import functools
import torch
import torch.nn as nn
import torch.nn.functional as F
from transformers import GPT2Config, GPT2Model, GPT2PreTrainedModel
from transformers.modeling_outputs import CausalLMOutputWithCrossAttentions
def null_position_embeddings(range, dim):
return torch.zeros((range.shape[0], range.shape[1], dim), device=range.device)
class GPT2InferenceModel(GPT2PreTrainedModel):
"""Override GPT2LMHeadModel to allow for prefix conditioning."""
def __init__(self, config, gpt, pos_emb, embeddings, norm, linear, kv_cache):
super().__init__(config)
self.transformer = gpt
self.pos_embedding = pos_emb
self.embeddings = embeddings
self.final_norm = norm
self.lm_head = nn.Sequential(norm, linear)
self.kv_cache = kv_cache
def store_prefix_emb(self, prefix_emb):
self.cached_prefix_emb = prefix_emb
def prepare_inputs_for_generation(self, input_ids, past_key_values=None, **kwargs):
token_type_ids = kwargs.get("token_type_ids", None) # usually None
if not self.kv_cache:
past_key_values = None
# only last token for inputs_ids if past is defined in kwargs
if past_key_values is not None:
input_ids = input_ids[:, -1].unsqueeze(-1)
if token_type_ids is not None:
token_type_ids = token_type_ids[:, -1].unsqueeze(-1)
attention_mask = kwargs.get("attention_mask", None)
position_ids = kwargs.get("position_ids", None)
if attention_mask is not None and position_ids is None:
# create position_ids on the fly for batch generation
position_ids = attention_mask.long().cumsum(-1) - 1
position_ids.masked_fill_(attention_mask == 0, 1)
if past_key_values is not None:
position_ids = position_ids[:, -1].unsqueeze(-1)
else:
position_ids = None
return {
"input_ids": input_ids,
"past_key_values": past_key_values,
"use_cache": kwargs.get("use_cache"),
"position_ids": position_ids,
"attention_mask": attention_mask,
"token_type_ids": token_type_ids,
}
def forward(
self,
input_ids=None,
past_key_values=None,
attention_mask=None,
token_type_ids=None,
position_ids=None,
head_mask=None,
inputs_embeds=None,
encoder_hidden_states=None,
encoder_attention_mask=None,
labels=None,
use_cache=None,
output_attentions=None,
output_hidden_states=None,
return_dict=None,
):
assert self.cached_prefix_emb is not None
assert inputs_embeds is None # Not supported by this inference model.
assert labels is None # Training not supported by this inference model.
return_dict = return_dict if return_dict is not None else self.config.use_return_dict
# assert len(past_key_values) + len(input_ids) == attention_mask.shape[1]
# Create embedding
prefix_len = self.cached_prefix_emb.shape[1]
if input_ids.shape[1] != 1:
gen_inputs = input_ids[:, prefix_len:]
gen_emb = self.embeddings(gen_inputs)
gen_emb = gen_emb + self.pos_embedding(gen_emb)
if self.cached_prefix_emb.shape[0] != gen_emb.shape[0]:
prefix_emb = self.cached_prefix_emb.repeat_interleave(
gen_emb.shape[0] // self.cached_prefix_emb.shape[0], 0
)
else:
prefix_emb = self.cached_prefix_emb.to(gen_emb.dtype)
emb = torch.cat([prefix_emb, gen_emb], dim=1)
else:
emb = self.embeddings(input_ids)
emb = emb + self.pos_embedding.get_fixed_embedding(
attention_mask.shape[1] - (prefix_len + 1), attention_mask.device
)
transformer_outputs = self.transformer(
inputs_embeds=emb,
past_key_values=past_key_values,
attention_mask=attention_mask,
token_type_ids=token_type_ids,
position_ids=position_ids,
head_mask=head_mask,
encoder_hidden_states=encoder_hidden_states,
encoder_attention_mask=encoder_attention_mask,
use_cache=use_cache,
output_attentions=output_attentions,
output_hidden_states=output_hidden_states,
return_dict=return_dict,
)
hidden_states = transformer_outputs[0]
lm_logits = self.lm_head(hidden_states)
if not return_dict:
return (lm_logits,) + transformer_outputs[1:]
return CausalLMOutputWithCrossAttentions(
loss=None,
logits=lm_logits,
past_key_values=transformer_outputs.past_key_values,
hidden_states=transformer_outputs.hidden_states,
attentions=transformer_outputs.attentions,
cross_attentions=transformer_outputs.cross_attentions,
)
@staticmethod
def _reorder_cache(past, beam_idx):
"""
This function is used to re-order the :obj:`past_key_values` cache if
:meth:`~transformers.PreTrainedModel.beam_search` or :meth:`~transformers.PreTrainedModel.beam_sample` is
called. This is required to match :obj:`past_key_values` with the correct beam_idx at every generation step.
