/ner_axon_example

Primary LanguageElixir

BasicAiReg

mix deps.get

LSTM Based NER RNN

Issues with the shapes

Cant solve the input -> output shape issue::

17:18:03.243 [info] StreamExecutor device (0): Host, Default Version ** (RuntimeError) branch operand 0 must match the shape of the only parameter of branch computation 0: got operand: (token[], f32[66,256], s64[128,100,1], f32[256,1024], f32[256,1024], /index=5/f32[256,1024], f32[256,1024], f32[1024,1024], f32[1024,1024], f32[1024,1024], /index=10/f32[1024,1024], f32[1024], f32[1024], f32[1024], f32[1024], /index=15/s64[128,100,1], s64[128,100,1], s64[], s64[], f32[], /index=20/f32[9], f32[1024,9], f32[66,256], f32[1024], f32[1024], /index=25/f32[1024], f32[1024], f32[1024,1024], f32[1024,1024], f32[1024,1024], /index=30/f32[1024,1024], f32[256,1024], f32[1024,9], f32[9], s64[128,100,1]); computation: (p: (token[], f32[66,256], s64[128,100,1], f32[256,1024], f32[256,1024], /index=5/f32[256,1024], f32[256,1024], f32[1024,1024], f32[1024,1024], f32[1024,1024], /index=10/f32[1024,1024], f32[1024], f32[1024], f32[1024], f32[1024], /index=15/s64[], f32[128,1,1024], f32[128,1,1024], f32[128,100,1024], f32[128,100,256], /index=20/f32[256,1024], f32[256,1024], f32[256,1024], f32[256,1024], f32[1024,1024], /index=25/f32[1024,1024], f32[1024,1024], f32[1024,1024], f32[1024], f32[1024], /index=30/f32[1024], f32[1024], f32[1024,9], f32[9], s64[128,100,1])) -> (token[], f32[]).

mix run -e "Ner.run()"

Basic from scratch

https://keras.io/examples/nlp/text_classification_from_scratch/

mix run -e "NerB.run()"

Results

14:50:14.056 [info] StreamExecutor device (0): Host, Default Version Epoch: 0, Batch: 0, accuracy: 0.2031250 loss: 0.0000000 Epoch: 1, Batch: 0, accuracy: 0.6015625 loss: 0.3883106 Epoch: 2, Batch: 0, accuracy: 0.6484375 loss: 0.3281645 Epoch: 3, Batch: 0, accuracy: 0.6484375 loss: 0.3025800 Epoch: 4, Batch: 0, accuracy: 0.6484375 loss: 0.2894246 Epoch: 5, Batch: 0, accuracy: 0.6484375 loss: 0.2812856 Epoch: 6, Batch: 0, accuracy: 0.6484375 loss: 0.2759556 Epoch: 7, Batch: 0, accuracy: 0.6484375 loss: 0.2718235 Epoch: 8, Batch: 0, accuracy: 0.6484375 loss: 0.2685038 Epoch: 9, Batch: 0, accuracy: 0.6484375 loss: 0.2659863 Epoch: 10, Batch: 0, accuracy: 0.6484375 loss: 0.2639405 Epoch: 11, Batch: 0, accuracy: 0.6484375 loss: 0.2620718 Epoch: 12, Batch: 0, accuracy: 0.6484375 loss: 0.2605137 Epoch: 13, Batch: 0, accuracy: 0.6484375 loss: 0.2592957