Spaces:
Running
on
Zero
Running
on
Zero
Update infer.py
Browse files
infer.py
CHANGED
@@ -50,9 +50,10 @@ def load_model(model_name, audio_tokenizer_path):
|
|
50 |
use_flash_attention_2=True,
|
51 |
use_cache=True,
|
52 |
)
|
53 |
-
model = model.cuda
|
54 |
tokenizer = AutoTokenizer.from_pretrained(model_name)
|
55 |
-
tokenizer_voila = VoilaTokenizer(model_path=audio_tokenizer_path, device="
|
|
|
56 |
return model, tokenizer, tokenizer_voila, model_type
|
57 |
|
58 |
def is_audio_output_task(task_type):
|
@@ -90,11 +91,11 @@ def eval_model(model, tokenizer, tokenizer_voila, model_type, task_type, history
|
|
90 |
yield all_tokens[:,i]
|
91 |
|
92 |
if model_type == "autonomous":
|
93 |
-
input_generator = get_input_generator(torch.as_tensor(streaming_user_input_audio_tokens).cuda
|
94 |
-
input_ids = [torch.as_tensor([input]).transpose(1,2).cuda
|
95 |
input_ids = torch.cat(input_ids, dim=2) # concat to [bs, seq, num_codebooks*2]
|
96 |
else:
|
97 |
-
input_ids = torch.as_tensor([input_ids]).transpose(1,2).cuda
|
98 |
gen_params = {
|
99 |
"input_ids": input_ids,
|
100 |
"ref_embs": ref_embs,
|
@@ -110,8 +111,8 @@ def eval_model(model, tokenizer, tokenizer_voila, model_type, task_type, history
|
|
110 |
"audio_top_k": 50,
|
111 |
}
|
112 |
if model_type == "audio":
|
113 |
-
audio_datas = torch.tensor([audio_datas], dtype=torch.bfloat16).cuda
|
114 |
-
audio_data_masks = torch.tensor([audio_data_masks]).cuda
|
115 |
gen_params["audio_datas"] = audio_datas
|
116 |
gen_params["audio_data_masks"] = audio_data_masks
|
117 |
elif model_type == "autonomous":
|
@@ -141,7 +142,7 @@ def eval_model(model, tokenizer, tokenizer_voila, model_type, task_type, history
|
|
141 |
'text': tokenizer.decode(text_outputs),
|
142 |
}
|
143 |
if is_audio_output_task(task_type):
|
144 |
-
audio_values = tokenizer_voila.decode(torch.tensor(audio_outputs).cuda
|
145 |
out['audio'] = (audio_values.detach().cpu().numpy(), 16000)
|
146 |
return out
|
147 |
|
@@ -185,10 +186,12 @@ if __name__ == "__main__":
|
|
185 |
# step2: encode ref
|
186 |
ref_embs, ref_embs_mask = None, None
|
187 |
if is_audio_output_task(args.task_type):
|
188 |
-
spkr_model = SpeakerEmbedding(device="
|
|
|
|
|
189 |
wav, sr = torchaudio.load(args.ref_audio)
|
190 |
ref_embs = spkr_model(wav, sr)
|
191 |
-
ref_embs_mask = torch.tensor([1]).cuda
|
192 |
|
193 |
out = eval_model(model, tokenizer, tokenizer_voila, model_type, args.task_type, history, ref_embs, ref_embs_mask)
|
194 |
print(f"Output str: {out['text']}")
|
|
|
50 |
use_flash_attention_2=True,
|
51 |
use_cache=True,
|
52 |
)
|
53 |
+
model = model.to("cuda")
|
54 |
tokenizer = AutoTokenizer.from_pretrained(model_name)
|
55 |
+
tokenizer_voila = VoilaTokenizer(model_path=audio_tokenizer_path, device="cpu")
|
56 |
+
tokenizer_voila.to("cuda")
|
57 |
return model, tokenizer, tokenizer_voila, model_type
|
58 |
|
59 |
def is_audio_output_task(task_type):
|
|
|
91 |
yield all_tokens[:,i]
|
92 |
|
93 |
if model_type == "autonomous":
|
94 |
+
input_generator = get_input_generator(torch.as_tensor(streaming_user_input_audio_tokens).to('cuda'))
|
95 |
+
input_ids = [torch.as_tensor([input]).transpose(1,2).to('cuda') for input in input_ids] # transpose to [bs, seq, num_codebooks]
|
96 |
input_ids = torch.cat(input_ids, dim=2) # concat to [bs, seq, num_codebooks*2]
|
97 |
else:
|
98 |
+
input_ids = torch.as_tensor([input_ids]).transpose(1,2).to('cuda') # transpose to [bs, seq, num_codebooks]
|
99 |
gen_params = {
|
100 |
"input_ids": input_ids,
|
101 |
"ref_embs": ref_embs,
|
|
|
111 |
"audio_top_k": 50,
|
112 |
}
|
113 |
if model_type == "audio":
|
114 |
+
audio_datas = torch.tensor([audio_datas], dtype=torch.bfloat16).to('cuda')
|
115 |
+
audio_data_masks = torch.tensor([audio_data_masks]).to('cuda')
|
116 |
gen_params["audio_datas"] = audio_datas
|
117 |
gen_params["audio_data_masks"] = audio_data_masks
|
118 |
elif model_type == "autonomous":
|
|
|
142 |
'text': tokenizer.decode(text_outputs),
|
143 |
}
|
144 |
if is_audio_output_task(task_type):
|
145 |
+
audio_values = tokenizer_voila.decode(torch.tensor(audio_outputs).to('cuda'))
|
146 |
out['audio'] = (audio_values.detach().cpu().numpy(), 16000)
|
147 |
return out
|
148 |
|
|
|
186 |
# step2: encode ref
|
187 |
ref_embs, ref_embs_mask = None, None
|
188 |
if is_audio_output_task(args.task_type):
|
189 |
+
spkr_model = SpeakerEmbedding(device="cpu")
|
190 |
+
spkr_model.model.to("cuda")
|
191 |
+
spkr_model.device = "cuda"
|
192 |
wav, sr = torchaudio.load(args.ref_audio)
|
193 |
ref_embs = spkr_model(wav, sr)
|
194 |
+
ref_embs_mask = torch.tensor([1]).to('cuda')
|
195 |
|
196 |
out = eval_model(model, tokenizer, tokenizer_voila, model_type, args.task_type, history, ref_embs, ref_embs_mask)
|
197 |
print(f"Output str: {out['text']}")
|