Distribute.py with multiple GPU's hangs indefinitely

Hey all,

I’m sure this is me just missing a piece of the puzzle but I’m stumped.

If I run the following:

CUDA_VISIBLE_DEVICES="1" python TTS/bin/distribute.py --script train_tacotron.py --config_path=/workspace/tacotron/config.json --continue_path=/workspace/tacotron/

Then I get a normally running instance. (I see files found and then processing begins)

Alternatively, if I run:

CUDA_VISIBLE_DEVICES="1,2" python TTS/bin/distribute.py --script train_tacotron.py --config_path=/workspace/tacotron/config.json --continue_path=/workspace/tacotron/

Then I only get the following output (it then hangs indefinitely):

['/workspace/TTS/TTS/bin/train_tacotron.py', '--continue_path=/workspace/tacotron/', '--restore_path=', '--config_path=/workspace/tacotron/config.json', '--group_id=group_2021_01_25-154310', '--rank=0']
['/workspace/TTS/TTS/bin/train_tacotron.py', '--continue_path=/workspace/tacotron/', '--restore_path=', '--config_path=/workspace/tacotron/config.json', '--group_id=group_2021_01_25-154310', '--rank=1']
2021-01-25 15:43:11.260445: I tensorflow/stream_executor/platform/default/dso_loader.cc:49] Successfully opened dynamic library libcudart.so.11.0
2021-01-25 15:43:11.308820: I tensorflow/stream_executor/platform/default/dso_loader.cc:49] Successfully opened dynamic library libcudart.so.11.0
 > Using CUDA:  True
 > Number of GPUs:  2
 > Training continues for /workspace/tacotron/
 > Setting up Audio Processor...
 | > sample_rate:48000
 | > resample:False
 | > num_mels:80
 | > min_level_db:-100
 | > frame_shift_ms:None
 | > frame_length_ms:None
 | > ref_level_db:0
 | > fft_size:1024
 | > power:1.5
 | > preemphasis:0.0
 | > griffin_lim_iters:60
 | > signal_norm:True
 | > symmetric_norm:True
 | > mel_fmin:50.0
 | > mel_fmax:7600.0
 | > spec_gain:1.0
 | > stft_pad_mode:reflect
 | > max_norm:4.0
 | > clip_norm:True
 | > do_trim_silence:True
 | > trim_db:25
 | > do_sound_norm:False
 | > stats_path:/workspace/scale_stats.npy
 | > hop_length:256
 | > win_length:1024
 | > Found 39490 files in /data/VCTK/VCTK-Corpus

Netstat shows the processes coming up and listening:
tcp 0 0* LISTEN

I use the same config for both:
(Note, I experience the same behavior if I’m pointed at port 54321 or 55555 as reported here. (Used an alternative port to rule out any issues with initial 54321 port))


“github_branch”:"* master",
“model”: “Tacotron2”, // one of the model in models/
“run_name”: “vctk-r=2-ddc”,
“run_description”: “tacotron2 on vctl r=2 with ddc only without guided attention”,
“compute_input_seq_cache”: false,
“mixed_precision”: false,

    "fft_size": 1024,         // number of stft frequency levels. Size of the linear spectogram frame.
    "win_length": 1024,      // stft window length in ms.
    "hop_length": 256,       // stft window hop-lengh in ms.
    "frame_length_ms": null, // stft window length in ms.If null, 'win_length' is used.
    "frame_shift_ms": null,  // stft window hop-lengh in ms. If null, 'hop_length' is used.

    // Audio processing parameters
    // "sample_rate": 22050,   // DATASET-RELATED: wav sample-rate. If different than the original data, it is resampled.
"sample_rate": 48000,
    "preemphasis": 0.0,     // pre-emphasis to reduce spec noise and make it more structured. If 0.0, no -pre-emphasis.
    "ref_level_db": 0,     // reference level db, theoretically 20db is the sound of air.

