[Model] add soulx-singer support (#3862)
Signed-off-by: Mr.Dong <1042542469@qq.com> Signed-off-by: Yueqian Lin <linyueqian@outlook.com> Co-authored-by: Mr.Dong <1042542469@qq.com> Co-authored-by: Yueqian Lin <70319226+linyueqian@users.noreply.github.com> Co-authored-by: Yueqian Lin <linyueqian@outlook.com>
This commit is contained in:
50 files changed
+9533
-13
No files matched your search
@@ -0,0 +1,24 @@
|
||||
import torch
|
||||
import torch.nn as nn
|
||||
|
||||
from vllm_omni.diffusion.models.soulx_singer.modules.flow_matching import FlowMatchingTransformer
|
||||
|
||||
|
||||
class CFMDecoder(nn.Module):
|
||||
def __init__(self, config):
|
||||
super().__init__()
|
||||
self.model = FlowMatchingTransformer(cfg=config, **config)
|
||||
|
||||
def forward(self, mel, x_mask, decoder_inp, is_prompt):
|
||||
outputs = self.model(mel, x_mask, decoder_inp, is_prompt)
|
||||
|
||||
noise, x, flow_pred, final_mask, prompt_len = outputs["output"]
|
||||
return noise, x, flow_pred, final_mask, prompt_len
|
||||
|
||||
def reverse_diffusion(self, pt_mel, pt_decoder_inp, gt_decoder_inp, n_timesteps=32, cfg=1):
|
||||
diffusion_cond = torch.cat([pt_decoder_inp, gt_decoder_inp], dim=1)
|
||||
diffusion_cond_emb = self.model.cond_emb(diffusion_cond)
|
||||
diffusion_prompt = pt_mel
|
||||
|
||||
generated = self.model.reverse_diffusion(diffusion_cond_emb, diffusion_prompt, n_timesteps=n_timesteps, cfg=cfg)
|
||||
return generated
|
||||
Reference in new issue
Block a user