"""
return tuple(
tuple(past_state.index_select(0, beam_idx.to(past_state.device)) for past_state in layer_past)
for layer_past in past
)
class LearnedPositionEmbeddings(nn.Module):
def __init__(self, seq_len, model_channels, init_std=0.02, relative=False):
super().__init__()
self.emb = nn.Embedding(seq_len, model_channels)
nn.init.normal_(self.emb.weight, mean=0.0, std=init_std)
self.relative = relative
def forward(self, x):
seq_len = x.shape[1]
if self.relative:
start = torch.randint(seq_len, (1,), device=x.device).item()
positions = torch.arange(start, start + seq_len, device=x.device)
else:
positions = torch.arange(seq_len, device=x.device)
return self.emb(positions)
def get_fixed_embedding(self, ind, dev):
return self.emb(torch.tensor([ind], device=dev)).unsqueeze(0)
def init_gpt(layers, model_channels, heads, max_mel_seq_len, max_text_seq_len, max_prompt_len, checkpointing):
"""
Initializes a GPT-2 model and its position embeddings for a text-to-speech system.
Args:
layers (int): Number of layers in the GPT-2 model.
model_channels (int): Dimension of the GPT-2 model.
heads (int): Number of heads in the GPT-2 model.
max_mel_seq_len (int): Maximum sequence length for the mel spectrogram.
max_text_seq_len (int): Maximum sequence length for the text.
max_prompt_len (int): Maximum length of the prompt.
checkpointing (bool): Whether to use gradient checkpointing.
Returns:
gpt (GPT2Model): GPT-2 model.
mel_pos_emb (LearnedPositionEmbeddings): Position embeddings for the mel spectrogram.
text_pos_emb (LearnedPositionEmbeddings): Position embeddings for the text.
"""
gpt_config = GPT2Config(
vocab_size=123,
n_positions=max_mel_seq_len + max_text_seq_len + max_prompt_len,
n_ctx=max_mel_seq_len + max_text_seq_len + max_prompt_len,
n_embd=model_channels,
n_layer=layers,
n_head=heads,
gradient_checkpointing=checkpointing,
use_cache=not checkpointing,
)
gpt = GPT2Model(gpt_config)
del gpt.wpe
del gpt.wte
gpt.wpe = functools.partial(null_position_embeddings, dim=model_channels)
audio_pos_emb = (
LearnedPositionEmbeddings(max_mel_seq_len, model_channels)
if max_mel_seq_len != -1
else functools.partial(null_position_embeddings, dim=model_channels)
)
text_pos_emb = (
LearnedPositionEmbeddings(max_text_seq_len, model_channels)
if max_mel_seq_len != -1
else functools.partial(null_position_embeddings, dim=model_channels)
)
return gpt, audio_pos_emb, text_pos_emb
class XTTSGPTEncoder(nn.Module):
"""XTTS GPT Encoder model implementation.
Args:
start_text_token (int): Index of the start token in the text vocabulary.
stop_text_token (int): Index of the stop token in the text vocabulary.
n_layers (int): Number of layers in the GPT-2 model.
n_model_channels (int): Dimension of the GPT-2 model.
n_heads (int): Number of heads in the GPT-2 model.
max_text_tokens (int): Maximum number of text tokens.
max_audio_tokens (int): Maximum number of audio tokens.
max_prompt_tokens (int): Maximum number of prompt tokens.
audio_len_compression (int): Compression factor for the audio length.
number_text_tokens (int): Number of text tokens.
number_audio_codes (int): Number of audio codes.
start_mel_token (int): Index of the start token in the mel code vocabulary.
stop_mel_token (int): Index of the stop token in the mel code vocabulary.
checkpointing (bool): Whether or not to use gradient checkpointing at training.
"""
_inference_flag = False
def __init__(
self,
start_text_token=261,
stop_text_token=0,
n_layers=8,
n_model_channels=512,
n_heads=8,
max_text_tokens=120,
max_audio_tokens=250,
max_prompt_tokens=70,
audio_len_compression=1024,
number_text_tokens=256,
number_audio_codes=8194,
start_mel_token=8192,
stop_mel_token=8193,
checkpointing=True,
label_smoothing=0.0,
):
super().__init__()
self.label_smoothing = label_smoothing
self.number_text_tokens = number_text_tokens
self.start_text_token = start_text_token
self.stop_text_token = stop_text_token
self.number_audio_codes = number_audio_codes
self.start_mel_token = start_mel_token
self.stop_mel_token = stop_mel_token
self.start_prompt_token = start_mel_token
self.stop_prompt_token = stop_mel_token
self.n_layers = n_layers
self.n_heads = n_heads
self.n_model_channels = n_model_channels
self.max_audio_tokens = -1 if max_audio_tokens == -1 else max_audio_tokens + 2 + self.max_conditioning_inputs
self.max_text_tokens = -1 if max_text_tokens == -1 else max_text_tokens + 2
self.max_prompt_tokens = max_prompt_tokens
self.audio_len_compression = audio_len_compression
# embedding layers
self.text_embedding = nn.Embedding(self.number_text_tokens, n_model_channels)
self.audio_embedding = nn.Embedding(self.number_audio_codes, n_model_channels)
self.prompt_embedding = nn.Embedding(self.number_audio_codes, n_model_channels)
self.prompt_pos_embedding = LearnedPositionEmbeddings(24 * 9, n_model_channels)
# initialize the GPT-2 model
(
self.gpt,
self.audio_pos_embedding,
self.text_pos_embedding,
) = init_gpt(
n_layers,
n_model_channels,
n_heads,
self.max_audio_tokens,
self.max_text_tokens,
self.max_prompt_tokens,
checkpointing,
)
# output layers
self.final_norm = nn.LayerNorm(n_model_channels)
self.text_head = nn.Linear(n_model_channels, self.number_text_tokens)
self.mel_head = nn.Linear(n_model_channels, self.number_audio_codes)
def get_grad_norm_parameter_groups(self):
return {
"conditioning_encoder": list(self.conditioning_encoder.parameters()),
"gpt": list(self.gpt.parameters()),
"heads": list(self.text_head.parameters()) + list(self.mel_head.parameters()),
}
def init_model_for_inference(self, kv_cache=True, use_deepspeed=False, use_deepspeed_f16=False):
self._inference_flag = True
seq_length = self.max_prompt_tokens + self.max_audio_tokens + self.max_text_tokens
gpt_config = GPT2Config(
vocab_size=self.max_audio_tokens,
n_positions=seq_length,
n_ctx=seq_length,
n_embd=self.n_model_channels,
n_layer=self.n_layers,
n_head=self.n_heads,
gradient_checkpointing=False,
use_cache=True,
)
self.inference_model = GPT2InferenceModel(
gpt_config,
self.gpt,
self.audio_pos_embedding,
self.audio_embedding,
self.final_norm,
self.mel_head,
kv_cache=kv_cache,
)
self.gpt.wte = self.audio_embedding
def set_inputs_and_targets(self, input, start_token, stop_token):
inp = F.pad(input, (1, 0), value=start_token)
tar = F.pad(input, (0, 1), value=stop_token)
return inp, tar
def set_audio_tokens_padding(self, audio_tokens, audio_token_lens):