    // Silence trimming
    "do_trim_silence": true,// enable trimming of slience of audio as you load it. LJspeech (false), TWEB (false), Nancy (true)
    "trim_db": 25,          // threshold for timming silence. Set this according to your dataset.

// Griffin-Lim
"power": 1.5,           // value to sharpen wav signals after GL algorithm.
"griffin_lim_iters": 60,// #griffin-lim iterations. 30-60 is a good range. Larger the value, slower the generation.

    // MelSpectrogram parameters
    "num_mels": 80,         // size of the mel spec frame.
    "mel_fmin": 50.0,        // minimum freq level for mel-spec. ~50 for male and ~95 for female voices. Tune for dataset!!
    "mel_fmax": 7600.0,     // maximum freq level for mel-spec. Tune for dataset!!
    "spec_gain": 1.0,         // scaler value appplied after log transform of spectrogram.

    // Normalization parameters
    "signal_norm": true,    // normalize spec values. Mean-Var normalization if 'stats_path' is defined otherwise range normalization defined by the other params.
    "min_level_db": -100,   // lower bound for normalization
    "symmetric_norm": true, // move normalization to range [-1, 1]
    "max_norm": 4.0,        // scale normalization to range [-max_norm, max_norm] or [0, max_norm]
    "clip_norm": true,      // clip normalized values into the range.
    "stats_path": "/workspace/scale_stats.npy"    // DO NOT USE WITH MULTI_SPEAKER MODEL. scaler stats file computed by 'compute_statistics.py'. If it is defined, mean-std based notmalization is used and other normalization params are ignored

// if custom character set is not defined,
// default set in symbols.py is used
    "pad": "_",
    "eos": "&",
    "bos": "*",
    //"characters": "ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz!'(),-.:;? ",
    //"characters": "ABCDEFGHIJKLMçãàáâêéíóôõúûabcdefghijklmnopqrstuvwxyz!'(),-.:;? ",
    "characters": "ABCDEFGHIJKLMNOPQRSTUVWXYZÇÃÀÁÂÊÉÍÓÔÕÚÛabcdefghijklmnopqrstuvwxyzçãàáâêéíóôõúû!(),-.:;? ",
    "punctuations":"!'(),-.:;? ",
    "phonemes":"iyɨʉɯuɪʏʊeøɘəɵɤoɛœɜɞʌɔæɐaɶɑɒᵻʘɓǀɗǃʄǂɠǁʛpbtdʈɖcɟkɡqɢʔɴŋɲɳnɱmʙrʀⱱɾɽɸβfvθðszʃʒʂʐçʝxɣχʁħʕhɦɬɮʋɹɻjɰlɭʎʟˈˌːˑʍwɥʜʢʡɕʑɺɧɚ˞ɫ'̃' "

** “distributed”:{**
** “backend”: “nccl”,**
** “url”: “tcp://localhost:55555”**
** },**

"reinit_layers": [],    // give a list of layer names to restore from the given checkpoint. If not defined, it reloads all heuristically matching layers.

"batch_size": 16,       // Batch size for training. Lower values than 32 might cause hard to learn attention. It is overwritten by 'gradual_training'.
"r": 2,                 // Number of decoder frames to predict per iteration. Set the initial values if gradual training is enabled.  
"gradual_training": null, //set gradual training steps [first_step, r, batch_size]. If it is null, gradual training is disabled. For Tacotron, you might need to reduce the 'batch_size' as you proceeed.

"loss_masking": false,       // enable / disable loss masking against the sequence padding.
"decoder_loss_alpha": 0.5,  // decoder loss weight. If > 0, it is enabled
"postnet_loss_alpha": 0.25, // postnet loss weight. If > 0, it is enabled
"ga_alpha": 0.0,           // weight for guided attention loss. If > 0, guided attention is enabled.
"decoder_diff_spec_alpha": 0.25,     // differential spectral loss weight. If > 0, it is enabled
"postnet_diff_spec_alpha": 0.25,     // differential spectral loss weight. If > 0, it is enabled
"decoder_ssim_alpha": 0.5,     // differential spectral loss weight. If > 0, it is enabled
"postnet_ssim_alpha": 0.25,     // differential spectral loss weight. If > 0, it is enabled

"run_eval": true,
"test_delay_epochs": 1,  //Until attention is aligned, testing only wastes computation time.
"test_sentences_file": null,//"../../../datasets/BRSpeech-3-Speakers-Paper/BRSpeech-3-Speakers-Paper/TTS-Portuguese_Corpus/test_setences.txt",  // set a file to load sentences to be used for testing. If it is null then we use default english sentences.

"noam_schedule": false,        // use noam warmup and lr schedule.
"grad_clip": 1.0,                // upper limit for gradients for clipping.
"epochs": 1000,                // total number of epochs to train.
"lr": 0.0001,                  // Initial learning rate. If Noam decay is active, maximum learning rate.
"wd": 0.000001,         // Weight decay weight.
"warmup_steps": 4000,          // Noam decay steps to increase the learning rate from 0 to "lr"
"seq_len_norm": true,	   // Normalize eash sample loss with its length to alleviate imbalanced datasets. Use it if your dataset is small or has skewed distribution of sequence lengths.

"memory_size": -1,              // ONLY TACOTRON - size of the memory queue used fro storing last decoder predictions for auto-regression. If < 0, memory queue is disabled and decoder only uses the last prediction frame. 
"prenet_type": "original",     // "original" or "bn".
"prenet_dropout": true,        // enable/disable dropout at prenet. 

"attention_type": "original",  // 'original' or 'graves'
"attention_heads": 4,          // number of attention heads (only for 'graves')
"attention_norm": "softmax",   // softmax or sigmoid. Suggested to use softmax for Tacotron2 and sigmoid for Tacotron.
"windowing": false,            // Enables attention windowing. Used only in eval mode.
"use_forward_attn": false,      // if it uses forward attention. In general, it aligns faster.
"forward_attn_mask": false,    // Additional masking forcing monotonicity only in eval mode.
"transition_agent": false,     // enable/disable transition agent of forward attention.
"location_attn": true,        // enable_disable location sensitive attention. It is enabled for TACOTRON by default.
"bidirectional_decoder": false,  // use https://arxiv.org/abs/1907.09006. Use it, if attention does not work well with your dataset.
"double_decoder_consistency": true,  // use DDC explained here https://erogol.com/solving-attention-problems-of-tts-models-with-double-decoder-consistency-draft/
"ddc_r": 4,                           // reduction rate for coarse decoder.
"apex_amp_level": null, // level of optimization with NVIDIA's apex feature for automatic mixed FP16/FP32 precision (AMP), NOTE: currently only O1 is supported, and use "O1" to activate.

"stopnet": true,               // Train stopnet predicting the end of synthesis. 
"separate_stopnet": true,     // Train stopnet seperately if 'stopnet==true'. It prevents stopnet loss to influence the rest of the model. It causes a better model, but it trains SLOWER.

"print_step": 50,       // Number of steps to log traning on console.
"tb_plot_step": 50,    // Number of steps to plot TB training figures.
"print_eval": false,     // If True, it prints intermediate loss values in evalulation.
"save_step": 500,      // Number of training steps expected to save traninpg stats and checkpoints.
"checkpoint": true,     // If true, it saves checkpoints per "save_step"
"tb_model_param_stats": false,     // true, plots param stats per layer on tensorboard. Might be memory consuming, but good for debugging. 

