audio_processing/quantization/core_vq_lsx_version.py:66
Functionuniform_init
(*shape: int)
boson_multimodal/audio_processing/quantization/core_vq.py:56
Functionwrapped
(*args, **kwargs)
boson_multimodal/model/higgs_audio/utils.py:483
← previous301–316 of 316, ranked by callers