deepgeek/CosyVoice
2
0
Fork 0
mirror of https://github.com/FunAudioLLM/CosyVoice.git synced 2026-02-04 09:29:25 +08:00
Code Issues Packages Projects Releases Wiki Activity

Fix CosyVoice3 config error

Browse Source
This commit is contained in:
di.wu
2025-12-17 14:56:21 +08:00
parent f5816b4e51
commit ff0d05c380
2 changed files with 3 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
cosyvoice/llm/llm.py
View File

@@ -698,7 +698,7 @@ class CosyVoice3LM(Qwen2LM):
lm_output, lm_output_mask = self.llm(lm_input, lm_input_len.to(device)) lm_output, lm_output_mask = self.llm(lm_input, lm_input_len.to(device))
logits = self.llm_decoder(lm_output) logits = self.llm_decoder(lm_output)
loss = self.criterion_ce(logits, lm_target.to(device)) loss = self.criterion_ce(logits, lm_target.to(device))
acc = th_accuracy(logits.view(-1, self.speech_token_size + 3), lm_target, ignore_label=IGNORE_ID) acc = th_accuracy(logits.view(-1, self.speech_token_size + 200), lm_target, ignore_label=IGNORE_ID)
return {'loss': loss, 'acc': acc} return {'loss': loss, 'acc': acc}
@torch.inference_mode() @torch.inference_mode()

4
examples/libritts/cosyvoice3/conf/cosyvoice3.yaml
View File

@@ -20,7 +20,7 @@ num_decoding_left_chunks: -1 # streaming inference flow decoder left chunk size,
# model params # model params
# for all class/function included in this repo, we use !<name> or !<new> for intialization, so that user may find all corresponding class/function according to one single yaml. # for all class/function included in this repo, we use !<name> or !<new> for intialization, so that user may find all corresponding class/function according to one single yaml.
# for system/third_party class/function, we do not require this. # for system/third_party class/function, we do not require this.
llm: !new:cosyvoice.llm.llm.Qwen2LM llm: !new:cosyvoice.llm.llm.CosyVoice3LM
llm_input_size: !ref <llm_input_size> llm_input_size: !ref <llm_input_size>
llm_output_size: !ref <llm_output_size> llm_output_size: !ref <llm_output_size>
speech_token_size: 6561 speech_token_size: 6561
@@ -231,4 +231,4 @@ train_conf_gan:
grad_clip: 5 grad_clip: 5
accum_grad: 1 # in gan training, accum_grad must be 1 accum_grad: 1 # in gan training, accum_grad must be 1
log_interval: 100 log_interval: 100
save_per_step: -1 save_per_step: -1