Comments (5)
may-
commented on June 13, 2024
Hello @matthew-burnard,
Could you please add fp16: True in the config file, i.e. configs/transformer_reverse.yaml or configs/rnn_reverse.yaml, and try to run the same train command again?
training:
[...]
use_cuda: False
print_valid_sents: [0, 3, 6]
keep_best_ckpts: 2
fp16: True # <- add this line!In my CPU environment, I've confirmed it works with the modification above:
2022-10-27 22:42:32,654 - INFO - joeynmt.training - Train stats:
device: cpu
n_gpu: 0
16-bits training: True
gradient accumulation: 1
batch size per device: 10
effective batch size (w. parallel & accumulation): 10
2022-10-27 22:42:32,654 - INFO - joeynmt.training - EPOCH 1
2022-10-27 22:44:04,839 - INFO - joeynmt.training - Epoch 1, Step: 100, Batch Loss: 52.500000, Batch Acc: 0.072272, Tokens per Sec: 149, Lr: 0.001000
2022-10-27 22:45:38,308 - INFO - joeynmt.training - Epoch 1, Step: 200, Batch Loss: 46.500000, Batch Acc: 0.082955, Tokens per Sec: 151, Lr: 0.001000
2022-10-27 22:47:09,797 - INFO - joeynmt.training - Epoch 1, Step: 300, Batch Loss: 53.500000, Batch Acc: 0.098496, Tokens per Sec: 151, Lr: 0.001000
2022-10-27 22:48:42,891 - INFO - joeynmt.training - Epoch 1, Step: 400, Batch Loss: 63.250000, Batch Acc: 0.103837, Tokens per Sec: 149, Lr: 0.001000
2022-10-27 22:50:16,389 - INFO - joeynmt.training - Epoch 1, Step: 500, Batch Loss: 46.750000, Batch Acc: 0.122228, Tokens per Sec: 151, Lr: 0.001000
2022-10-27 22:51:48,991 - INFO - joeynmt.training - Epoch 1, Step: 600, Batch Loss: 34.250000, Batch Acc: 0.141818, Tokens per Sec: 150, Lr: 0.001000
2022-10-27 22:53:23,885 - INFO - joeynmt.training - Epoch 1, Step: 700, Batch Loss: 46.750000, Batch Acc: 0.171323, Tokens per Sec: 149, Lr: 0.001000
2022-10-27 22:54:56,047 - INFO - joeynmt.training - Epoch 1, Step: 800, Batch Loss: 23.750000, Batch Acc: 0.241879, Tokens per Sec: 148, Lr: 0.001000
2022-10-27 22:56:30,321 - INFO - joeynmt.training - Epoch 1, Step: 900, Batch Loss: 27.625000, Batch Acc: 0.356111, Tokens per Sec: 149, Lr: 0.001000
2022-10-27 22:58:06,413 - INFO - joeynmt.training - Epoch 1, Step: 1000, Batch Loss: 32.750000, Batch Acc: 0.504765, Tokens per Sec: 144, Lr: 0.001000
2022-10-27 22:58:06,413 - INFO - joeynmt.prediction - Predicting 1000 example(s)... (Greedy decoding with min_output_length=1, max_output_length=30, return_prob='none', generate_unk=True, repetition_penalty=-1, no_repeat_ngram_size=-1)
Predicting...: 100%|████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 1000/1000 [02:10<00:00, 7.63it/s]
2022-10-27 23:00:17,941 - INFO - joeynmt.metrics - nrefs:1|case:mixed|eff:no|tok:13a|smooth:exp|version:2.3.1
2022-10-27 23:00:17,941 - INFO - joeynmt.prediction - Evaluation result (greedy) bleu: 44.19, loss: 31.47, ppl: 6.88, acc: 0.65, generation: 131.0036[sec], evaluation: 0.5055[sec]
2022-10-27 23:00:17,943 - INFO - joeynmt.training - Hooray! New best validation result [bleu]!
