forked from mrq/ai-voice-cloning
VALL-E config edits
This commit is contained in:
parent
2e33bf071a
commit
34ef0467b9
|
@ -1,41 +1,61 @@
|
||||||
{
|
{
|
||||||
"autotuning": {
|
"optimizer": {
|
||||||
"enabled": false,
|
"type": "AdamW",
|
||||||
"results_dir": "./config/autotune/results",
|
"params": {
|
||||||
"exps_dir": "./config/autotune/exps",
|
"lr": 2e-05,
|
||||||
"overwrite": false,
|
"betas": [
|
||||||
"metric": "throughput",
|
0.9,
|
||||||
"start_profile_step": 10,
|
0.96
|
||||||
"end_profile_step": 20,
|
],
|
||||||
"fast": false,
|
"eps": 1e-07,
|
||||||
"max_train_batch_size": 32,
|
"weight_decay": 0.01
|
||||||
"mp_size": 1,
|
}
|
||||||
"num_tuning_micro_batch_sizes": 3,
|
},
|
||||||
"tuner_type": "model_based",
|
"scheduler":{
|
||||||
"tuner_early_stopping": 5,
|
"type":"WarmupLR",
|
||||||
"tuner_num_trials": 50,
|
"params":{
|
||||||
"arg_mappings": {
|
"warmup_min_lr":0,
|
||||||
"train_micro_batch_size_per_gpu": "--per_device_train_batch_size",
|
"warmup_max_lr":2e-5,
|
||||||
"gradient_accumulation_steps ": "--gradient_accumulation_steps"
|
"warmup_num_steps":100,
|
||||||
|
"warmup_type":"linear"
|
||||||
}
|
}
|
||||||
},
|
},
|
||||||
"zero_optimization": {
|
"fp16":{
|
||||||
"stage": 0,
|
"enabled":true,
|
||||||
"offload_param": {
|
"loss_scale":0,
|
||||||
"device": "nvme",
|
"loss_scale_window":1000,
|
||||||
"nvme_path": "/tmp/zero/",
|
"initial_scale_power":16,
|
||||||
"pin_memory": false,
|
"hysteresis":2,
|
||||||
"buffer_count": 5,
|
"min_loss_scale":1
|
||||||
"buffer_size": 1e9,
|
},
|
||||||
"max_in_cpu": 1e9
|
"autotuning":{
|
||||||
},
|
"enabled":false,
|
||||||
"overlap_comm": true,
|
"results_dir":"./config/autotune/results",
|
||||||
"reduce_bucket_size": "auto",
|
"exps_dir":"./config/autotune/exps",
|
||||||
"contiguous_gradients": true,
|
"overwrite":false,
|
||||||
"sub_group_size": 1e8,
|
"metric":"throughput",
|
||||||
"stage3_prefetch_bucket_size": "auto",
|
"start_profile_step":10,
|
||||||
"stage3_param_persistence_threshold": "auto",
|
"end_profile_step":20,
|
||||||
"stage3_max_live_parameters": "auto",
|
"fast":false,
|
||||||
"stage3_max_reuse_distance": "auto"
|
"max_train_batch_size":32,
|
||||||
|
"mp_size":1,
|
||||||
|
"num_tuning_micro_batch_sizes":3,
|
||||||
|
"tuner_type":"model_based",
|
||||||
|
"tuner_early_stopping":5,
|
||||||
|
"tuner_num_trials":50,
|
||||||
|
"arg_mappings":{
|
||||||
|
"train_micro_batch_size_per_gpu":"--per_device_train_batch_size",
|
||||||
|
"gradient_accumulation_steps ":"--gradient_accumulation_steps"
|
||||||
|
}
|
||||||
|
},
|
||||||
|
"zero_optimization":{
|
||||||
|
"stage":0,
|
||||||
|
"reduce_bucket_size":"auto",
|
||||||
|
"contiguous_gradients":true,
|
||||||
|
"sub_group_size":1e8,
|
||||||
|
"stage3_prefetch_bucket_size":"auto",
|
||||||
|
"stage3_param_persistence_threshold":"auto",
|
||||||
|
"stage3_max_live_parameters":"auto",
|
||||||
|
"stage3_max_reuse_distance":"auto"
|
||||||
}
|
}
|
||||||
}
|
}
|
|
@ -3,14 +3,17 @@ ckpt_root: ./training/${voice}/finetune/ckpt/
|
||||||
log_root: ./training/${voice}/finetune/logs/
|
log_root: ./training/${voice}/finetune/logs/
|
||||||
|
|
||||||
data_dirs: [./training/${voice}/valle/]
|
data_dirs: [./training/${voice}/valle/]
|
||||||
spkr_name_getter: "lambda p: p.parts[-3]"
|
spkr_name_getter: "lambda p: p.parts[-3]" # "lambda p: p.parts[-1].split('-')[0]"
|
||||||
|
|
||||||
model: ${model_name}
|
model: ${model_name}
|
||||||
batch_size: ${batch_size}
|
batch_size: ${batch_size}
|
||||||
eval_batch_size: ${validation_batch_size}
|
gradient_accumulation_steps: ${gradient_accumulation_size}
|
||||||
|
eval_batch_size: ${batch_size}
|
||||||
|
|
||||||
max_iter: ${iterations}
|
max_iter: ${iterations}
|
||||||
save_ckpt_every: ${save_rate}
|
save_ckpt_every: ${save_rate}
|
||||||
eval_every: ${validation_rate}
|
eval_every: ${validation_rate}
|
||||||
|
|
||||||
|
max_phones: 256
|
||||||
|
|
||||||
sampling_temperature: 1.0
|
sampling_temperature: 1.0
|
|
@ -488,7 +488,7 @@ def setup_gradio():
|
||||||
)
|
)
|
||||||
with gr.Row():
|
with gr.Row():
|
||||||
TRAINING_SETTINGS["batch_size"] = gr.Number(label="Batch Size", value=128, precision=0)
|
TRAINING_SETTINGS["batch_size"] = gr.Number(label="Batch Size", value=128, precision=0)
|
||||||
TRAINING_SETTINGS["gradient_accumulation_size"] = gr.Number(label="Gradient Accumulation Size", value=4, precision=0, visible=args.tts_backend=="tortoise")
|
TRAINING_SETTINGS["gradient_accumulation_size"] = gr.Number(label="Gradient Accumulation Size", value=4, precision=0)
|
||||||
with gr.Row():
|
with gr.Row():
|
||||||
TRAINING_SETTINGS["save_rate"] = gr.Number(label="Save Frequency (in epochs)", value=5, precision=0)
|
TRAINING_SETTINGS["save_rate"] = gr.Number(label="Save Frequency (in epochs)", value=5, precision=0)
|
||||||
TRAINING_SETTINGS["validation_rate"] = gr.Number(label="Validation Frequency (in epochs)", value=5, precision=0)
|
TRAINING_SETTINGS["validation_rate"] = gr.Number(label="Validation Frequency (in epochs)", value=5, precision=0)
|
||||||
|
|
Loading…
Reference in New Issue
Block a user