# Set padding areas within MEL (currently it is coded with the MEL code for <zero>).
for b in range(len(audio_token_lens)):
actual_end = audio_token_lens[b]
if actual_end < audio_tokens.shape[-1]:
audio_tokens[b, actual_end:] = self.stop_mel_token
return audio_tokens
def get_logits(
self,
speech_conditioning_inputs,
first_inputs,
first_head,
second_inputs=None,
second_head=None,
prompt=None,
get_attns=False,
return_latent=False,
attn_mask_text=None,
attn_mask_mel=None,
):
if prompt is not None and speech_conditioning_inputs is not None:
offset = speech_conditioning_inputs.shape[1] + prompt.shape[1]
if second_inputs is not None:
emb = torch.cat(
[speech_conditioning_inputs, prompt, first_inputs, second_inputs],
dim=1,
)
else:
emb = torch.cat([speech_conditioning_inputs, prompt, first_inputs], dim=1)
elif speech_conditioning_inputs is not None:
offset = speech_conditioning_inputs.shape[1]
if second_inputs is not None:
emb = torch.cat([speech_conditioning_inputs, first_inputs, second_inputs], dim=1)
else:
emb = torch.cat([speech_conditioning_inputs, first_inputs], dim=1)
elif prompt is not None:
offset = prompt.shape[1]
if second_inputs is not None:
emb = torch.cat([prompt, first_inputs, second_inputs], dim=1)
else:
emb = torch.cat([prompt, first_inputs], dim=1)
# with torch.backends.cuda.sdp_kernel(enable_flash=True, enable_math=False, enable_mem_efficient=False):
attn_mask = None
if attn_mask_text is not None:
attn_mask = torch.cat([attn_mask_text, attn_mask_mel], dim=1)
if prompt is not None:
attn_mask_prompt = torch.ones(prompt.shape[0], offset, dtype=torch.bool, device=emb.device)
attn_mask = torch.cat([attn_mask_prompt, attn_mask], dim=1)
gpt_out = self.gpt(
inputs_embeds=emb,
return_dict=True,
output_attentions=get_attns,
attention_mask=attn_mask,
)
if get_attns:
return gpt_out.attentions
enc = gpt_out.last_hidden_state[:, offset:]
enc = self.final_norm(enc)
if return_latent:
return enc[:, : first_inputs.shape[1]], enc[:, -second_inputs.shape[1] :]
first_logits = enc[:, : first_inputs.shape[1]]
first_logits = first_head(first_logits)
first_logits = first_logits.permute(0, 2, 1)
if second_inputs is not None:
second_logits = enc[:, -second_inputs.shape[1] :]
second_logits = second_head(second_logits)
second_logits = second_logits.permute(0, 2, 1)
return first_logits, second_logits
else:
return first_logits
def get_conditioning(self, speech_conditioning_input):
speech_conditioning_input = (
speech_conditioning_input.unsqueeze(1)
if len(speech_conditioning_input.shape) == 3
else speech_conditioning_input
)
conds = []
for j in range(speech_conditioning_input.shape[1]):
conds.append(self.conditioning_encoder(speech_conditioning_input[:, j]))
conds = torch.stack(conds, dim=1)
conds = conds.mean(dim=1)
return conds
def get_prompts(self, prompt_codes):
prompt = F.pad(prompt_codes, (1, 0), value=self.start_prompt_token)
prompt = F.pad(prompt_codes, (0, 1), value=self.stop_prompt_token)
return prompt
def forward(
self,
text_inputs,
text_lengths,
audio_codes,
wav_lengths,
prompt_codes,
return_attentions=False,
return_latent=False,
):
max_text_len = text_lengths.max()
# Due to the convolution in DVAE, codes do not end with silence at the right place. Rather it predicts some intermediate values
# Like [..., 186, 45, 45, 83] where actually it should end with 186.
# We take last 3 codes to prevent abrupt ending of the audio.
# TODO: This is might need some testing.
mel_lengths = torch.ceil(wav_lengths / self.mel_length_compression).long() + 3
# If len(codes) + 3 is larger than maxiumum allowed length, we truncate the codes.
max_mel_len = mel_lengths.max()
if max_mel_len > audio_codes.shape[-1]:
audio_codes = F.pad(audio_codes, (0, max_mel_len - audio_codes.shape[-1]))
# silence aware lengths, skip the silence tokens at the end of the mel codes.
silence = True
for idx, l in enumerate(mel_lengths):
length = l.item()
while silence:
if audio_codes[idx, length - 1] != 83:
break
length -= 1
mel_lengths[idx] = length
# Lovely assertions
assert (
max_mel_len <= audio_codes.shape[-1]
), f" ❗ max_mel_len ({max_mel_len}) > audio_codes.shape[-1] ({audio_codes.shape[-1]})"
assert (
max_text_len <= text_inputs.shape[-1]
), f" ❗ max_text_len ({max_text_len}) > text_inputs.shape[-1] ({text_inputs.shape[-1]})"
# Append stop token to text inputs
text_inputs = F.pad(text_inputs[:, :max_text_len], (0, 1), value=self.stop_text_token)
# Append silence token to mel codes
audio_codes = F.pad(audio_codes[:, :max_mel_len], (0, 1), value=self.stop_mel_token)
# Pad mel codes with STOP_MEL_TOKEN
audio_codes = self.set_mel_padding(audio_codes, mel_lengths)
# Compute speech conditioning input
conds = None
if speech_conditioning_input is not None:
if not return_latent:
# Compute speech conditioning input
speech_conditioning_input = (
speech_conditioning_input.unsqueeze(1)
if len(speech_conditioning_input.shape) == 3
else speech_conditioning_input
)
conds = []
for j in range(speech_conditioning_input.shape[1]):
conds.append(self.conditioning_encoder(speech_conditioning_input[:, j]))
conds = torch.stack(conds, dim=1)
if self.average_conditioning_embeddings:
conds = conds.mean(dim=1).unsqueeze(1)
else:
# already computed
conds = speech_conditioning_input.unsqueeze(1)
# Build input and target tensors
# Prepend start token to inputs and append stop token to targets
text_inputs, _ = self.set_inputs_and_targets(text_inputs, self.start_text_token, self.stop_text_token)
audio_codes, _ = self.set_inputs_and_targets(audio_codes, self.start_mel_token, self.stop_mel_token)
# Set attn_mask
attn_mask_text = None
attn_mask_mel = None
if not return_latent:
attn_mask_text = torch.ones(
text_inputs.shape[0],
text_inputs.shape[1],
dtype=torch.bool,
device=text_inputs.device,
)
attn_mask_mel = torch.ones(
audio_codes.shape[0],
audio_codes.shape[1],
dtype=torch.bool,
device=audio_codes.device,
)
for idx, l in enumerate(text_lengths):
attn_mask_text[idx, l + 1 :] = 0.0
for idx, l in enumerate(mel_lengths):
attn_mask_mel[idx, l + 1 :] = 0.0
# Compute text embeddings + positional embeddings
# print(" > text input latent:", text_inputs)
text_emb = self.text_embedding(text_inputs) + self.text_pos_embedding(text_inputs)
# Compute mel embeddings + positional embeddings
audio_emb = self.audio_embedding(audio_codes) + self.audio_embedding(audio_codes)
# Compute prompt embeddings + positional embeddings
prompt = self.get_prompts(prompt_codes)
# prompt_emb = self.audio_embedding(prompt).detach() + self.mel_pos_embedding(prompt).detach()
prompt_emb = self.prompt_embedding(prompt) + self.prompt_pos_embedding(prompt)
# dropout prompt embeddings
prompt_emb = F.dropout(prompt_emb, p=0.1, training=self.training)
# Get logits
sub = -4 # don't ask me why 😄
if self.training:
sub = -1
_, audio_logits = self.get_logits(
conds,
text_emb,
self.text_head,
audio_emb,
self.mel_head,
prompt=prompt_emb,
get_attns=return_attentions,
return_latent=return_latent,
attn_mask_text=attn_mask_text,
attn_mask_mel=attn_mask_mel,
)
return audio_logits[:, :sub] # sub to prevent bla.