2022-10-27 23:00:17,963 - INFO - joeynmt.training - Example #0
2022-10-27 23:00:17,964 - INFO - joeynmt.training - Source: 33 9 15 3 14 33 32 42 23 12 14 17 4 35 0 48 46 36 46 27 2 34 35 17 36 39 7 14 9 0
2022-10-27 23:00:17,964 - INFO - joeynmt.training - Reference: 0 9 14 7 39 36 17 35 34 2 27 46 36 46 48 0 35 4 17 14 12 23 42 32 33 14 3 15 9 33
2022-10-27 23:00:17,964 - INFO - joeynmt.training - Hypothesis: 0 9 14 7 9 36 17 35 34 2 2 27 12 21 46 0 23 17 4 4 4 4 9
2022-10-27 23:00:17,964 - INFO - joeynmt.training - Example #3
2022-10-27 23:00:17,966 - INFO - joeynmt.training - Source: 10 43 37 32 6 9 25 36 21 29 16 7 18 27 30 46 37 15 7 48 18
2022-10-27 23:00:17,966 - INFO - joeynmt.training - Reference: 18 48 7 15 37 46 30 27 18 7 16 29 21 36 25 9 6 32 37 43 10
2022-10-27 23:00:17,966 - INFO - joeynmt.training - Hypothesis: 18 48 7 15 37 46 30 27 18 13 13 29 29 36 25 9 32 32 43 43
2022-10-27 23:00:17,966 - INFO - joeynmt.training - Example #6
2022-10-27 23:00:17,967 - INFO - joeynmt.training - Source: 0 38 14 26 20 34 10 36 11 32 29 21
2022-10-27 23:00:17,967 - INFO - joeynmt.training - Reference: 21 29 32 11 36 10 34 20 26 14 38 0
2022-10-27 23:00:17,967 - INFO - joeynmt.training - Hypothesis: 21 29 32 11 36 10 34 26 26 14
Sorry, I haven't tested it in CPU before. And we know the tutorial is kind of outdated. We are working on the tutorial update for joey version 2, but it will take more time...
from joeynmt.
may-
commented on June 13, 2024
reminder for myself:
- force to assign fp16 = True, if use_cuda = False
- update reverse task config files to use fp16
- update reverse task tutorial
from joeynmt.
maya-burnard
commented on June 13, 2024
Confirmed that this solves the problem on my end.
from joeynmt.
may-
commented on June 13, 2024
Nice! Thank you for letting us know this. We will fix it.
from joeynmt.
may-
commented on June 13, 2024
solved in #202
from joeynmt.
Related Issues (20)
- Multi-GPU training. HOT 5
- JoeyNMT v1 procedure is no more compatible with JoeyNMT v2 HOT 12
- better config validation
- RuntimeError: indices should be either on cpu or on the same device as the indexed tensor (cpu) HOT 1
- AttributeError: module 'packaging' has no attribute 'version' HOT 2
- Unit test FAIL: testSentencepieceTokenizer (test.unit.test_tokenizer.TestTokenizer) HOT 4
- trg_mask generate problem HOT 4
- Running build_vocab.py for wmt17_bpe with or without --joint? HOT 3
- RuntimeError: indices should be either on cpu or on the same device as the indexed tensor (cpu) HOT 5
- run predict function in Colab, get ConfigurationError: Invalid `batch_type` option.
- (enhancement) Deploying trained models on HuggingFace Space HOT 2
- Basic iwslt config train failure due to directory errors HOT 1
- Early stopping criteria is only checked for the `ReduceLROnPlateau` scheduler HOT 5
- Link in Tutorial to Collab dead HOT 4
- Tutorial - Test Set Evaluation HOT 5
- Columns and DataType Not Explicitly Set on line 387 of datasets.py
- Unit Test Fails - Windows Installation HOT 4
- serving & ONNX compat ?
- Implementing Knowledge distillation HOT 3
Recommend Projects
-
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
-
Vue.js
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
-
Typescript
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
-
TensorFlow
An Open Source Machine Learning Framework for Everyone
-
Django
The Web framework for perfectionists with deadlines.
-
Laravel
A PHP framework for web artisans
-
D3
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
-
Recommend Topics
-
javascript
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
-
web
Some thing interesting about web. New door for the world.
-
server
A server is a program made to process requests and deliver data to clients.
-
Machine learning
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
-
Visualization
Some thing interesting about visualization, use data art
-
Game
Some thing interesting about game, make everyone happy.
Recommend Org
-
Facebook
We are working to build community through open source technology. NB: members must have two-factor auth.
-
Microsoft
Open source projects and samples from Microsoft.
-
Google
Google ❤️ Open Source for everyone.
-
Alibaba
Alibaba Open Source for everyone
-
D3
Data-Driven Documents codes.
-
Tencent
China tencent open source team.
from joeynmt.