UGH

vall_e/models/ar.py

@@ -117,10 +117,11 @@ class AR(Base):
 		device = text_list[0].device
 		batch_size = len(text_list)
 
-		resps_list: list[Tensor] = [ torch.zeros(0, device=device).to(torch.int16) for _ in text_list ]
+		sequence_list = [ torch.zeros(0, device=device).to(torch.int16) for _ in text_list ]
 		stopped = torch.zeros(batch_size, device=device).bool()
 
 		state = {} if cfg.inference.recurrent_forward else None
+
 		sampling_beam_width_use_logs = True
 		scores = [ 1.0 ] * sampling_beam_width
 
@@ -129,12 +130,12 @@ class AR(Base):
 
 		# get next in sequence
 		for n in trange(max_steps // max(1, self.recurrent_chunk_size)):
-
+			resps_list = self._unsqueeze_list(sequence_list)
 			logits = super().forward(
 				text_list=text_list,
 				proms_list=proms_list,
-				resps_list=self._unsqueeze_list(resps_list),
-				quant_levels=None,
+				resps_list=resps_list,
+				
 				state=state
 			)
 
@@ -173,8 +174,7 @@ class AR(Base):
 			for i, ri in enumerate(r):
 				if self.stop_token in ri:
 					stopped[i] = True
-
-				resps_list[i] = torch.cat([resps_list[i], ri])
+				sequence_list[i] = torch.cat([sequence_list[i], ri])
 
 			# stop token found
 			stopped |= r == self.stop_token
@@ -191,10 +191,9 @@ class AR(Base):
 
 			sequence_list = [sequence_list[best_idx]]
 
-		res = [self._prune(r) for r in resps_list]
 		if self.interleave:
-			res = [self._deinterleave(r) for r in res]
-		return res
+			sequence_list = [self._deinterleave(r) for r in sequence_list]
+		return [self._prune(r) for r in sequence_list]
 
 
 def example_usage():

vall_e/models/base.py

@@ -150,7 +150,7 @@ class MultiEmbedding(nn.Module):
 	"""
 
 	def __init__(self, max_n_levels, n_tokens, token_dim, monolithic=False):
-		super().__init__(max_n_levels, token_dim)
+		super().__init__()
 		self.monolithic = monolithic
 		self.max_n_levels = max_n_levels
 		self.n_tokens = n_tokens

vall_e/models/nar.py

@@ -69,6 +69,7 @@ class NAR(Base):
 		sampling_repetition_penalty: float = 1.0,
 		sampling_repetition_penalty_decay: float = 0.0,
 		sampling_length_penalty: float = 0.0,
+		sampling_beam_width: int = 0,
 	):
 		"""
 		Args:
@@ -129,9 +130,9 @@ class NAR(Base):
 
 				resps_list = super().sample(
 					logits=logits,
-					resps_list=resps_list,
+					resps_list=prev_list,
 					quant_levels=quant_levels,
-					
+
 					temperature=sampling_temperature,
 					top_p=sampling_top_p,
 					top_k=sampling_top_k,