def compute_embeddings(
self,
speech_conditioning_latent,
text_inputs,
input_tokens=None,
prompt_codes=None,
pad_input_text=False,
):
"""Compute all the embeddings needed for inference."""
if pad_input_text and text_inputs.shape[1] < 250:
text_inputs = F.pad(text_inputs, (0, 250 - text_inputs.shape[1]), value=self.stop_text_token)
else:
text_inputs = F.pad(text_inputs, (0, 1), value=self.stop_text_token)
text_inputs = F.pad(text_inputs, (1, 0), value=self.start_text_token)
emb = self.text_embedding(text_inputs) + self.text_pos_embedding(text_inputs)
print(" > Text inputs:", text_inputs)
if prompt_codes is not None:
prompt_codes = self.get_prompts(prompt_codes)
# prompt_emb = self.audio_embedding(prompt_codes) + self.mel_pos_embedding(prompt_codes)
prompt_emb = self.prompt_embedding(prompt_codes) + self.prompt_pos_embedding(prompt_codes)
print(" > Prompt inputs:", prompt_codes)
print(" > Prompt inputs shape:", prompt_codes.shape)
emb = torch.cat([prompt_emb, emb], dim=1)
if speech_conditioning_latent is not None:
conds = speech_conditioning_latent.unsqueeze(1)
emb = torch.cat([conds, emb], dim=1)
self.inference_model.store_prefix_emb(emb)
fake_inputs = torch.full(
(
emb.shape[0],
emb.shape[1] + 1, # +1 for the start_mel_token
),
fill_value=1,
dtype=torch.long,
device=text_inputs.device,
)
fake_inputs[:, -1] = self.start_mel_token
if input_tokens is not None:
fake_inputs = torch.cat([fake_inputs, input_tokens], dim=1)
return fake_inputs
def inference(
self,
text_inputs,
input_tokens=None,
prompt_codes=None,
pad_input_text=False,
**hf_generate_kwargs,
):
if pad_input_text and text_inputs.shape[1] < 250:
text_inputs = F.pad(text_inputs, (0, 250 - text_inputs.shape[1]), value=self.stop_text_token)
else:
text_inputs = F.pad(text_inputs, (0, 1), value=self.stop_text_token)
text_inputs = F.pad(text_inputs, (1, 0), value=self.start_text_token)
emb = self.text_embedding(text_inputs) + self.text_pos_embedding(text_inputs)
if prompt_codes is not None:
prompt_codes = self.get_prompts(prompt_codes)
prompt_emb = self.prompt_embedding(prompt_codes) + self.prompt_pos_embedding(prompt_codes)
emb = torch.cat([prompt_emb, emb], dim=1)
self.inference_model.store_prefix_emb(emb)
fake_inputs = torch.full(
(
emb.shape[0],
emb.shape[1] + 1, # +1 for the start_mel_token
),
fill_value=1,
dtype=torch.long,
device=text_inputs.device,
)
fake_inputs[:, -1] = self.start_mel_token
if input_tokens is not None:
fake_inputs = torch.cat([fake_inputs, input_tokens], dim=1)
gen = self.inference_model.generate(
fake_inputs,
bos_token_id=self.start_mel_token,
pad_token_id=self.stop_mel_token,
eos_token_id=self.stop_mel_token,
max_length=self.max_audio_tokens * 2 + self.max_prompt_tokens + self.max_text_tokens,
**hf_generate_kwargs,
)
if "return_dict_in_generate" in hf_generate_kwargs:
return gen.sequences[:, fake_inputs.shape[1] :], gen
return gen[:, fake_inputs.shape[1] :]