some cleanup, fixed the wrapper attention to explicitly use other sdpa backends

README.md

@@ -16,9 +16,6 @@ Besides a working PyTorch environment, the only hard requirement is [`espeak-ng`
 
 ## Install
 
-> [!NOTE]
-> There seems to be some form of regression in fancier attention mechanisms in some environments where you might need to explicitly set `attention` to `flash_attention_2` or `sdpa`.
-
 Simply run `pip install git+https://git.ecker.tech/mrq/vall-e` or `pip install git+https://github.com/e-c-k-e-r/vall-e`.
 
 I've tested this repo under Python versions `3.10.9`, `3.11.3`, and `3.12.3`.
@@ -30,7 +27,7 @@ I've tested this repo under Python versions `3.10.9`, `3.11.3`, and `3.12.3`.
 
 My pre-trained weights can be acquired from [here](https://huggingface.co/ecker/vall-e).
 
-A script to setup a proper environment and download the weights can be invoked with `./scripts/setup.sh`
+A script to setup a proper environment and download the weights can be invoked with `./scripts/setup.sh`. This will automatically create a `venv`, and download the weights and config file to the right place.
 
 ## Train
 

scripts/setup-training.sh

@@ -1,7 +0,0 @@
-#!/bin/bash
-`dirname $0`/setup.sh
-
-wget -P ./training/valle/ "https://huggingface.co/ecker/vall-e/resolve/main/data.tar.gz"
-wget -P ./training/valle/ "https://huggingface.co/ecker/vall-e/resolve/main/.cache.tar.gz"
-tar -xzf ./training/valle/data.tar.gz -C "./training/valle/" data.h5
-tar -xzf ./training/valle/.cache.tar.gz -C "./training/valle/"

vall_e/config.py

@@ -223,6 +223,10 @@ class ModelExperimentalSettings:
 	causal_size: int = 1 # experimental setting to see if I can just do parallel decoding in chunks instead of one-at-a-time without resorting to exotic solutions
 	# VALL-E 2's approach of "combining token embeddings to group them" sounds terribad for a shared AR/NAR model
 	# however, introducing partial parallel decoding for the AR maybe maybe MAYBE might help try and unify the AR/NAR tasks better, MAYBE
+	# it just seems like a bitch to try and train something worthwhile with it, since there's crackles every other token
+
+	p_len_train: float = 0.05 # odds of injecting a "len" task within the model for NAR-len
+	# to-to: just incorporate this as a task instead
 
 # I really need to clean this up
 @dataclass()

vall_e/models/ar_nar.py

@@ -26,73 +26,6 @@ def clamp(n, lo, hi):
 	return max(lo, min(n, hi))
 
 class AR_NAR(Base):
-	@property
-	def capabilities(self) -> list[str]:
-		if hasattr(self, "config") and self.config:
-			return self.config.capabilities
-		return cfg.model.capabilities
-
-	@property
-	def causal(self):
-		return "ar" in self.capabilities
-
-	@property
-	def n_resp_levels(self) -> int:
-		if hasattr(self, "config") and self.config:
-			return self.config.resp_levels
-		return cfg.model.resp_levels
-
-	@property
-	def n_max_levels(self) -> int:
-		if hasattr(self, "config") and self.config:
-			return self.config.max_levels
-		return cfg.model.max_levels
-
-	@property
-	def n_tasks(self) -> int:
-		if hasattr(self, "config") and self.config:
-			return self.config.tasks
-		return cfg.model.tasks
-	
-	@property
-	def n_langs(self) -> int:
-		if hasattr(self, "config") and self.config:
-			return self.config.langs
-		return cfg.model.langs
-
-	@property
-	def n_tones(self) -> int:
-		if hasattr(self, "config") and self.config:
-			return self.config.tones
-		return cfg.model.tones
-
-	@property
-	def causal_size(self) -> int:
-		# 1 for the stop token
-		# governs how much to shift the logits by
-		# could *technically* make it work to where it can also predict *ALL* RVQ levels in one step, but experimental.py is the better way to go about it
-		if hasattr(self, "config") and self.config:
-			return self.config.experimental.causal_size
-		return cfg.model.experimental.causal_size
-
-	@property
-	def version(self) -> int:
-		if hasattr(self, "config") and self.config:
-			return self.config.version
-		return cfg.model.version
-
-	def _prune(self, l: Tensor, stop = None):
-		if stop is None:
-			stop = self.stop_token
-		indices = (l == stop).nonzero()
-		if len(indices) == 0:
-			return l
-		return l[: indices.min().item()]
-
-	@staticmethod
-	def _unsqueeze_list(x_list, axis=-1):
-		return [x.unsqueeze(dim=axis) for x in x_list]
-
 	def forward(
 		self,
 		text_list: list[Tensor],
@@ -299,7 +232,7 @@ class AR_NAR(Base):
 
