From fe0fe9127593e99ca79a4159d73b7f3963cad112 Mon Sep 17 00:00:00 2001 From: bingochaos <523834173@qq.com> Date: Fri, 26 Sep 2025 19:21:53 +0800 Subject: [PATCH] run with cpu --- engines/infer.py | 10 +++++----- 1 file changed, 5 insertions(+), 5 deletions(-) diff --git a/engines/infer.py b/engines/infer.py index ffd6cfe..11192f1 100644 --- a/engines/infer.py +++ b/engines/infer.py @@ -65,7 +65,7 @@ class InferBase: n_parameters = sum(p.numel() for p in model.parameters() if p.requires_grad) self.logger.info(f"Num params: {n_parameters}") model = create_ddp_model( - model.cuda(), + model, broadcast_buffers=False, find_unused_parameters=self.cfg.find_unused_parameters, ) @@ -117,9 +117,9 @@ class Audio2ExpressionInfer(InferBase): with torch.no_grad(): input_dict = {} input_dict['id_idx'] = F.one_hot(torch.tensor(self.cfg.id_idx), - self.cfg.model.backbone.num_identity_classes).cuda(non_blocking=True)[None,...] + self.cfg.model.backbone.num_identity_classes)[None,...] speech_array, ssr = librosa.load(self.cfg.audio_input, sr=16000) - input_dict['input_audio_array'] = torch.FloatTensor(speech_array).cuda(non_blocking=True)[None,...] + input_dict['input_audio_array'] = torch.FloatTensor(speech_array)[None,...] end = time.time() output_dict = self.model(input_dict) @@ -198,9 +198,9 @@ class Audio2ExpressionInfer(InferBase): try: input_dict = {} input_dict['id_idx'] = F.one_hot(torch.tensor(self.cfg.id_idx), - self.cfg.model.backbone.num_identity_classes).cuda(non_blocking=True)[ + self.cfg.model.backbone.num_identity_classes)[ None, ...] - input_dict['input_audio_array'] = torch.FloatTensor(input_audio).cuda(non_blocking=True)[None, ...] + input_dict['input_audio_array'] = torch.FloatTensor(input_audio)[None, ...] output_dict = self.model(input_dict) out_exp = output_dict['pred_exp'].squeeze().cpu().numpy()[start_frame:, :] except: