"TRUST me I'm an Engineer"
2020-2022
M.S. Student at Data Science & Business Analytics Lab., School of Industrial Management Engineering, Korea University.2015-2020
Senior Engineer at SK holdings C&C., Korea.
License: Apache License 2.0
"TRUST me I'm an Engineer"
2020-2022
M.S. Student at Data Science & Business Analytics Lab., School of Industrial Management Engineering, Korea University.2015-2020
Senior Engineer at SK holdings C&C., Korea.์๋
ํ์ธ์ ์ฐ์ ๋ฉ์ง ์ฝ๋ ๊ณต์ ํด์ฃผ์
์ ๊ฐ์ฌํฉ๋๋ค
bash script/inference/evaluate_multimodel.sh ์คํ ์ ๋ฐ์ํ ์๋ฌ์ ๋ํด์ ์ง๋ฌธ๋๋ฆฝ๋๋ค
Traceback (most recent call last):
File "inference/beam_search.py", line 902, in
wer, cer, swer = cli_main()
File "inference/beam_search.py", line 835, in cli_main
args = options.parse_args_and_arch(parser)
File "/usr/local/lib/python3.8/dist-packages/fairseq/options.py", line 158, in parse_args_and_arch
TASK_REGISTRY[args.task].add_args(parser)
KeyError: 'audio_multitraining'
์ ์๋ฌ๊ฐ ์ด๋ค ์๋ฌ์ธ์ง ์์ญ๋๊น?
Hi,
I download the fairseq checkpoint from readme.
The inference/beam_search.py
is for asr I think, so I trying to running the bash script/inference/evaluate_multimodel.sh
, but missing some file MANIFEST_PATH
(some dictionary in it?)
Would like to share the MANIFEST
file?
Thank you.
์๋ ํ์ธ์. ์ข์ ์ฐ๊ตฌ๋ฅผ ๊ณต์ ํด์ฃผ์ ์ ๊ฐ์ฌ๋๋ฆฝ๋๋ค.
๋ ผ๋ฌธ์ report๋ CER ์ ์๋ฅผ ๊ณ์ฐํ๊ธฐ ์ํ ๊ตฌํ์ด ์ด๋ ๋ถ๋ถ์ธ์ง ํ์คํ์ง ์์ ์ง๋ฌธ๋๋ฆฝ๋๋ค.
์ถ์ธก์ปจ๋ฐ inference.py์์ ์๋์ ๋ถ๋ถ์ธ ๊ฒ ๊ฐ์๋ฐ์,
https://github.com/JoungheeKim/K-wav2vec/blob/main/inference/beam_search.py#L227
ํน์ ์ด์ฒ๋ผ editdistance library์ editdistance.eval() ํจ์๋ฅผ ์ฌ์ฉํ๊ณ , ํจ์์ ์ ๋ ฅ์ผ๋ก๋ ์์ ๋จ์๋ก normalize๋ ๋ ๊ฐ์ string์ด ์ฌ์ฉ๋๊ฒ ๋ง์๊น์?
๊ฐ์ฌํฉ๋๋ค.
huggingface ๋ชจ๋ธ์ ๋ณด๋ฉด vocab_size๊ฐ 32์ธ๊ฑธ๋ก ๋ณด์,
fairseq์์ ์ ๊ณตํ๋ vocab_list์ 4๊ฐ์ง speacial token์ ์ฌ์ฉํ ๊ฒ์ผ๋ก ๋ณด์ด๋๋ฐ์.
ํน์ ๊ทธ๋ฌ๋ฉด fine-tuningํ์ จ์๋, ksponspeech๋ ์ ๋ถ ํ๊ธ๋ก ๋์ด์์ํ ๋ฐ, ํ๊ธ์์-์ํ๋ฒณ ์์ฐจํ๊ธฐ๋ฒ์ ์ํด์ ๋ฐ์ดํฐ ๋ณํํด์ ์ฌ์ฉํ์ จ๋์?
๋ํ, ์ฝฉ๊ธ๋ฆฌ์ฌ๋ ์ซ์๊ฐ์๊ฒฝ์ฐ (1/ํ๋) ์ด๋ ๊ฒ ์๋๋ฐ, ์ด๋ฐ ๊ฒ๋ค๋ ๋ค ๋ฐ์ดํฐ ์ ์ฒ๋ฆฌํด์ ๋์ค ํ๋๋ก ์ ํํด์ ์ฌ์ฉํ์ ๊ฑด๊ฐ์?
์๋
ํ์ธ์
๋์์ ๋ฐ๊ณ ์ ๊ธ์ ๋จ๊ฒผ์ต๋๋ค. ์๋์ ๊ฐ์ด ptํ์ผ์ ๋ถ๋ฌ์ค๋ ค๊ณ ํ๋๋ฐ ConfigKeyError ์๋ฌ๊ฐ ๋๋๋ฐ
์ด๋ ๊ฒ ๋ถ๋ฌ์ค๋ ๊ฒ ๋ง๋์ง์?
import fairseq
cp_path = "wav2vec/pre_models/checkpoint_best.pt"
model, cfg, task = fairseq.checkpoint_utils.load_model_ensemble_and_task([cp_path])
model = model[0]
model.eval()
[์๋ฌ]
ConfigKeyError: Key 'eval_wer' not in 'AudioPretrainingConfig'
full_key: eval_wer
reference_type=Optional[AudioPretrainingConfig]
object_type=AudioPretrainingConfig
bash script/pretrain/run_further_pretrain.sh ์ฝ๋๋ฅผ ์ด์ฉํ์ฌ further pre-train์ ์งํํ๋ ค๊ณ ํ๋๋ฐ ๋ค์๊ณผ ๊ฐ์ ์๋ฌ๊ฐ ๋ฐ์ํฉ๋๋ค.
RuntimeError: one of the variables needed for gradient computation has been modified by an inplace operation: [torch.cuda.HalfTensor [8, 768, 180]], which is output 0 of AsStridedBackward0, is at version 2; expected version 1 instead. Hint: enable anomaly detection to find the operation that failed to compute its gradient, with torch.autograd.set_detect_anomaly(True).
ํน์ ์ด๋ป๊ฒ ํด๊ฒฐํ ์ ์์๊น์?
python -W ignore fairseq_cli/hydra_train.py \ task.data=path_to_data \ checkpoint.save_dir=save_dir \ task.del_silence=True \ model.additional_layers=2 \ model.w2v_path=path_to pretrain_weight \ distributed_training.distributed_world_size=2 \ --config-dir configs/finetune/add \ --config-name 960h
get stuck in
if not cfg.distributed_training.pipeline_model_parallel: self._criterion = self._criterion.to(device=self.device) self._model = self._model.to(device=self.device)
in fairseq.trainer.py line 72 with no errors..
what command should I use to train with multi-gpu?
์๋
ํ์ธ์.
ํ๋ฅญํ ์ฝ๋์ pre-trained ๋ชจ๋ธ ๊ณต์ ํด์ฃผ์
์ ์ ๋ง ๊ฐ์ฌ๋๋ฆฝ๋๋ค.
๋ค๋ฆ์ด ์๋๋ผ readme์ ์๋ ํ๊ตญ์ด ์ฌ์ ํ์ตํ ํ๊น
ํ์ด์ค ๋ชจ๋ธ์ ์๋ฒ์ ๋ค์ด๋ก๋ ๋ฐ์์ ๋๋ ค๋ณด๊ณ ์๋๋ฐ์
๋ชจ๋ธ์ local path ํตํด์ ์ load๋๋๋ฐ,
feature extractor๋ฅผ ๋ถ๋ฌ์ฌ ๋ .from_pretrained(local_path) ๋ก ๋ถ๋ฌ์ค๋๊น ์๊พธ ์๋์ ๊ฐ์ OSError๊ฐ ๋น๋๋ค.
... make sure {local_path} is the correct path to a directory containing a preprocessor_config.json file
๊ณต์ ํด์ฃผ์ ๋ชจ๋ธ ์์ถ ํ์ผ์๋ config.json๊ณผ pytorch_model.bin๋ง ์๋๋ฐ,
ํน์ feature extractor๋ ๊ทธ๋ฅ ํ๊น
ํ์ด์ค ํ๋ธ์ ์๋ ์์ด wav2vec2.0 ์ฒดํฌํฌ์ธํธ ๋๋ ๋ค๋ฅธ ํ๊ตญ์ด wav2vec2.0 ์ฒดํฌํฌ์ธํธ๋ก ๋ถ๋ฌ์ค๋ฉด ๋๋ ๊ฒ์ธ์ง ๊ถ๊ธํฉ๋๋ค.
๊ฐ์ฌํฉ๋๋ค.
Hi. I was wondering whether the pretrained checkpoint for K-wav2vec 2.0 was going to be released (the one that was further pretrained on the English checkpoint).
Thanks in advance.
A declarative, efficient, and flexible JavaScript library for building user interfaces.
๐ Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
An Open Source Machine Learning Framework for Everyone
The Web framework for perfectionists with deadlines.
A PHP framework for web artisans
Bring data to life with SVG, Canvas and HTML. ๐๐๐
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
Some thing interesting about web. New door for the world.
A server is a program made to process requests and deliver data to clients.
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
Some thing interesting about visualization, use data art
Some thing interesting about game, make everyone happy.
We are working to build community through open source technology. NB: members must have two-factor auth.
Open source projects and samples from Microsoft.
Google โค๏ธ Open Source for everyone.
Alibaba Open Source for everyone
Data-Driven Documents codes.
China tencent open source team.