repeat extend the prom to fill the initial tokens for nar-len (it somewhat works, the model just needs to train more)

vall_e/inference.py

@@ -356,6 +356,7 @@ class TTS():
 					)
 				elif model_len is not None:
 					len_list = model_len( text_list=[phns], proms_list=[prom], max_steps=10, disable_tqdm=not tqdm ) # don't need more than that
+					len_list = [ min(l, max_ar_steps) for l in len_list ]
 					resps_list = model_nar( text_list=[phns], proms_list=[prom], len_list=len_list,
 						max_levels=max_nar_levels,
 						sampling_temperature=nar_temp,

vall_e/models/nar.py

@@ -18,7 +18,8 @@ from einops import rearrange
 from torch import Tensor
 from tqdm import trange
 
-from ..emb.qnt import trim
+from ..emb.qnt import trim, repeat_extend_audio
+
 import logging
 
 def clamp(n, lo, hi):
@@ -216,7 +217,8 @@ class NAR(Base):
 			
 			# fill with mock tokens
 			# to-do: repeat with the input prompt, as per training
-			prev_list = [ torch.tensor([ self.stop_token for _ in range(resp_len) ], device=device, dtype=torch.int16) for resp_len in len_list ]
+			#prev_list = [ torch.tensor([ self.stop_token for _ in range(resp_len) ], device=device, dtype=torch.int16) for resp_len in len_list ]
+			prev_list = [ repeat_extend_audio( prom, resp_len ) for resp_len, prom in zip(len_list, proms_list) ]
 
 			# to-do: figure out why this fails when I copy some things from ar_nar
 			for n in trange( max_levels, desc="NAR", disable=disable_tqdm ):
@@ -251,7 +253,8 @@ class NAR(Base):
 					prev_list=prev_list,
 					quant_levels=quant_levels,
 
-					temperature=sampling_temperature,
+					#temperature=sampling_temperature,
+					temperature=1.0 if n == 0 else sampling_temperature,
 					min_temperature=sampling_min_temperature,
 					top_p=sampling_top_p,
 					top_k=sampling_top_k,