mrq/vall-e
5
9
Fork 0
Code Issues 8 Pull Requests Packages Projects Releases Wiki Activity

might just be better to explicitly define prompt duration ranges, especially under a "train small contexts then increase it" training paradigm

Browse Source
This commit is contained in:
mrq 2024-05-11 09:50:54 -05:00
parent bd0a36ba8d
commit 4d93a16ef7
2 changed files with 10 additions and 9 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

7
vall_e/config.py
View File

@ -144,11 +144,13 @@ class Dataset:
phones_range: list[int] = field(default_factory=lambda: [4, 256])
duration_range: list[float] = field(default_factory=lambda: [1.0, 12.0])
prompt_duration_range: list[float] = field(default_factory=lambda: [3.0, 6.0])
min_utterances: int = 2
random_utterance: float = 1.0
max_prompts: int = 3
prompt_duration: float = 3.0
prompt_duration: float = 0.0 # legacy
max_resps: int = 1
p_resp_append: float = 1.0
@ -676,6 +678,9 @@ class Config(_Config):
if self.hyperparameters.scheduler == "":
self.hyperparameters.torch_scheduler = True
if self.dataset.prompt_duration != 0:
self.dataset.prompt_duration_range = [self.dataset.prompt_duration, self.dataset.prompt_duration]
# Preserves the old behavior
class NaiveTokenizer:
def get_vocab( self ):

12
vall_e/data.py
View File

@ -368,12 +368,8 @@ class Dataset(_Dataset):
)
"""
# shuffle it up a bit
prom_length = 0
if cfg.experimental and False:
trim_length = max(2, int(np.random.normal(loc=5, scale=1.25) * cfg.dataset.frames_per_second))
else:
trim_length = int(cfg.dataset.prompt_duration * cfg.dataset.frames_per_second) + random.randint(-cfg.dataset.frames_per_second, cfg.dataset.frames_per_second)
trim_length = random.randint(cfg.dataset.prompt_duration_range[0], cfg.dataset.prompt_duration_range[1]) * cfg.dataset.frames_per_second
for _ in range(cfg.dataset.max_prompts):
path = random.choice(choices)
@ -388,7 +384,7 @@ class Dataset(_Dataset):
else:
qnt = _load_quants(path)
if cfg.dataset.prompt_duration > 0 and trim_length < qnt.shape[0]:
if 0 < trim_length and trim_length < qnt.shape[0]:
qnt = trim( qnt, trim_length )
prom_list.append(qnt)
@ -401,7 +397,7 @@ class Dataset(_Dataset):
# as you technically can't just append encodec sequences together like this without issues
prom = torch.cat(prom_list)
if cfg.dataset.prompt_duration > 0 and trim_length < prom.shape[0]:
if 0 < trim_length and trim_length < prom.shape[0]:
prom = trim( prom, trim_length )
return prom
@ -474,7 +470,7 @@ class Dataset(_Dataset):
resps = torch.concat([ resps, qnt ])
task = "tts"
trim_length = int(cfg.dataset.prompt_duration * cfg.dataset.frames_per_second)
trim_length = random.randint(cfg.dataset.prompt_duration_range[0], cfg.dataset.prompt_duration_range[1]) * cfg.dataset.frames_per_second
proms = self.sample_prompts(spkr_name, ignore=path) if random.random() < cfg.dataset.random_utterance else resps