deepgeek/CosyVoice
2
0
Fork 0
mirror of https://github.com/FunAudioLLM/CosyVoice.git synced 2026-02-05 09:59:23 +08:00
Code Issues Packages Projects Releases Wiki Activity

update dpo

Browse Source
This commit is contained in:
lyuxiang.lx
2025-06-13 16:14:05 +08:00
parent cc234bd322
commit 63856565f3
23 changed files with 345 additions and 2024 deletions
Download Patch File Download Diff File Expand all files Collapse all files

3
cosyvoice/cli/model.py
View File

@@ -103,7 +103,7 @@ class CosyVoiceModel:
def llm_job(self, text, prompt_text, llm_prompt_speech_token, llm_embedding, uuid):
with self.llm_context, torch.cuda.amp.autocast(self.fp16 is True and hasattr(self.llm, 'vllm') is False):
if isinstance(text, Generator):
assert isinstance(self, CosyVoice2Model), 'streaming input text is only implemented for CosyVoice2!'
assert isinstance(self, CosyVoice2Model) and not hasattr(self.llm, 'vllm'), 'streaming input text is only implemented for CosyVoice2 and do not support vllm!'
for i in self.llm.inference_bistream(text=text,
prompt_text=prompt_text.to(self.device),
prompt_text_len=torch.tensor([prompt_text.shape[1]], dtype=torch.int32).to(self.device),
@@ -279,6 +279,7 @@ class CosyVoice2Model(CosyVoiceModel):
enable_prompt_embeds=True,
gpu_memory_utilization=0.2)
self.llm.vllm = LLMEngine.from_engine_args(engine_args)
self.llm.lock = threading.Lock()
del self.llm.llm.model.model.layers
def token2wav(self, token, prompt_token, prompt_feat, embedding, token_offset, uuid, stream=False, finalize=False, speed=1.0):