 		# get next in sequence
 		for n in trange(max_steps // max(1, self.causal_size), desc="AR", disable=disable_tqdm):
-			resps_list = self._unsqueeze_list(sequence_list)
+			resps_list = [x.unsqueeze(dim=-1) for x in sequence_list]
 
 			inputs = self.inputs(
 				text_list=text_list,

vall_e/models/arch/__init__.py

@@ -30,7 +30,7 @@ except Exception as e:
 	pass
 
 try:
-	from .llama import LlamaModel, LlamaConfig, AVAILABLE_ATTENTIONS, LlamaAttention, LlamaAttention_Base, LlamaForCausalLM
+	from .llama import LlamaModel, LlamaConfig, AVAILABLE_ATTENTIONS, LlamaAttention, LlamaAttention_Adapted, LlamaForCausalLM
 	AVAILABLE_ARCHES.append("llama")
 except Exception as e:
 	ERROR_ARCHES["llama"] = e
@@ -61,11 +61,4 @@ try:
 	from .mamba_vasqu import Mamba2Model_HF, Mamba2Config_HF
 	AVAILABLE_ARCHES.append("mamba2-hf")
 except Exception as e:
-	ERROR_ARCHES["mamba2-hf"] = e
+	ERROR_ARCHES["mamba2-hf"] = e
-
-# desu should remove, perf was very lacking in comparison to regular bitnet
-try:
-	from .mmfreelm import *
-	AVAILABLE_ARCHES.append("mmfreelm")
-except Exception as e:
-	ERROR_ARCHES["mmfreelm"] = e

vall_e/models/arch/llama.py

@@ -7,9 +7,18 @@ from torch import Tensor, nn
 from transformers.cache_utils import Cache
 
 from transformers import LlamaModel, LlamaConfig, LlamaForCausalLM
-from transformers.models.llama.modeling_llama import LlamaAttention as LlamaAttention_Base, apply_rotary_pos_emb
+from transformers.models.llama.modeling_llama import LlamaAttention, apply_rotary_pos_emb, repeat_kv
 
-AVAILABLE_ATTENTIONS = ["mem_efficient", "math"]
+AVAILABLE_ATTENTIONS = ["sdpa"]
+
+if torch.backends.cuda.flash_sdp_enabled():
+	AVAILABLE_ATTENTIONS.append("flash")	
+
+if torch.backends.cuda.mem_efficient_sdp_enabled():
+	AVAILABLE_ATTENTIONS.append("mem_efficient")	
+
+if torch.backends.cuda.math_sdp_enabled():
+	AVAILABLE_ATTENTIONS.append("math")	
 
 try:
 	from xformers.ops import LowerTriangularMask
@@ -23,11 +32,11 @@ try:
 	from transformers.utils import is_flash_attn_2_available
 
 	if is_flash_attn_2_available():
-		AVAILABLE_ATTENTIONS.append("flash")
+		AVAILABLE_ATTENTIONS.append("flash_attention_2")
 except Exception as e:
 	print("Error while querying for `flash_attn_2` support", e)
 
-class LlamaAttention(LlamaAttention_Base):
+class LlamaAttention_Adapted(LlamaAttention):
 	def __init__(self, *args, **kwargs):
 		if 'mode' in kwargs:
 			self.mode = kwargs['mode']
@@ -35,8 +44,101 @@ class LlamaAttention(LlamaAttention_Base):
 		else:
 			self.mode = "math"
 
+		if self.mode == "math":
+			self.mode = torch.nn.attention.SDPBackend.MATH
+		elif self.mode == "mem_efficient":
+			self.mode = torch.nn.attention.SDPBackend.EFFICIENT_ATTENTION
+		elif self.mode == "flash":
+			self.mode = torch.nn.attention.SDPBackend.FLASH_ATTENTION
+		elif self.mode == "cudnn":
+			self.mode = torch.nn.attention.SDPBackend.CUDNN_ATTENTION
+
 		super().__init__(*args, **kwargs)
 
+	 # Adapted from LlamaAttention.forward
+	def forward(
+		self,
+		hidden_states: torch.Tensor,
+		attention_mask: Optional[torch.Tensor] = None,
+		position_ids: Optional[torch.LongTensor] = None,
+		past_key_value: Optional[Cache] = None,
+		output_attentions: bool = False,
+		use_cache: bool = False,
+		cache_position: Optional[torch.LongTensor] = None,
+		position_embeddings: Optional[Tuple[torch.Tensor, torch.Tensor]] = None,  # will become mandatory in v4.45
+		**kwargs,
+	) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor]]]:
+		if output_attentions:
+			# TODO: Improve this warning with e.g. `model.config.attn_implementation = "manual"` once this is implemented.
+			return super().forward(
+				hidden_states=hidden_states,
+				attention_mask=attention_mask,
+				position_ids=position_ids,
+				past_key_value=past_key_value,
+				output_attentions=output_attentions,
+				use_cache=use_cache,
+				cache_position=cache_position,
+				position_embeddings=position_embeddings,
+			)
+
+		bsz, q_len, _ = hidden_states.size()
+
+		query_states = self.q_proj(hidden_states)
+		key_states = self.k_proj(hidden_states)
+		value_states = self.v_proj(hidden_states)
+
+		query_states = query_states.view(bsz, q_len, self.num_heads, self.head_dim).transpose(1, 2)
+		key_states = key_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2)
+		value_states = value_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2)
+
+		if position_embeddings is None:
+			cos, sin = self.rotary_emb(value_states, position_ids)
+		else:
+			cos, sin = position_embeddings
+		query_states, key_states = apply_rotary_pos_emb(query_states, key_states, cos, sin)
+
+		if past_key_value is not None:
+			# sin and cos are specific to RoPE models; cache_position needed for the static cache
+			cache_kwargs = {"sin": sin, "cos": cos, "cache_position": cache_position}
+			key_states, value_states = past_key_value.update(key_states, value_states, self.layer_idx, cache_kwargs)
+
+		key_states = repeat_kv(key_states, self.num_key_value_groups)
+		value_states = repeat_kv(value_states, self.num_key_value_groups)
+
+		causal_mask = attention_mask
+		if attention_mask is not None:
+			causal_mask = causal_mask[:, :, :, : key_states.shape[-2]]
+
+		# SDPA with memory-efficient backend is currently (torch==2.1.2) bugged with non-contiguous inputs with custom attn_mask,
+		# Reference: https://github.com/pytorch/pytorch/issues/112577.
+		if query_states.device.type == "cuda" and causal_mask is not None:
+			query_states = query_states.contiguous()
+			key_states = key_states.contiguous()
+			value_states = value_states.contiguous()
+
+		# We dispatch to SDPA's Flash Attention or Efficient kernels via this `is_causal` if statement instead of an inline conditional assignment
+		# in SDPA to support both torch.compile's dynamic shapes and full graph options. An inline conditional prevents dynamic shapes from compiling.
+		is_causal = True if causal_mask is None and q_len > 1 else False
+
+		#with torch.backends.cuda.sdp_kernel(enable_flash=self.mode == "flash", enable_math=self.mode == "math", enable_mem_efficient=self.mode == "mem_efficient"):
+		with torch.nn.attention.sdpa_kernel(self.mode):
+			attn_output = torch.nn.functional.scaled_dot_product_attention(
+				query_states,
+				key_states,
+				value_states,
+				attn_mask=causal_mask,
+				dropout_p=self.attention_dropout if self.training else 0.0,
+				is_causal=is_causal,
+			)
+
+		attn_output = attn_output.transpose(1, 2).contiguous()
+		attn_output = attn_output.view(bsz, q_len, -1)
+
+		attn_output = self.o_proj(attn_output)
+
+		return attn_output, None, past_key_value
+
+	"""
 	def forward(
 		self,
 		hidden_states: torch.Tensor,
@@ -88,4 +190,5 @@ class LlamaAttention(LlamaAttention_Base):
 		attn_output = attn_output.reshape(bsz, q_len, self.hidden_size)
 		attn_output = self.o_proj(attn_output)
 
-		return attn_output, attn_weights, past_key_value
+		return attn_output, attn_weights, past_key_value
+	"""

vall_e/models/arch/mmfreelm.py

@@ -1,6 +0,0 @@
-# https://github.com/ridgerchu/matmulfreellm
-
-import torch
-import torch.nn.functional as F
-
-from mmfreelm.models import HGRNBitConfig, HGRNBitModel

vall_e/models/base.py

@@ -297,60 +297,22 @@ class Metrics(nn.Module):
 		)
 
 class Base(nn.Module):
-	# to-do: clean up this property mess
-	@property
-	def causal(self) -> bool:
-		raise NotImplementedError
-
-	@property
-	def n_resp_levels(self) -> int:
-		raise NotImplementedError
-
-	@property
-	def n_max_levels(self) -> int:
-		raise NotImplementedError
-	
-	@property
-	def n_langs(self) -> int:
-		raise NotImplementedError
-	
-	@property
-	def n_tasks(self) -> int:
-		raise NotImplementedError
-
-	@property
-	def n_tones(self) -> int:
-		raise NotImplementedError
-
-	@property
-	def causal_size(self) -> int:
-		raise NotImplementedError
-
-	@property
-	def version(self) -> int:
-		return 2
-
-	@property
-	def capabilities(self) -> list[str]:
-		raise NotImplementedError
-
-	@property
-	def stop_token(self):
-		if "len" in self.capabilities:
-			return 0
-		if not self.causal:
-			raise ValueError("Not using stop token!")
-		return self.n_audio_tokens
-
-	@property
-	def ignore_index(self):
-		return -100
-
 	def loss_factor(self, k):
 		if self.config is None:
 			return 1.0
 		return self.config.loss_factors[k] if k in self.config.loss_factors else 1.0
 
+	def _prune(self, l: Tensor, stop = None):
+		if stop is None:
+			stop = self.stop_token
+
+		indices = (l == stop).nonzero()
+
+		if len(indices) == 0:
+			return l
+
+		return l[: indices.min().item()]
+
 	# these probably need to live in an interleaved model, as pattern-ing is targeted for a sole AR model
 	"""
 	def codes_to_pattern(self, codes):
@@ -404,7 +366,6 @@ class Base(nn.Module):
 		super().__init__()
 		self.training = training
 		self.config = config
-		self.gradient_checkpointing = self.config.gradient_checkpointing if self.config is not None else True
 
 		self.n_text_tokens = n_text_tokens
 		self.n_audio_tokens = n_audio_tokens
@@ -416,19 +377,35 @@ class Base(nn.Module):
 		
 		self.l_padding = l_padding
 
-		arch_type = self.config.arch_type if self.config is not None else "llama"
+		self.ignore_index = -100
 
-		self.arch_type = arch_type
+		self.n_resp_levels = self.config.resp_levels if self.config else n_resp_levels
+		self.n_max_levels = self.config.max_levels if self.config else n_resp_levels
+		self.capabilities = self.config.capabilities if self.config else ["ar", "nar"]
+		self.gradient_checkpointing = self.config.gradient_checkpointing if self.config is not None else True
+
+		self.stop_token = self.n_audio_tokens # id 1024
+		self.causal = "ar" in self.capabilities or "len" in self.capabilities
+		self.version = self.config.version if self.config is not None else 5
+		self.causal_size = self.config.experimental.causal_size if self.config is not None else (1 if "ar" in self.capabilities else 0)
+
+		self.arch_type = self.config.arch_type if self.config is not None else "llama"
 
 		# check if requested arch is unavailable
 		if self.arch_type in ERROR_ARCHES:
 			raise ERROR_ARCHES[self.arch_type]
+		
+		attention_backend = self.config.attention if self.config is not None else "auto"
 		audio_embedding_sums = self.config.experimental.audio_embedding_sums if self.config is not None else False
 		split_classifiers = self.config.experimental.split_classifiers if self.config is not None else False
 		tie_classifier_to_embedding = self.config.experimental.tie_classifier_to_embedding if self.config is not None else False
 		audio_embedding_mode = self.config.experimental.audio_embedding_mode if self.config is not None else ""
 		unified_position_ids = self.config.experimental.unified_position_ids if self.config is not None else True
 
+		n_tasks = self.config.tasks if self.config is not None else 8
+		n_langs = self.config.langs if self.config is not None else 2
+		n_tones = self.config.tones if self.config is not None else 1
+
 		if "len" not in self.capabilities:
 			# +1 to include the stop token
 			n_resp_tokens = n_audio_tokens + ( 1 if self.causal_size > 0 else 0 )
@@ -457,7 +434,7 @@ class Base(nn.Module):
 		self.dropout_token = nn.Parameter(torch.zeros(d_model)) # zeros sounds nicer than randn for a special value
 
 		if self.version == 1: # legacy
-			n_audio_tokens += (self.n_tasks - 1) # old models have the task tokens in the prom
+			n_audio_tokens += (n_tasks - 1) # old models have the task tokens in the prom
 			self.proms_emb = MultiEmbedding(self.n_resp_levels, n_audio_tokens, d_model)
 			self.resps_emb = MultiEmbedding(self.n_resp_levels, n_resp_tokens, d_model, monolithic=self.monolithic)
 		elif self.version < 5:
@@ -485,11 +462,11 @@ class Base(nn.Module):
 
 		# useless since I actually removed using these with the input processing overhaul...
 		if self.version >= 3:
-			self.langs_emb = Embedding(self.n_langs, d_model) if self.n_langs > 0 else None
+			self.langs_emb = Embedding(n_langs, d_model) if n_langs > 0 else None
-			self.tasks_emb = Embedding(self.n_tasks, d_model) if self.n_tasks > 0 else None
+			self.tasks_emb = Embedding(n_tasks, d_model) if n_tasks > 0 else None
 		# never actually got added... I kept forgetting to classify all my audio for speaker's tone
 		if self.version >= 4:
-			self.tones_emb = Embedding(self.n_tones, d_model) if self.n_tones > 0 else None
+			self.tones_emb = Embedding(n_tones, d_model) if n_tones > 0 else None
 
 		# mamba requires this if a model does both AR and NAR tasks
 		# this *might* help for AR and NAR tasks since we explicitly specify the current RVQ level for a sequence, rather than having it "encoded" in the embeddings
@@ -501,31 +478,28 @@ class Base(nn.Module):
 			self.len_emb = Embedding(11, d_model) if "len" in self.capabilities else None
 
 		# there seems to have been a regression where anything touching the wrapped LlamaAttention class breaks
-		"""
+
-		# ick, there has to be a better way
+		if attention_backend == "auto":
-		if self.config.attention == "auto":
+			if "flash_attention_2" in AVAILABLE_ATTENTIONS:
-			if "flash" in AVAILABLE_ATTENTIONS:
+				attention_backend = "flash_attention_2"
-				self.config.attention = "flash"
+			elif "flash" in AVAILABLE_ATTENTIONS:
-			elif "xformers" in AVAILABLE_ATTENTIONS:
+				attention_backend = "flash"
-				self.config.attention = "xformers"
+			elif "mem_efficient" in AVAILABLE_ATTENTIONS:
+				attention_backend = "mem_efficient"
+			elif "math" in AVAILABLE_ATTENTIONS:
+				attention_backend = "math"
 			else:
-				self.config.attention = "sdpa"
+				attention_backend = "sdpa"
 
-		hf_attention = self.config.attention if self.config is not None else None
+		if attention_backend == "xformers":
-
+			attention_backend = "mem_efficient"
-		if self.config.attention in ["xformers", "mem_efficient", "math", "flash"]:
-			hf_attention = None
-			if self.config.attention not in AVAILABLE_ATTENTIONS:
-				raise ValueError(f"Requesting attention `{self.config.attention}` but is not available. Currently available: {AVAILABLE_ATTENTIONS}")
-		"""
-
-		if self.config.attention == "auto":
-			if "flash" in AVAILABLE_ATTENTIONS:
-				self.config.attention = "flash_attention_2"
-			else:
-				self.config.attention = "sdpa"
 		