"text_cleaner": "english_cleaners",
"enable_eos_bos_chars": false, // enable/disable beginning of sentence and end of sentence chars.
"num_loader_workers": 8,        // number of training data loader processes. Don't set it too big. 4-8 are good values.
"num_val_loader_workers": 8,    // number of evaluation data loader processes.
"batch_group_size": 0,  //Number of batches to shuffle after bucketing.
"min_seq_len": 2,       // DATASET-RELATED: minimum text length to use in training
"max_seq_len": 153,     // DATASET-RELATED: maximum text length

// "output_path": "/data5/rw/pit/keep/",      // DATASET-RELATED: output path for all training outputs.
"output_path": "/workspace/output/",

"phoneme_cache_path": "/workspace/checkpoints/mozilla_en_phonemes/",  // phoneme computation is slow, therefore, it caches results in the given folder.
"use_phonemes": true,           // use phonemes instead of raw characters. It is suggested for better pronounciation.
"phoneme_language": "en",     // depending on your target language, pick one from  https://github.com/bootphon/phonemizer#languages

// "speaker_encoder_config_path":"/root/speaker_encoder/LibriTTS-common-voice-voxceleb_angle_proto/config.json", // config.json for the speaker encoder
// "speaker_encoder_checkpoint_path": "/root/speaker_encoder/LibriTTS-common-voice-voxceleb_angle_proto/320k.pth.tar", // Speaker Encoder Checkpoint full path
"use_speaker_embedding": true,     // use speaker embedding to enable multi-speaker learning.
"use_external_speaker_embedding_file": true, // if true, forces the model to use external embedding per sample instead of nn.embeddings, that is, it supports external embeddings such as those used at: https://arxiv.org/abs /1806.04558
"external_speaker_embedding_file": "/root/speakers.json", // if not null and use_external_speaker_embedding_file is true, it is used to load a specific embedding file and thus uses these embeddings instead of nn.embeddings, that is, it supports external embeddings such as those used at: https://arxiv.org/abs /1806.04558
"use_gst": false,       // TACOTRON ONLY: use global style tokens

"gst":	{			                // gst parameter if gst is enabled
    "gst_style_input": null,        // Condition the style input either on a 
                                    // -> wave file [path to wave] or 
                                    // -> dictionary using the style tokens {'token1': 'value', 'token2': 'value'} example {"0": 0.15, "1": 0.15, "5": -0.15}  
                                    // with the dictionary being len(dict) <= len(gst_style_tokens).
    "disable_gst": false, // if true its force GST predict zero matrix.
    "gst_use_speaker_embedding": true, // if true pass speaker embedding in attention input GST.
    "gst_embedding_dim": 128,       
    "gst_num_heads": 4,
    "gst_style_tokens": 5
    "datasets":   // List of datasets. They all merged and they get different s$
            "name": "vctk",
            "path": "/data/VCTK/VCTK-Corpus",
            // para o teste no vctk eu escolhi 1 locutor de cada ACCEPTS, 
            "meta_file_train": ["p225", "p234", "p238", "p245", "p248", "p261", "p294", "p302", "p326", "p335", "p347"], // for vtck if list, ignore speakers id in list for train, its useful for test cloning with new speakers
            "meta_file_val": null


I have been unable to determine what add’l configurations are necessary to get nccl online. Any thoughts would be welcome. Note that these are running in docker instances on a CUDA-powered quad-GPU server. Not attempting to run across multiple machines at this point.


As a follow-up, I found out that we have 2 dual-processor GPU’s on the box. It appears that I can run a single process using 1 processor from either GPU, but when I attempt to use both processors on the same GPU for the same process, we hang.

(I found that CUDA_VISIBLE_DEVICES=0,2 for one process and then another kicked off with CUDA_VISIBLE_DEVICES=1,3 got both processes running)

Now, to figure out how to segment the data in such a way that I can run two parallel processes in an efficient manner.

Feedback / thoughts are very welcome.

1 Like

I am not sure if I could follow-up the problem. Right now you solved the first issue with running multi-gpu training and asking some advice for arranging the dataset. Is that right ?

Aye, apologies. I was thinking out loud more than anything.