Windows specific fixes (to-do: find libespeak-ng.dll automatically because it cannot be trusted to do it by default)

setup.py

@@ -37,7 +37,7 @@ setup(
     packages=find_packages(),
     install_requires=(
         # training backends
-        ["deepspeed>=0.7.7"] if not sys.platform.startswith("win") else [])
+        ["deepspeed>=0.7.7"] if not sys.platform.startswith("win") else ["psutil"])
         + [
         # logging niceties
         "coloredlogs>=15.0.1",

vall_e/config.py

@@ -948,6 +948,14 @@ class Config(BaseConfig):
 				_logger.warning(f"Deprecated flag found: {'cfg.model.interleave'}")
 				del model["interleave"]
 
+			if "p_rvq_levels" in model["experimental"]:
+				model["experimental"]["rvq_levels_p"] = model["experimental"]["p_rvq_levels"]
+				del model["experimental"]["p_rvq_levels"]
+
+			if "p_len_train" in model["experimental"]:
+				model["experimental"]["len_train_p"] = model["experimental"]["p_len_train"]
+				del model["experimental"]["p_len_train"]
+
 		self.models = [ Model(**model) if isinstance(model, dict) else model for model in self.models ]
 		self.loras = [ LoRA(**lora)  if isinstance(lora, dict) else lora for lora in self.loras ]
 

vall_e/demo.py

@@ -156,6 +156,12 @@ def main():
 
 		comparison_kwargs["disabled"]["layer_skip"] = False
 		comparison_kwargs["enabled"]["layer_skip"] = True
+	elif args.comparison == "refine-on-stop":
+		comparison_kwargs["suffix"] = "refine-on-stop"
+		comparison_kwargs["titles"] = [f"Without Ro<S>", "With Ro<S>"]
+
+		comparison_kwargs["disabled"]["refine_on_stop"] = False
+		comparison_kwargs["enabled"]["refine_on_stop"] = True
 	elif args.comparison == "ar-temp":
 		current_temp = args.ar_temp
 		other_temp = 1.0

vall_e/engines/base.py

@@ -48,8 +48,12 @@ from ..utils import wrapper as ml
 
 _logger = logging.getLogger(__name__)
 
-if not distributed_initialized() and cfg.trainer.backend == "local": # and world_size() > 1:
-	init_distributed(torch.distributed.init_process_group)
+# windows throws an error here
+try:
+	if not distributed_initialized() and cfg.trainer.backend == "local": # and world_size() > 1:
+		init_distributed(torch.distributed.init_process_group)
+except Exception as e:
+	pass
 
 # A very naive engine implementation using barebones PyTorch
 class Engine():

vall_e/models/arch/llama.py

@@ -5,7 +5,7 @@ import torch
 import logging
 import random
 
-from typing import Literal, overload, Optional, Tuple, Union, List, Unpack
+from typing import Literal, overload, Optional, Tuple, Union, List
 
 from torch import Tensor, nn
 from transformers.cache_utils import Cache

vall_e/webui.py

@@ -1,4 +1,5 @@
 import os
+import sys
 import re
 import math
 import argparse
@@ -22,6 +23,8 @@ from .emb.qnt import decode_to_wave
 from .data import get_lang_symmap, get_random_prompt
 
 
+is_windows = sys.platform.startswith("win")
+
 tts = None
 
 layout = {}
@@ -68,6 +71,9 @@ def get_model_paths( paths=[Path("./training/"), Path("./models/"), Path("./data
 				continue
 			configs.append( sft )
 
+	if is_windows:
+		configs = [ str(p) for p in configs ]
+
 	return configs
 
 def get_dtypes():
@@ -199,7 +205,10 @@ def do_inference_tts( progress=gr.Progress(track_tqdm=True), *args, **kwargs ):
 	parser.add_argument("--refine-on-stop", action="store_true")
 	args, unknown = parser.parse_known_args()
 
-	tmp = tempfile.NamedTemporaryFile(suffix='.wav')
+	if is_windows:
+		tmp = tempfile.NamedTemporaryFile(suffix='.wav', delete=False)
+	else:
+		tmp = tempfile.NamedTemporaryFile(suffix='.wav')
 
 	"""
 	if not args.references: