From a50bbbb476ef9df6ee03fc444632b9e573332b2c Mon Sep 17 00:00:00 2001 From: ModelHub XC Date: Wed, 20 May 2026 23:12:50 +0800 Subject: [PATCH] =?UTF-8?q?=E5=88=9D=E5=A7=8B=E5=8C=96=E9=A1=B9=E7=9B=AE?= =?UTF-8?q?=EF=BC=8C=E7=94=B1ModelHub=20XC=E7=A4=BE=E5=8C=BA=E6=8F=90?= =?UTF-8?q?=E4=BE=9B=E6=A8=A1=E5=9E=8B?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Model: Jzuluaga/wav2vec2-large-960h-lv60-self-en-atc-uwb-atcc-and-atcosim Source: Original Platform --- .gitattributes | 35 + README.md | 244 + added_tokens.json | 4 + all_results.json | 14 + alphabet.json | 1 + config.json | 108 + eval_results.json | 9 + language_model/atcosim_uwb_atcc_4g.binary | 3 + language_model/attrs.json | 1 + language_model/unigrams.txt | 0 log/eval_model_train | 15 + log/train_log | 21388 ++++++++++++++++++++ preprocessor_config.json | 10 + pytorch_model.bin | 3 + special_tokens_map.json | 6 + tokenizer_config.json | 11 + train_results.json | 8 + trainer_state.json | 265 + training_args.bin | 3 + vocab.json | 31 + 20 files changed, 22159 insertions(+) create mode 100644 .gitattributes create mode 100644 README.md create mode 100644 added_tokens.json create mode 100644 all_results.json create mode 100644 alphabet.json create mode 100644 config.json create mode 100644 eval_results.json create mode 100644 language_model/atcosim_uwb_atcc_4g.binary create mode 100644 language_model/attrs.json create mode 100644 language_model/unigrams.txt create mode 100644 log/eval_model_train create mode 100644 log/train_log create mode 100644 preprocessor_config.json create mode 100644 pytorch_model.bin create mode 100644 special_tokens_map.json create mode 100644 tokenizer_config.json create mode 100644 train_results.json create mode 100644 trainer_state.json create mode 100644 training_args.bin create mode 100644 vocab.json diff --git a/.gitattributes b/.gitattributes new file mode 100644 index 0000000..f6925f7 --- /dev/null +++ b/.gitattributes @@ -0,0 +1,35 @@ +*.7z filter=lfs diff=lfs merge=lfs -text +*.arrow filter=lfs diff=lfs merge=lfs -text +*.bin filter=lfs diff=lfs merge=lfs -text +*.bz2 filter=lfs diff=lfs merge=lfs -text +*.ckpt filter=lfs diff=lfs merge=lfs -text +*.ftz filter=lfs diff=lfs merge=lfs -text +*.gz filter=lfs diff=lfs merge=lfs -text +*.h5 filter=lfs diff=lfs merge=lfs -text +*.joblib filter=lfs diff=lfs merge=lfs -text +*.lfs.* filter=lfs diff=lfs merge=lfs -text +*.mlmodel filter=lfs diff=lfs merge=lfs -text +*.model filter=lfs diff=lfs merge=lfs -text +*.msgpack filter=lfs diff=lfs merge=lfs -text +*.npy filter=lfs diff=lfs merge=lfs -text +*.npz filter=lfs diff=lfs merge=lfs -text +*.onnx filter=lfs diff=lfs merge=lfs -text +*.ot filter=lfs diff=lfs merge=lfs -text +*.parquet filter=lfs diff=lfs merge=lfs -text +*.pb filter=lfs diff=lfs merge=lfs -text +*.pickle filter=lfs diff=lfs merge=lfs -text +*.pkl filter=lfs diff=lfs merge=lfs -text +*.pt filter=lfs diff=lfs merge=lfs -text +*.pth filter=lfs diff=lfs merge=lfs -text +*.rar filter=lfs diff=lfs merge=lfs -text +*.safetensors filter=lfs diff=lfs merge=lfs -text +saved_model/**/* filter=lfs diff=lfs merge=lfs -text +*.tar.* filter=lfs diff=lfs merge=lfs -text +*.tflite filter=lfs diff=lfs merge=lfs -text +*.tgz filter=lfs diff=lfs merge=lfs -text +*.wasm filter=lfs diff=lfs merge=lfs -text +*.xz filter=lfs diff=lfs merge=lfs -text +*.zip filter=lfs diff=lfs merge=lfs -text +*.zst filter=lfs diff=lfs merge=lfs -text +*tfevents* filter=lfs diff=lfs merge=lfs -text +language_model/atcosim_uwb_atcc_4g.binary filter=lfs diff=lfs merge=lfs -text diff --git a/README.md b/README.md new file mode 100644 index 0000000..96e0321 --- /dev/null +++ b/README.md @@ -0,0 +1,244 @@ +--- +language: en +license: apache-2.0 +tags: +- audio +- automatic-speech-recognition +- en-atc +- en +- generated_from_trainer +datasets: +- Jzuluaga/atcosim_corpus +- Jzuluaga/uwb_atcc +metrics: +- wer +base_model: facebook/wav2vec2-large-960h-lv60-self +model-index: +- name: wav2vec2-large-960h-lv60-self-en-atc-uwb-atcc-and-atcosim + results: + - task: + type: automatic-speech-recognition + name: Speech Recognition + dataset: + name: UWB-ATCC dataset (Air Traffic Control Communications) + type: Jzuluaga/uwb_atcc + config: test + split: test + metrics: + - type: wer + value: 17.48 + name: TEST WER + verified: false + - type: wer + value: 14.26 + name: TEST WER (+LM) + verified: false + - task: + type: automatic-speech-recognition + name: Speech Recognition + dataset: + name: ATCOSIM corpus (Air Traffic Control Communications) + type: Jzuluaga/atcosim_corpus + config: test + split: test + metrics: + - type: wer + value: 1.85 + name: TEST WER + verified: false + - type: wer + value: 1.13 + name: TEST WER (+LM) + verified: false +--- + + +This model is a fine-tuned version of [facebook/wav2vec2-large-960h-lv60-self](https://huggingface.co/facebook/wav2vec2-large-960h-lv60-self) on the EXPERIMENTS/DATA/ATCOSIM_UWB_ATCC/TRAIN - NA dataset. +It achieves the following results on the evaluation set: + +# wav2vec2-large-960h-lv60-self-en-atc-uwb-atcc-and-atcosim + +This model is a fine-tuned version of [facebook/wav2vec2-large-960h-lv60-self](https://huggingface.co/facebook/wav2vec2-large-960h-lv60-self) on two corpus: +- [UWB-ATCC corpus](https://huggingface.co/datasets/Jzuluaga/uwb_atcc), and +- [ATCOSIM corpus](https://huggingface.co/datasets/Jzuluaga/atcosim_corpus). + + + GitHub + + + GitHub + + + +It achieves the following results on the evaluation set (two tests sets joined together: UWB-ATCC and ATCOSIM): +- Loss: 0.4042 +- Wer: 0.1049 + +Paper: [How Does Pre-trained Wav2Vec 2.0 Perform on Domain Shifted ASR? An Extensive Benchmark on Air Traffic Control Communications](https://arxiv.org/abs/2203.16822). + +Authors: Juan Zuluaga-Gomez, Amrutha Prasad, Iuliia Nigmatulina, Saeed Sarfjoo, Petr Motlicek, Matthias Kleinert, Hartmut Helmke, Oliver Ohneiser, Qingran Zhan + +Abstract: Recent work on self-supervised pre-training focus on leveraging large-scale unlabeled speech data to build robust end-to-end (E2E)acoustic models (AM) that can be later fine-tuned on downstream tasks e.g., automatic speech recognition (ASR). Yet, few works investigated the impact on performance when the data properties substantially differ between the pre-training and fine-tuning phases, termed domain shift. We target this scenario by analyzing the robustness of Wav2Vec 2.0 and XLS-R models on downstream ASR for a completely unseen domain, air traffic control (ATC) communications. We benchmark these two models on several open-source and challenging ATC databases with signal-to-noise ratio between 5 and 20 dB. Relative word error rate (WER) reductions between 20% to 40% are obtained in comparison to hybrid-based ASR baselines by only fine-tuning E2E acoustic models with a smaller fraction of labeled data. We analyze WERs on the low-resource scenario and gender bias carried by one ATC dataset. + +Code — GitHub repository: https://github.com/idiap/w2v2-air-traffic + +## Usage + +You can use our Google Colab notebook to run and evaluate our model: https://github.com/idiap/w2v2-air-traffic/blob/master/src/eval_xlsr_atc_model.ipynb + +## Intended uses & limitations + +This model was fine-tuned on air traffic control data. We don't expect that it keeps the same performance on some others datasets, e.g., LibriSpeech or CommonVoice. + + +## Training and evaluation data + +See Table 1 (page 3) in our paper: [How Does Pre-trained Wav2Vec 2.0 Perform on Domain Shifted ASR? An Extensive Benchmark on Air Traffic Control Communications](https://arxiv.org/abs/2203.16822). We described there the partitions of how to use our model. + +- We use the UWB-ATCC + ATCOSIM corpus to fine-tune this model. You can download the raw data here: + - https://lindat.mff.cuni.cz/repository/xmlui/handle/11858/00-097C-0000-0001-CCA1-0 and, + - https://www.spsc.tugraz.at/databases-and-tools/atcosim-air-traffic-control-simulation-speech-corpus.html +- However, do not worry, we have prepared the database in `Datasets format`: + - Here, [UWB-ATCC corpus on HuggingFace](https://huggingface.co/datasets/Jzuluaga/uwb_atcc). + - Here: [ATCOSIM CORPUS on HuggingFace](https://huggingface.co/datasets/Jzuluaga/atcosim_corpus). +- If you want to prepare a database in HuggingFace format, you can follow the data loader script in: [data_loader_atc.py](https://huggingface.co/datasets/Jzuluaga/uwb_atcc/blob/main/atc_data_loader.py). + +## Writing your own inference script + +If you use language model, you need to install the KenLM bindings with: + +```bash +conda activate your_environment +pip install https://github.com/kpu/kenlm/archive/master.zip +``` + +The snippet of code: + +```python +from datasets import load_dataset, load_metric, Audio +import torch +from transformers import AutoModelForCTC, Wav2Vec2Processor, Wav2Vec2ProcessorWithLM +import torchaudio.functional as F + +USE_LM = False +DATASET_ID = "Jzuluaga/uwb_atcc" +MODEL_ID = "Jzuluaga/wav2vec2-large-960h-lv60-self-en-atc-uwb-atcc-and-atcosim" + +# 1. Load the dataset +# we only load the 'test' partition, however, if you want to load the 'train' partition, you can change it accordingly +uwb_atcc_corpus_test = load_dataset(DATASET_ID, "test", split="test") + +# 2. Load the model +model = AutoModelForCTC.from_pretrained(MODEL_ID) + +# 3. Load the processors, we offer support with LM, which should yield better resutls +if USE_LM: + processor = Wav2Vec2ProcessorWithLM.from_pretrained(MODEL_ID) +else: + processor = Wav2Vec2Processor.from_pretrained(MODEL_ID) +# 4. Format the test sample +sample = next(iter(uwb_atcc_corpus_test)) +file_sampling_rate = sample['audio']['sampling_rate'] +# resample if neccessary +if file_sampling_rate != 16000: + resampled_audio = F.resample(torch.tensor(sample["audio"]["array"]), file_sampling_rate, 16000).numpy() +else: + resampled_audio = torch.tensor(sample["audio"]["array"]).numpy() +input_values = processor(resampled_audio, return_tensors="pt").input_values + +# 5. Run the forward pass in the model +with torch.no_grad(): + logits = model(input_values).logits + +# get the transcription with processor +if USE_LM: + transcription = processor.batch_decode(logits.numpy()).text +else: + pred_ids = torch.argmax(logits, dim=-1) + transcription = processor.batch_decode(pred_ids) +# print the output +print(transcription) +``` + +# Cite us + +If you use this code for your research, please cite our paper with: + +``` +@article{zuluaga2022how, + title={How Does Pre-trained Wav2Vec2. 0 Perform on Domain Shifted ASR? An Extensive Benchmark on Air Traffic Control Communications}, + author={Zuluaga-Gomez, Juan and Prasad, Amrutha and Nigmatulina, Iuliia and Sarfjoo, Saeed and others}, + journal={IEEE Spoken Language Technology Workshop (SLT), Doha, Qatar}, + year={2022} + } +``` +and, + +``` +@article{zuluaga2022bertraffic, + title={BERTraffic: BERT-based Joint Speaker Role and Speaker Change Detection for Air Traffic Control Communications}, + author={Zuluaga-Gomez, Juan and Sarfjoo, Seyyed Saeed and Prasad, Amrutha and others}, + journal={IEEE Spoken Language Technology Workshop (SLT), Doha, Qatar}, + year={2022} + } +``` + +and, + +``` +@article{zuluaga2022atco2, + title={ATCO2 corpus: A Large-Scale Dataset for Research on Automatic Speech Recognition and Natural Language Understanding of Air Traffic Control Communications}, + author={Zuluaga-Gomez, Juan and Vesel{\`y}, Karel and Sz{\"o}ke, Igor and Motlicek, Petr and others}, + journal={arXiv preprint arXiv:2211.04054}, + year={2022} +} +``` + + +## Training procedure + +### Training hyperparameters + +The following hyperparameters were used during training: +- learning_rate: 0.0001 +- train_batch_size: 24 +- eval_batch_size: 12 +- seed: 42 +- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08 +- lr_scheduler_type: linear +- lr_scheduler_warmup_steps: 1000 +- training_steps: 10000 +- mixed_precision_training: Native AMP + +### Training results + +| Training Loss | Epoch | Step | Validation Loss | Wer | +|:-------------:|:-----:|:-----:|:---------------:|:------:| +| No log | 0.63 | 500 | 2.2638 | 0.9359 | +| 2.6089 | 1.27 | 1000 | 0.7277 | 0.2407 | +| 2.6089 | 1.9 | 1500 | 0.5800 | 0.1745 | +| 0.6019 | 2.53 | 2000 | 0.4887 | 0.1514 | +| 0.6019 | 3.17 | 2500 | 0.4666 | 0.1421 | +| 0.4722 | 3.8 | 3000 | 0.4426 | 0.1451 | +| 0.4722 | 4.44 | 3500 | 0.4176 | 0.1248 | +| 0.4278 | 5.07 | 4000 | 0.4365 | 0.1239 | +| 0.4278 | 5.7 | 4500 | 0.3816 | 0.1177 | +| 0.369 | 6.34 | 5000 | 0.4113 | 0.1172 | +| 0.369 | 6.97 | 5500 | 0.3863 | 0.1230 | +| 0.341 | 7.6 | 6000 | 0.3850 | 0.1116 | +| 0.341 | 8.24 | 6500 | 0.4014 | 0.1141 | +| 0.3119 | 8.87 | 7000 | 0.3953 | 0.1078 | +| 0.3119 | 9.51 | 7500 | 0.4018 | 0.1080 | +| 0.3008 | 10.14 | 8000 | 0.3964 | 0.1074 | +| 0.3008 | 10.77 | 8500 | 0.3917 | 0.1078 | +| 0.2741 | 11.41 | 9000 | 0.3961 | 0.1057 | +| 0.2741 | 12.04 | 9500 | 0.3974 | 0.1053 | +| 0.2531 | 12.67 | 10000 | 0.4042 | 0.1049 | + + +### Framework versions + +- Transformers 4.24.0 +- Pytorch 1.13.0+cu117 +- Datasets 2.6.1 +- Tokenizers 0.13.2 diff --git a/added_tokens.json b/added_tokens.json new file mode 100644 index 0000000..cc95e40 --- /dev/null +++ b/added_tokens.json @@ -0,0 +1,4 @@ +{ + "": 30, + "": 29 +} diff --git a/all_results.json b/all_results.json new file mode 100644 index 0000000..ef41f80 --- /dev/null +++ b/all_results.json @@ -0,0 +1,14 @@ +{ + "epoch": 12.67, + "eval_loss": 0.40417608618736267, + "eval_runtime": 163.5465, + "eval_samples": 4723, + "eval_samples_per_second": 28.879, + "eval_steps_per_second": 2.409, + "eval_wer": 0.10485405449172099, + "train_loss": 0.5960800704956055, + "train_runtime": 13155.4001, + "train_samples": 18925, + "train_samples_per_second": 18.243, + "train_steps_per_second": 0.76 +} \ No newline at end of file diff --git a/alphabet.json b/alphabet.json new file mode 100644 index 0000000..84c78d8 --- /dev/null +++ b/alphabet.json @@ -0,0 +1 @@ +{"labels": [" ", "a", "b", "c", "d", "e", "f", "g", "h", "i", "j", "k", "l", "m", "n", "o", "p", "q", "r", "s", "t", "u", "v", "w", "x", "y", "z", "\u2047", "", "", ""], "is_bpe": false} \ No newline at end of file diff --git a/config.json b/config.json new file mode 100644 index 0000000..f0cf319 --- /dev/null +++ b/config.json @@ -0,0 +1,108 @@ +{ + "_name_or_path": "facebook/wav2vec2-large-960h-lv60-self", + "activation_dropout": 0.0, + "adapter_kernel_size": 3, + "adapter_stride": 2, + "add_adapter": false, + "apply_spec_augment": true, + "architectures": [ + "Wav2Vec2ForCTC" + ], + "attention_dropout": 0.0, + "bos_token_id": 1, + "classifier_proj_size": 256, + "codevector_dim": 256, + "contrastive_logits_temperature": 0.1, + "conv_bias": true, + "conv_dim": [ + 512, + 512, + 512, + 512, + 512, + 512, + 512 + ], + "conv_kernel": [ + 10, + 3, + 3, + 3, + 3, + 2, + 2 + ], + "conv_stride": [ + 5, + 2, + 2, + 2, + 2, + 2, + 2 + ], + "ctc_loss_reduction": "mean", + "ctc_zero_infinity": true, + "diversity_loss_weight": 0.1, + "do_stable_layer_norm": true, + "eos_token_id": 2, + "feat_extract_activation": "gelu", + "feat_extract_dropout": 0.0, + "feat_extract_norm": "layer", + "feat_proj_dropout": 0.05, + "feat_quantizer_dropout": 0.0, + "final_dropout": 0.0, + "hidden_act": "gelu", + "hidden_dropout": 0.0, + "hidden_dropout_prob": 0.1, + "hidden_size": 1024, + "initializer_range": 0.02, + "intermediate_size": 4096, + "layer_norm_eps": 1e-05, + "layerdrop": 0.0, + "mask_feature_length": 12, + "mask_feature_min_masks": 0, + "mask_feature_prob": 0.0, + "mask_time_length": 12, + "mask_time_min_masks": 2, + "mask_time_prob": 0.075, + "model_type": "wav2vec2", + "num_adapter_layers": 3, + "num_attention_heads": 16, + "num_codevector_groups": 2, + "num_codevectors_per_group": 320, + "num_conv_pos_embedding_groups": 16, + "num_conv_pos_embeddings": 128, + "num_feat_extract_layers": 7, + "num_hidden_layers": 24, + "num_negatives": 100, + "output_hidden_size": 1024, + "pad_token_id": 28, + "proj_codevector_dim": 256, + "tdnn_dilation": [ + 1, + 2, + 3, + 1, + 1 + ], + "tdnn_dim": [ + 512, + 512, + 512, + 512, + 1500 + ], + "tdnn_kernel": [ + 5, + 3, + 3, + 1, + 1 + ], + "torch_dtype": "float32", + "transformers_version": "4.24.0", + "use_weighted_layer_sum": false, + "vocab_size": 31, + "xvector_output_dim": 512 +} diff --git a/eval_results.json b/eval_results.json new file mode 100644 index 0000000..9cb4148 --- /dev/null +++ b/eval_results.json @@ -0,0 +1,9 @@ +{ + "epoch": 12.67, + "eval_loss": 0.40417608618736267, + "eval_runtime": 163.5465, + "eval_samples": 4723, + "eval_samples_per_second": 28.879, + "eval_steps_per_second": 2.409, + "eval_wer": 0.10485405449172099 +} \ No newline at end of file diff --git a/language_model/atcosim_uwb_atcc_4g.binary b/language_model/atcosim_uwb_atcc_4g.binary new file mode 100644 index 0000000..b1ae401 --- /dev/null +++ b/language_model/atcosim_uwb_atcc_4g.binary @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:974c8f8db889c30ac4060a741144847c7b71cad0a8bd6714d81a265794b235d0 +size 1360621 diff --git a/language_model/attrs.json b/language_model/attrs.json new file mode 100644 index 0000000..3c07595 --- /dev/null +++ b/language_model/attrs.json @@ -0,0 +1 @@ +{"alpha": 0.5, "beta": 1.5, "unk_score_offset": -10.0, "score_boundary": true} \ No newline at end of file diff --git a/language_model/unigrams.txt b/language_model/unigrams.txt new file mode 100644 index 0000000..e69de29 diff --git a/log/eval_model_train b/log/eval_model_train new file mode 100644 index 0000000..224f7b4 --- /dev/null +++ b/log/eval_model_train @@ -0,0 +1,15 @@ +# Running on vgnf007 +# Started at Wed 30 Nov 07:36:30 CET 2022 +# python3 src/eval_model.py --language-model experiments/data/atcosim_uwb_atcc/train/lm/atcosim_uwb_atcc_4g.binary --pretrained-model experiments/results/bs_exp/linear/wav2vec2-large-960h-lv60-self/atcosim_uwb_atcc/0.0ld_0.0ad_0.0attd_0.05fpd_0.075mtp_12mtl_0.0mfp_12mfl_1acc/ --print-output true --test-set experiments/data/atcosim_corpus/train +Integrating a LM by shallow fusion, results should be better +*** Loading the Wav2Vec 2.0 model, loading... *** +Traceback (most recent call last): + File "/remote/idiap.svm/temp.speech01/jzuluaga/experiments/journal/asr/github/w2v2-air-traffic/src/eval_model.py", line 250, in + main() + File "/remote/idiap.svm/temp.speech01/jzuluaga/experiments/journal/asr/github/w2v2-air-traffic/src/eval_model.py", line 152, in main + processor, processor_ctc_kenlm, model = get_kenlm_processor(path_model, path_lm) + File "/remote/idiap.svm/temp.speech01/jzuluaga/experiments/journal/asr/github/w2v2-air-traffic/src/eval_model.py", line 65, in get_kenlm_processor + path_tokenizer + "/vocab.json", +TypeError: unsupported operand type(s) for +: 'PosixPath' and 'str' +# Accounting: time=25 threads=1 +# Finished at Wed 30 Nov 07:36:55 CET 2022 with status 1 diff --git a/log/train_log b/log/train_log new file mode 100644 index 0000000..419a05c --- /dev/null +++ b/log/train_log @@ -0,0 +1,21388 @@ +# Running on vgni027 +# Started at Mon 28 Nov 08:09:01 CET 2022 +# python3 src/run_speech_recognition_ctc.py --report_to=none --run_name=experiments/results/bs_exp/linear/wav2vec2-large-960h-lv60-self/atcosim_uwb_atcc/0.0ld_0.0ad_0.0attd_0.05fpd_0.075mtp_12mtl_0.0mfp_12mfl_1acc/ --preprocessing_num_workers=5 --model_name_or_path=facebook/wav2vec2-large-960h-lv60-self --dataset_name=experiments/data/atcosim_uwb_atcc/train --min_duration_in_seconds=0.2 --max_duration_in_seconds=20 --eval_dataset_name=experiments/data/atcosim_uwb_atcc/test --train_split_name=train --output_dir=experiments/results/bs_exp/linear/wav2vec2-large-960h-lv60-self/atcosim_uwb_atcc/0.0ld_0.0ad_0.0attd_0.05fpd_0.075mtp_12mtl_0.0mfp_12mfl_1acc/ --num_train_epochs=50 --per_device_train_batch_size=24 --per_device_eval_batch_size=12 --gradient_accumulation_steps=1 --learning_rate=1e-4 --weight_decay=0.001 --warmup_steps=1000 --evaluation_strategy=steps --text_column_name=text --audio_column_name=audio --length_column_name=input_length '--chars_to_ignore=, ? . ! \; \: " “ % ‘ ” � — ’ … –' --save_steps=1000 --eval_steps=500 --logging_steps=1000 --layerdrop=0.0 --activation_dropout=0.0 --attention_dropout=0.0 --save_total_limit=1 --feat_proj_dropout=0.05 --mask_time_prob=0.075 --mask_time_length=12 --mask_feature_prob=0.0 --mask_feature_length=12 --gradient_checkpointing --freeze_feature_encoder --fp16 --group_by_length --do_train --do_eval --max_steps 10000 --overwrite_output_dir --freeze_feature_encoder +11/28/2022 08:09:10 - WARNING - __main__ - Process rank: -1, device: cuda:0, n_gpu: 1distributed training: False, 16-bits training: True +11/28/2022 08:09:10 - INFO - __main__ - Training/evaluation parameters TrainingArguments( +_n_gpu=1, +adafactor=False, +adam_beta1=0.9, +adam_beta2=0.999, +adam_epsilon=1e-08, +auto_find_batch_size=False, +bf16=False, +bf16_full_eval=False, +data_seed=None, +dataloader_drop_last=False, +dataloader_num_workers=0, +dataloader_pin_memory=True, +ddp_bucket_cap_mb=None, +ddp_find_unused_parameters=None, +ddp_timeout=1800, +debug=[], +deepspeed=None, +disable_tqdm=False, +do_eval=True, +do_predict=False, +do_train=True, +eval_accumulation_steps=None, +eval_delay=0, +eval_steps=500, +evaluation_strategy=steps, +fp16=True, +fp16_backend=auto, +fp16_full_eval=False, +fp16_opt_level=O1, +fsdp=[], +fsdp_min_num_params=0, +fsdp_transformer_layer_cls_to_wrap=None, +full_determinism=False, +gradient_accumulation_steps=1, +gradient_checkpointing=True, +greater_is_better=None, +group_by_length=True, +half_precision_backend=auto, +hub_model_id=None, +hub_private_repo=False, +hub_strategy=every_save, +hub_token=, +ignore_data_skip=False, +include_inputs_for_metrics=False, +jit_mode_eval=False, +label_names=None, +label_smoothing_factor=0.0, +learning_rate=0.0001, +length_column_name=input_length, +load_best_model_at_end=False, +local_rank=-1, +log_level=passive, +log_level_replica=passive, +log_on_each_node=True, +logging_dir=experiments/results/bs_exp/linear/wav2vec2-large-960h-lv60-self/atcosim_uwb_atcc/0.0ld_0.0ad_0.0attd_0.05fpd_0.075mtp_12mtl_0.0mfp_12mfl_1acc/runs/Nov28_08-09-10_vgni027, +logging_first_step=False, +logging_nan_inf_filter=True, +logging_steps=1000, +logging_strategy=steps, +lr_scheduler_type=linear, +max_grad_norm=1.0, +max_steps=10000, +metric_for_best_model=None, +mp_parameters=, +no_cuda=False, +num_train_epochs=50.0, +optim=adamw_hf, +output_dir=experiments/results/bs_exp/linear/wav2vec2-large-960h-lv60-self/atcosim_uwb_atcc/0.0ld_0.0ad_0.0attd_0.05fpd_0.075mtp_12mtl_0.0mfp_12mfl_1acc/, +overwrite_output_dir=True, +past_index=-1, +per_device_eval_batch_size=12, +per_device_train_batch_size=24, +prediction_loss_only=False, +push_to_hub=False, +push_to_hub_model_id=None, +push_to_hub_organization=None, +push_to_hub_token=, +ray_scope=last, +remove_unused_columns=True, +report_to=[], +resume_from_checkpoint=None, +run_name=experiments/results/bs_exp/linear/wav2vec2-large-960h-lv60-self/atcosim_uwb_atcc/0.0ld_0.0ad_0.0attd_0.05fpd_0.075mtp_12mtl_0.0mfp_12mfl_1acc/, +save_on_each_node=False, +save_steps=1000, +save_strategy=steps, +save_total_limit=1, +seed=42, +sharded_ddp=[], +skip_memory_metrics=True, +tf32=None, +torchdynamo=None, +tpu_metrics_debug=False, +tpu_num_cores=None, +use_ipex=False, +use_legacy_prediction_loop=False, +use_mps_device=False, +warmup_ratio=0.0, +warmup_steps=1000, +weight_decay=0.001, +xpu_backend=None, +) +11/28/2022 08:09:10 - WARNING - datasets.builder - Using custom data configuration train-229a51d735ef0a2b +Downloading and preparing dataset atc_data_loader/train to /remote/idiap.svm/temp.speech01/jzuluaga/experiments/journal/asr/github/w2v2-air-traffic/.cache/experiments/results/bs_exp/linear/wav2vec2-large-960h-lv60-self/atcosim_uwb_atcc/0.0ld_0.0ad_0.0attd_0.05fpd_0.075mtp_12mtl_0.0mfp_12mfl_1acc//train/atc_data_loader/train-229a51d735ef0a2b/0.0.0/f2633cc53c6abe32cddd4152eebde1a4e3c9953e1446e190b8d9a13330cddaa4... + Generating train split: 0 examples [00:00, ? examples/s] Generating train split: 1 examples [00:53, 53.46s/ examples] Generating train split: 256 examples [00:53, 6.77 examples/s] Generating train split: 512 examples [00:54, 16.30 examples/s] Generating train split: 768 examples [00:54, 29.69 examples/s] Generating train split: 1024 examples [00:54, 48.26 examples/s] Generating train split: 1280 examples [00:54, 73.60 examples/s] Generating train split: 1536 examples [00:54, 108.14 examples/s] Generating train split: 1792 examples [00:55, 153.96 examples/s] Generating train split: 2048 examples [00:55, 214.00 examples/s] Generating train split: 2304 examples [00:55, 287.75 examples/s] Generating train split: 2560 examples [00:55, 375.64 examples/s] Generating train split: 2816 examples [00:56, 463.83 examples/s] Generating train split: 3072 examples [00:56, 570.48 examples/s] Generating train split: 3328 examples [00:56, 663.67 examples/s] Generating train split: 3584 examples [00:56, 783.88 examples/s] Generating train split: 3840 examples [00:56, 882.22 examples/s] Generating train split: 4096 examples [00:57, 926.29 examples/s] Generating train split: 4352 examples [00:57, 994.41 examples/s] Generating train split: 4608 examples [00:57, 1029.52 examples/s] Generating train split: 4864 examples [00:57, 1022.38 examples/s] Generating train split: 5120 examples [00:58, 1078.20 examples/s] Generating train split: 5376 examples [00:58, 1115.61 examples/s] Generating train split: 5632 examples [00:58, 1118.35 examples/s] Generating train split: 5888 examples [00:58, 1155.38 examples/s] Generating train split: 6144 examples [00:58, 1200.64 examples/s] Generating train split: 6400 examples [00:59, 1166.52 examples/s] Generating train split: 6656 examples [00:59, 1158.81 examples/s] Generating train split: 6912 examples [00:59, 1180.90 examples/s] Generating train split: 7168 examples [00:59, 1154.28 examples/s] Generating train split: 7424 examples [00:59, 1156.77 examples/s] Generating train split: 7680 examples [01:00, 1208.85 examples/s] Generating train split: 8144 examples [01:00, 1786.32 examples/s] Generating train split: 8533 examples [01:00, 2193.75 examples/s] Generating train split: 8960 examples [01:00, 2520.73 examples/s] Generating train split: 9459 examples [01:00, 3083.36 examples/s] Generating train split: 9816 examples [01:00, 3110.19 examples/s] Generating train split: 10240 examples [01:00, 3300.45 examples/s] Generating train split: 10752 examples [01:00, 3521.50 examples/s] Generating train split: 11264 examples [01:01, 3646.93 examples/s] Generating train split: 11776 examples [01:01, 3720.64 examples/s] Generating train split: 12261 examples [01:01, 4003.74 examples/s] Generating train split: 12673 examples [01:01, 3919.00 examples/s] Generating train split: 13073 examples [01:01, 3838.18 examples/s] Generating train split: 13568 examples [01:01, 3925.92 examples/s] Generating train split: 14076 examples [01:01, 4235.56 examples/s] Generating train split: 14525 examples [01:01, 4305.79 examples/s] Generating train split: 14961 examples [01:01, 4162.99 examples/s] Generating train split: 15381 examples [01:02, 4044.01 examples/s] Generating train split: 15872 examples [01:02, 4103.73 examples/s] Generating train split: 16376 examples [01:02, 4361.38 examples/s] Generating train split: 16816 examples [01:02, 4261.34 examples/s] Generating train split: 17245 examples [01:02, 4044.89 examples/s] Generating train split: 17664 examples [01:02, 3868.00 examples/s] Generating train split: 18153 examples [01:02, 4145.65 examples/s] Generating train split: 18573 examples [01:02, 3889.75 examples/s] Dataset atc_data_loader downloaded and prepared to /remote/idiap.svm/temp.speech01/jzuluaga/experiments/journal/asr/github/w2v2-air-traffic/.cache/experiments/results/bs_exp/linear/wav2vec2-large-960h-lv60-self/atcosim_uwb_atcc/0.0ld_0.0ad_0.0attd_0.05fpd_0.075mtp_12mtl_0.0mfp_12mfl_1acc//train/atc_data_loader/train-229a51d735ef0a2b/0.0.0/f2633cc53c6abe32cddd4152eebde1a4e3c9953e1446e190b8d9a13330cddaa4. Subsequent calls will reuse this data. +11/28/2022 08:10:24 - WARNING - datasets.builder - Using custom data configuration test-b7cef58c6dbd7d6c +Downloading and preparing dataset atc_data_loader/test to /remote/idiap.svm/temp.speech01/jzuluaga/experiments/journal/asr/github/w2v2-air-traffic/.cache/experiments/results/bs_exp/linear/wav2vec2-large-960h-lv60-self/atcosim_uwb_atcc/0.0ld_0.0ad_0.0attd_0.05fpd_0.075mtp_12mtl_0.0mfp_12mfl_1acc//test/atc_data_loader/test-b7cef58c6dbd7d6c/0.0.0/f2633cc53c6abe32cddd4152eebde1a4e3c9953e1446e190b8d9a13330cddaa4... + Generating test split: 0 examples [00:00, ? examples/s] Generating test split: 1 examples [00:12, 12.03s/ examples] Generating test split: 256 examples [00:12, 29.51 examples/s] Generating test split: 512 examples [00:12, 68.80 examples/s] Generating test split: 768 examples [00:12, 121.23 examples/s] Generating test split: 1024 examples [00:12, 188.38 examples/s] Generating test split: 1280 examples [00:13, 270.78 examples/s] Generating test split: 1536 examples [00:13, 362.18 examples/s] Generating test split: 1792 examples [00:13, 462.16 examples/s] Generating test split: 2048 examples [00:13, 613.65 examples/s] Generating test split: 2480 examples [00:13, 977.92 examples/s] Generating test split: 2816 examples [00:13, 1248.52 examples/s] Generating test split: 3316 examples [00:14, 1800.86 examples/s] Generating test split: 3658 examples [00:14, 2081.76 examples/s] Generating test split: 4096 examples [00:14, 2458.65 examples/s] Generating test split: 4591 examples [00:14, 2996.04 examples/s] loading configuration file config.json from cache at /idiap/temp/jzuluaga/cache/huggingface/models--facebook--wav2vec2-large-960h-lv60-self/snapshots/54074b1c16f4de6a5ad59affb4caa8f2ea03a119/config.json +Model config Wav2Vec2Config { + "_name_or_path": "facebook/wav2vec2-large-960h-lv60-self", + "activation_dropout": 0.1, + "adapter_kernel_size": 3, + "adapter_stride": 2, + "add_adapter": false, + "apply_spec_augment": true, + "architectures": [ + "Wav2Vec2ForCTC" + ], + "attention_dropout": 0.1, + "bos_token_id": 1, + "classifier_proj_size": 256, + "codevector_dim": 256, + "contrastive_logits_temperature": 0.1, + "conv_bias": true, + "conv_dim": [ + 512, + 512, + 512, + 512, + 512, + 512, + 512 + ], + "conv_kernel": [ + 10, + 3, + 3, + 3, + 3, + 2, + 2 + ], + "conv_stride": [ + 5, + 2, + 2, + 2, + 2, + 2, + 2 + ], + "ctc_loss_reduction": "sum", + "ctc_zero_infinity": false, + "diversity_loss_weight": 0.1, + "do_stable_layer_norm": true, + "eos_token_id": 2, + "feat_extract_activation": "gelu", + "feat_extract_dropout": 0.0, + "feat_extract_norm": "layer", + "feat_proj_dropout": 0.1, + "feat_quantizer_dropout": 0.0, + "final_dropout": 0.1, + "gradient_checkpointing": false, + "hidden_act": "gelu", + "hidden_dropout": 0.1, + "hidden_dropout_prob": 0.1, + "hidden_size": 1024, + "initializer_range": 0.02, + "intermediate_size": 4096, + "layer_norm_eps": 1e-05, + "layerdrop": 0.1, + "mask_feature_length": 10, + "mask_feature_min_masks": 0, + "mask_feature_prob": 0.0, + "mask_time_length": 10, + "mask_time_min_masks": 2, + "mask_time_prob": 0.05, + "model_type": "wav2vec2", + "num_adapter_layers": 3, + "num_attention_heads": 16, + "num_codevector_groups": 2, + "num_codevectors_per_group": 320, + "num_conv_pos_embedding_groups": 16, + "num_conv_pos_embeddings": 128, + "num_feat_extract_layers": 7, + "num_hidden_layers": 24, + "num_negatives": 100, + "output_hidden_size": 1024, + "pad_token_id": 0, + "proj_codevector_dim": 256, + "tdnn_dilation": [ + 1, + 2, + 3, + 1, + 1 + ], + "tdnn_dim": [ + 512, + 512, + 512, + 512, + 1500 + ], + "tdnn_kernel": [ + 5, + 3, + 3, + 1, + 1 + ], + "transformers_version": "4.24.0", + "use_weighted_layer_sum": false, + "vocab_size": 32, + "xvector_output_dim": 512 +} + +Dataset atc_data_loader downloaded and prepared to /remote/idiap.svm/temp.speech01/jzuluaga/experiments/journal/asr/github/w2v2-air-traffic/.cache/experiments/results/bs_exp/linear/wav2vec2-large-960h-lv60-self/atcosim_uwb_atcc/0.0ld_0.0ad_0.0attd_0.05fpd_0.075mtp_12mtl_0.0mfp_12mfl_1acc//test/atc_data_loader/test-b7cef58c6dbd7d6c/0.0.0/f2633cc53c6abe32cddd4152eebde1a4e3c9953e1446e190b8d9a13330cddaa4. Subsequent calls will reuse this data. + 0%| | 0/1 [00:00 to the vocabulary +Adding to the vocabulary +Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained. +loading configuration file preprocessor_config.json from cache at /idiap/temp/jzuluaga/cache/huggingface/models--facebook--wav2vec2-large-960h-lv60-self/snapshots/54074b1c16f4de6a5ad59affb4caa8f2ea03a119/preprocessor_config.json +loading configuration file config.json from cache at /idiap/temp/jzuluaga/cache/huggingface/models--facebook--wav2vec2-large-960h-lv60-self/snapshots/54074b1c16f4de6a5ad59affb4caa8f2ea03a119/config.json +Model config Wav2Vec2Config { + "_name_or_path": "facebook/wav2vec2-large-960h-lv60-self", + "activation_dropout": 0.1, + "adapter_kernel_size": 3, + "adapter_stride": 2, + "add_adapter": false, + "apply_spec_augment": true, + "architectures": [ + "Wav2Vec2ForCTC" + ], + "attention_dropout": 0.1, + "bos_token_id": 1, + "classifier_proj_size": 256, + "codevector_dim": 256, + "contrastive_logits_temperature": 0.1, + "conv_bias": true, + "conv_dim": [ + 512, + 512, + 512, + 512, + 512, + 512, + 512 + ], + "conv_kernel": [ + 10, + 3, + 3, + 3, + 3, + 2, + 2 + ], + "conv_stride": [ + 5, + 2, + 2, + 2, + 2, + 2, + 2 + ], + "ctc_loss_reduction": "sum", + "ctc_zero_infinity": false, + "diversity_loss_weight": 0.1, + "do_stable_layer_norm": true, + "eos_token_id": 2, + "feat_extract_activation": "gelu", + "feat_extract_dropout": 0.0, + "feat_extract_norm": "layer", + "feat_proj_dropout": 0.1, + "feat_quantizer_dropout": 0.0, + "final_dropout": 0.1, + "gradient_checkpointing": false, + "hidden_act": "gelu", + "hidden_dropout": 0.1, + "hidden_dropout_prob": 0.1, + "hidden_size": 1024, + "initializer_range": 0.02, + "intermediate_size": 4096, + "layer_norm_eps": 1e-05, + "layerdrop": 0.1, + "mask_feature_length": 10, + "mask_feature_min_masks": 0, + "mask_feature_prob": 0.0, + "mask_time_length": 10, + "mask_time_min_masks": 2, + "mask_time_prob": 0.05, + "model_type": "wav2vec2", + "num_adapter_layers": 3, + "num_attention_heads": 16, + "num_codevector_groups": 2, + "num_codevectors_per_group": 320, + "num_conv_pos_embedding_groups": 16, + "num_conv_pos_embeddings": 128, + "num_feat_extract_layers": 7, + "num_hidden_layers": 24, + "num_negatives": 100, + "output_hidden_size": 1024, + "pad_token_id": 0, + "proj_codevector_dim": 256, + "tdnn_dilation": [ + 1, + 2, + 3, + 1, + 1 + ], + "tdnn_dim": [ + 512, + 512, + 512, + 512, + 1500 + ], + "tdnn_kernel": [ + 5, + 3, + 3, + 1, + 1 + ], + "transformers_version": "4.24.0", + "use_weighted_layer_sum": false, + "vocab_size": 32, + "xvector_output_dim": 512 +} + +Feature extractor Wav2Vec2FeatureExtractor { + "do_normalize": true, + "feature_extractor_type": "Wav2Vec2FeatureExtractor", + "feature_size": 1, + "padding_side": "right", + "padding_value": 0.0, + "return_attention_mask": true, + "sampling_rate": 16000 +} + +loading weights file pytorch_model.bin from cache at /idiap/temp/jzuluaga/cache/huggingface/models--facebook--wav2vec2-large-960h-lv60-self/snapshots/54074b1c16f4de6a5ad59affb4caa8f2ea03a119/pytorch_model.bin +All model checkpoint weights were used when initializing Wav2Vec2ForCTC. + +Some weights of Wav2Vec2ForCTC were not initialized from the model checkpoint at facebook/wav2vec2-large-960h-lv60-self and are newly initialized: ['wav2vec2.masked_spec_embed'] +You should probably TRAIN this model on a down-stream task to be able to use it for predictions and inference. +Some weights of Wav2Vec2ForCTC were not initialized from the model checkpoint at facebook/wav2vec2-large-960h-lv60-self and are newly initialized because the shapes did not match: +- lm_head.weight: found shape torch.Size([32, 1024]) in the checkpoint and torch.Size([31, 1024]) in the model instantiated +- lm_head.bias: found shape torch.Size([32]) in the checkpoint and torch.Size([31]) in the model instantiated +You should probably TRAIN this model on a down-stream task to be able to use it for predictions and inference. + preprocess datasets #0: 0%| | 0/3786 [00:00 to the vocabulary +Adding to the vocabulary +max_steps is given, it will override any value given in num_train_epochs +Using cuda_amp half precision backend +The following columns in the training set don't have a corresponding argument in `Wav2Vec2ForCTC.forward` and have been ignored: input_length. If input_length are not expected by `Wav2Vec2ForCTC.forward`, you can safely ignore this message. +/idiap/user/jzuluaga/miniconda3/envs/w2v2/lib/python3.10/site-packages/transformers/optimization.py:306: FutureWarning: This implementation of AdamW is deprecated and will be removed in a future version. Use the PyTorch implementation torch.optim.AdamW instead, or set `no_deprecation_warning=True` to disable this warning + warnings.warn( +***** Running training ***** + Num examples = 18925 + Num Epochs = 13 + Instantaneous batch size per device = 24 + Total train batch size (w. parallel, distributed & accumulation) = 24 + Gradient Accumulation steps = 1 + Total optimization steps = 10000 + Number of trainable parameters = 311260319 + 0%| | 0/10000 [00:00", + "eos_token": "", + "pad_token": "[PAD]", + "unk_token": "[UNK]" +} diff --git a/tokenizer_config.json b/tokenizer_config.json new file mode 100644 index 0000000..3cb942d --- /dev/null +++ b/tokenizer_config.json @@ -0,0 +1,11 @@ +{ + "bos_token": "", + "do_lower_case": false, + "eos_token": "", + "pad_token": "[PAD]", + "processor_class": "Wav2Vec2ProcessorWithLM", + "replace_word_delimiter_char": " ", + "tokenizer_class": "Wav2Vec2CTCTokenizer", + "unk_token": "[UNK]", + "word_delimiter_token": "|" +} diff --git a/train_results.json b/train_results.json new file mode 100644 index 0000000..23e5b88 --- /dev/null +++ b/train_results.json @@ -0,0 +1,8 @@ +{ + "epoch": 12.67, + "train_loss": 0.5960800704956055, + "train_runtime": 13155.4001, + "train_samples": 18925, + "train_samples_per_second": 18.243, + "train_steps_per_second": 0.76 +} \ No newline at end of file diff --git a/trainer_state.json b/trainer_state.json new file mode 100644 index 0000000..a7bbb43 --- /dev/null +++ b/trainer_state.json @@ -0,0 +1,265 @@ +{ + "best_metric": null, + "best_model_checkpoint": null, + "epoch": 12.67427122940431, + "global_step": 10000, + "is_hyper_param_search": false, + "is_local_process_zero": true, + "is_world_process_zero": true, + "log_history": [ + { + "epoch": 0.63, + "eval_loss": 2.2638113498687744, + "eval_runtime": 149.9992, + "eval_samples_per_second": 31.487, + "eval_steps_per_second": 2.627, + "eval_wer": 0.9359337678636492, + "step": 500 + }, + { + "epoch": 1.27, + "learning_rate": 9.970000000000001e-05, + "loss": 2.6089, + "step": 1000 + }, + { + "epoch": 1.27, + "eval_loss": 0.7277476787567139, + "eval_runtime": 148.226, + "eval_samples_per_second": 31.864, + "eval_steps_per_second": 2.658, + "eval_wer": 0.24067851503987692, + "step": 1000 + }, + { + "epoch": 1.9, + "eval_loss": 0.5800275802612305, + "eval_runtime": 148.0487, + "eval_samples_per_second": 31.902, + "eval_steps_per_second": 2.661, + "eval_wer": 0.1745475891664305, + "step": 1500 + }, + { + "epoch": 2.53, + "learning_rate": 8.892222222222223e-05, + "loss": 0.6019, + "step": 2000 + }, + { + "epoch": 2.53, + "eval_loss": 0.48867326974868774, + "eval_runtime": 148.4526, + "eval_samples_per_second": 31.815, + "eval_steps_per_second": 2.654, + "eval_wer": 0.15135014776729688, + "step": 2000 + }, + { + "epoch": 3.17, + "eval_loss": 0.466572642326355, + "eval_runtime": 148.0238, + "eval_samples_per_second": 31.907, + "eval_steps_per_second": 2.662, + "eval_wer": 0.14213999433221328, + "step": 2500 + }, + { + "epoch": 3.8, + "learning_rate": 7.78111111111111e-05, + "loss": 0.4722, + "step": 3000 + }, + { + "epoch": 3.8, + "eval_loss": 0.44257038831710815, + "eval_runtime": 149.7003, + "eval_samples_per_second": 31.55, + "eval_steps_per_second": 2.632, + "eval_wer": 0.14505485607870128, + "step": 3000 + }, + { + "epoch": 4.44, + "eval_loss": 0.41759932041168213, + "eval_runtime": 149.0588, + "eval_samples_per_second": 31.685, + "eval_steps_per_second": 2.643, + "eval_wer": 0.12481276061697907, + "step": 3500 + }, + { + "epoch": 5.07, + "learning_rate": 6.671111111111111e-05, + "loss": 0.4278, + "step": 4000 + }, + { + "epoch": 5.07, + "eval_loss": 0.4364745318889618, + "eval_runtime": 148.5505, + "eval_samples_per_second": 31.794, + "eval_steps_per_second": 2.652, + "eval_wer": 0.12388162422573985, + "step": 4000 + }, + { + "epoch": 5.7, + "eval_loss": 0.3815610408782959, + "eval_runtime": 149.3382, + "eval_samples_per_second": 31.626, + "eval_steps_per_second": 2.638, + "eval_wer": 0.1177280272053763, + "step": 4500 + }, + { + "epoch": 6.34, + "learning_rate": 5.560000000000001e-05, + "loss": 0.369, + "step": 5000 + }, + { + "epoch": 6.34, + "eval_loss": 0.4113306403160095, + "eval_runtime": 159.7075, + "eval_samples_per_second": 29.573, + "eval_steps_per_second": 2.467, + "eval_wer": 0.11716124853244808, + "step": 5000 + }, + { + "epoch": 6.97, + "eval_loss": 0.3862614035606384, + "eval_runtime": 152.4609, + "eval_samples_per_second": 30.978, + "eval_steps_per_second": 2.584, + "eval_wer": 0.1230112141208858, + "step": 5500 + }, + { + "epoch": 7.6, + "learning_rate": 4.448888888888889e-05, + "loss": 0.341, + "step": 6000 + }, + { + "epoch": 7.6, + "eval_loss": 0.384976863861084, + "eval_runtime": 159.7493, + "eval_samples_per_second": 29.565, + "eval_steps_per_second": 2.466, + "eval_wer": 0.1116149143759362, + "step": 6000 + }, + { + "epoch": 8.24, + "eval_loss": 0.401400089263916, + "eval_runtime": 155.917, + "eval_samples_per_second": 30.292, + "eval_steps_per_second": 2.527, + "eval_wer": 0.11406420792680458, + "step": 6500 + }, + { + "epoch": 8.87, + "learning_rate": 3.337777777777778e-05, + "loss": 0.3119, + "step": 7000 + }, + { + "epoch": 8.87, + "eval_loss": 0.39530250430107117, + "eval_runtime": 165.7417, + "eval_samples_per_second": 28.496, + "eval_steps_per_second": 2.377, + "eval_wer": 0.10782964252459415, + "step": 7000 + }, + { + "epoch": 9.51, + "eval_loss": 0.4018384516239166, + "eval_runtime": 163.4263, + "eval_samples_per_second": 28.9, + "eval_steps_per_second": 2.411, + "eval_wer": 0.10801182138374965, + "step": 7500 + }, + { + "epoch": 10.14, + "learning_rate": 2.2277777777777778e-05, + "loss": 0.3008, + "step": 8000 + }, + { + "epoch": 10.14, + "eval_loss": 0.3963571786880493, + "eval_runtime": 172.5171, + "eval_samples_per_second": 27.377, + "eval_steps_per_second": 2.284, + "eval_wer": 0.10744504271082142, + "step": 8000 + }, + { + "epoch": 10.77, + "eval_loss": 0.39167362451553345, + "eval_runtime": 165.6389, + "eval_samples_per_second": 28.514, + "eval_steps_per_second": 2.379, + "eval_wer": 0.10780940042913242, + "step": 8500 + }, + { + "epoch": 11.41, + "learning_rate": 1.1166666666666668e-05, + "loss": 0.2741, + "step": 9000 + }, + { + "epoch": 11.41, + "eval_loss": 0.3961273431777954, + "eval_runtime": 164.9191, + "eval_samples_per_second": 28.638, + "eval_steps_per_second": 2.389, + "eval_wer": 0.10568398040565159, + "step": 9000 + }, + { + "epoch": 12.04, + "eval_loss": 0.39744970202445984, + "eval_runtime": 164.8733, + "eval_samples_per_second": 28.646, + "eval_steps_per_second": 2.39, + "eval_wer": 0.10529938059187888, + "step": 9500 + }, + { + "epoch": 12.67, + "learning_rate": 5.555555555555556e-08, + "loss": 0.2531, + "step": 10000 + }, + { + "epoch": 12.67, + "eval_loss": 0.40417608618736267, + "eval_runtime": 164.6908, + "eval_samples_per_second": 28.678, + "eval_steps_per_second": 2.392, + "eval_wer": 0.10485405449172099, + "step": 10000 + }, + { + "epoch": 12.67, + "step": 10000, + "total_flos": 2.7136568655380324e+19, + "train_loss": 0.5960800704956055, + "train_runtime": 13155.4001, + "train_samples_per_second": 18.243, + "train_steps_per_second": 0.76 + } + ], + "max_steps": 10000, + "num_train_epochs": 13, + "total_flos": 2.7136568655380324e+19, + "trial_name": null, + "trial_params": null +} diff --git a/training_args.bin b/training_args.bin new file mode 100644 index 0000000..4861599 --- /dev/null +++ b/training_args.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3c84ee358d1ff2b725c335fdd3efaeb0ad280010e6d18e558b2ffafcfdf3c6f8 +size 3771 diff --git a/vocab.json b/vocab.json new file mode 100644 index 0000000..4540c2a --- /dev/null +++ b/vocab.json @@ -0,0 +1,31 @@ +{ + "[PAD]": 28, + "[UNK]": 27, + "a": 1, + "b": 2, + "c": 3, + "d": 4, + "e": 5, + "f": 6, + "g": 7, + "h": 8, + "i": 9, + "j": 10, + "k": 11, + "l": 12, + "m": 13, + "n": 14, + "o": 15, + "p": 16, + "q": 17, + "r": 18, + "s": 19, + "t": 20, + "u": 21, + "v": 22, + "w": 23, + "x": 24, + "y": 25, + "z": 26, + "|": 0 +}