-		hf_attention = self.config.attention if self.config is not None else None
+		hf_attention = attention_backend
+
+		if attention_backend in ["xformers", "mem_efficient", "math", "flash", "cudnn"]:
+			hf_attention = None
+			if attention_backend not in AVAILABLE_ATTENTIONS:
+				raise ValueError(f"Requesting attention `{attention_backend}` but is not available. Currently available: {AVAILABLE_ATTENTIONS}")
 
 		if self.arch_type == "transformer":
 			self.sin_emb = SinusoidalEmbedding(d_model)
@@ -654,18 +628,6 @@ class Base(nn.Module):
 				))
 
 			self.model = RetNetDecoder(RetNetConfig(**kwargs))
-
-			# do some funny stuff for LoRA training
-			"""
-			if self.gradient_checkpointing:
-				def make_inputs_require_grads(module, input, output):
-					for i, t in enumerate(input):
-						if not isinstance(t, torch.Tensor):
-							continue
-						t.requires_grad_(True)
-
-				self.model.register_forward_hook(make_inputs_require_grads)
-			"""
 		elif self.arch_type == "retnet-hf":
 			kwargs = dict(
 				vocab_size=n_resp_tokens,
@@ -757,10 +719,8 @@ class Base(nn.Module):
 		if hasattr( self.model, "embeddings" ):
 			del self.model.embeddings
 
-		"""
+		if attention_backend in ["mem_efficient", "math", "flash", "cudnn", "auto"]:
-		if self.config.attention in ["xformers", "auto", "mem_efficient", "math", "flash"]:
+			self.model = ml.replace_attention( self.model, klass=LlamaAttention_Adapted, target=LlamaAttention, mode=attention_backend )
-			self.model = ml.replace_attention( self.model, klass=LlamaAttention, target=LlamaAttention_Base, mode=self.config.attention )
-		"""
 
 		if not split_classifiers:
 			self.classifier = nn.Linear(d_model, n_resp_tokens)

vall_e/models/nar.py

@@ -21,71 +21,6 @@ from tqdm import trange
 from ..emb.qnt import trim
 
 class NAR(Base):
-	@property
-	def capabilities(self) -> list[str]:
-		if hasattr(self, "config") and self.config:
-			return self.config.capabilities
-		return cfg.model.capabilities
-
-	@property
-	def causal(self):
-		return "len" in self.capabilities
-
-	@property
-	def n_resp_levels(self) -> int:
-		if hasattr(self, "config") and self.config:
-			return self.config.resp_levels
-		return cfg.model.resp_levels
-
-	@property
-	def n_max_levels(self) -> int:
-		if hasattr(self, "config") and self.config:
-			return self.config.max_levels
-		return cfg.model.max_levels
-
-	@property
-	def n_tasks(self) -> int:
-		if hasattr(self, "config") and self.config:
-			return self.config.tasks
-		return cfg.model.tasks
-
-	@property
-	def n_langs(self) -> int:
-		if hasattr(self, "config") and self.config:
-			return self.config.langs
-		return cfg.model.langs
-
-	@property
-	def n_tones(self) -> int:
-		if hasattr(self, "config") and self.config:
-			return self.config.tones
-		return cfg.model.tones
-
-	@property
-	def causal_size(self) -> int:
-		# 1 for the stop token
-		# governs how much to shift the logits by
-		# could *technically* make it work to where it can also predict *ALL* RVQ levels in one step, but experimental.py is the better way to go about it
-		return 1 # if self.causal else 0
-
-	@property
-	def version(self) -> int:
-		if hasattr(self, "config") and self.config:
-			return self.config.version
-		return cfg.model.version
-
-	def _prune(self, l: Tensor, stop = None):
-		if stop is None:
-			stop = self.stop_token
-		indices = (l == stop).nonzero()
-		if len(indices) == 0:
-			return l
-		return l[: indices.min().item()]
-
-	@staticmethod
-	def _unsqueeze_list(x_list, axis=-1):
-		return [x.unsqueeze(dim=axis) for x in x_list]
-
 	def forward(
 		self,
 		text_list: list[Tensor],
@@ -121,7 +56,7 @@ class NAR(Base):
 
 		# is training
 		if resps_list is not None:
-			p_len_task = 0.25
+			p_len_task = self.config.experimental.p_len_train if self.config is not None else 0.05
 
 			n_levels_set = {r.shape[-1] for r in resps_list}
 			n_levels = next(iter(n_levels_set))