Skip to content
Lint & Tests

fix batch inference (#169) #658

Sign in to view logs

fix batch inference (#169)

fix batch inference (#169) #658

Workflow file for this run

.github/workflows/push.yml at b5e2266
name: Lint & Tests
on: [push, pull_request]
jobs:
lint-and-tests:
runs-on: ubuntu-latest
strategy:
matrix:
python-version: ['3.10']
steps:
- uses: actions/checkout@v2
- name: Set up Python ${{ matrix.python-version }}
uses: actions/setup-python@v2
with:
python-version: ${{ matrix.python-version }}
- name: Install dependencies
run: |
python -m pip install --upgrade pip
pip install --upgrade setuptools
pip install -e .
pip install -r requirements.opt.txt
pip install sacrebleu
pip install flake8
pip install rich
python -m pip install black==22.* flake8==3.8.*
if [ -f requirements.txt ]; then pip install -r requirements.txt; fi
- name: Check code with Black
run: |
black --check .
- name: Lint with flake8
run: |
flake8 .
- name: Unit tests
run: |
python -m unittest discover
- name: Recipes config check
run: |
python eole/tests/test_recipes.py recipes
- name: Test vocabulary build
run: |
python eole/bin/main.py build_vocab \
-config eole/tests/data/data.yaml \
-save_data /tmp/eole \
-n_sample 5000 \
-src_vocab /tmp/eole.vocab.src \
-tgt_vocab /tmp/eole.vocab.tgt \
&& rm -rf /tmp/sample
- name: Test field/transform dump
run: |
# The dumped fields are used later when testing tools
python eole/bin/main.py train \
-config eole/tests/data/data.yaml \
-save_data /tmp/eole.train.check \
-n_sample 30 \
-model '{"architecture": "rnn"}' \
-training '{"num_workers": 0, "bucket_size": 1024}' \
-src_vocab /tmp/eole.vocab.src \
-tgt_vocab /tmp/eole.vocab.tgt \
-src_vocab_size 1000 \
-tgt_vocab_size 1000
- name: Test RNN training
run: |
python eole/bin/main.py train \
-config eole/tests/data/data.yaml \
-src_vocab /tmp/eole.vocab.src \
-tgt_vocab /tmp/eole.vocab.tgt \
-src_vocab_size 1000 \
-tgt_vocab_size 1000 \
-model '{"architecture": "rnn", "hidden_size": 10, "embeddings": {"word_vec_size": 5, "position_encoding_type": None}}' \
-training '{"batch_size": 10, "num_workers": 0, "bucket_size": 1024, "train_steps": 10}' \
-report_every 5\
-tensorboard \
-tensorboard_log_dir /tmp/logs_train
python eole/tests/test_events.py --logdir /tmp/logs_train -tensorboard_checks train
- name: Test RNN training and validation
run: |
python eole/bin/main.py train \
-config eole/tests/data/data.yaml \
-src_vocab /tmp/eole.vocab.src \
-tgt_vocab /tmp/eole.vocab.tgt \
-src_vocab_size 1000 \
-tgt_vocab_size 1000 \
-model '{"architecture": "rnn", "hidden_size": 10, "embeddings": {"word_vec_size": 5, "position_encoding_type": None}}' \
-training '{"batch_size": 10, "num_workers": 0, "bucket_size": 1024, "train_steps": 10, "valid_steps": 5}' \
-report_every 5 \
-tensorboard \
-tensorboard_log_dir /tmp/logs_train_and_valid
python eole/tests/test_events.py --logdir /tmp/logs_train_and_valid -tensorboard_checks train
python eole/tests/test_events.py --logdir /tmp/logs_train_and_valid -tensorboard_checks valid
- name: Test RNN training with coverage
run: |
python eole/bin/main.py train \
-config eole/tests/data/data.yaml \
-src_vocab /tmp/eole.vocab.src \
-tgt_vocab /tmp/eole.vocab.tgt \
-src_vocab_size 1000 \
-tgt_vocab_size 1000 \
-report_every 5 \
-model '{"architecture": "rnn", "hidden_size": 10, "embeddings": {"word_vec_size": 5, "position_encoding_type": None}, "decoder": {"coverage_attn": True, "lambda_coverage": 0.1}}' \
-training '{"batch_size": 10, "num_workers": 0, "bucket_size": 1024, "train_steps": 10}'
- name: Test Transformer training with align
run: |
python eole/bin/main.py train \
-config eole/tests/data/align_data.yaml \
-src_vocab /tmp/eole.vocab.src \
-tgt_vocab /tmp/eole.vocab.tgt \
-src_vocab_size 1000 \
-tgt_vocab_size 1000 \
-model '{"layers": 4, "hidden_size": 16, "transformer_ff": 64, "embeddings": {"word_vec_size": 16}, "encoder": {"encoder_type": "transformer", "heads": 2}, "decoder": {"decoder_type": "transformer", "lambda_align": 0.05, "alignment_layer": 2, "alignment_heads": 0, "heads": 2}}' \
-training '{"batch_size": 10, "num_workers": 0, "bucket_size": 1024, "train_steps": 10, "dropout_steps": [0, 3, 7], "dropout": [0.3, 0.2, 0.1], "attention_dropout": [0.2, 0.2, 0.1]}' \
-report_every 5 \
- name : Test Transformer training and validation with dynamic scoring
run: |
python3 eole/bin/main.py train \
-config eole/tests/data/data.yaml \
-src_vocab /tmp/eole.vocab.src \
-tgt_vocab /tmp/eole.vocab.tgt \
-src_vocab_size 1000 \
-tgt_vocab_size 1000 \
-model '{"layers": 4, "hidden_size": 16, "transformer_ff": 16, "embeddings": {"word_vec_size": 16, "position_encoding_type": "SinusoidalInterleaved"}, "encoder": {"encoder_type": "transformer", "heads": 2}, "decoder": {"decoder_type": "transformer", "heads": 2}}' \
-training '{"batch_size": 10, "num_workers": 0, "bucket_size": 1024, "train_steps": 10, "valid_steps": 5, "accum_count": [2, 4, 8], "accum_steps": [0, 3, 7], "model_path": "/tmp/eole.model"}' \
-report_every 2 \
-valid_metrics "BLEU" "TER" \
-tensorboard \
-scoring_debug \
-tensorboard_log_dir /tmp/logs_dynamic-scoring \
-dump_preds /tmp/dump_preds
python eole/tests/test_events.py --logdir /tmp/logs_dynamic-scoring -tensorboard_checks valid_metrics
- name : Test Transformer training and validation with dynamic scoring and maxrelative
run: |
python3 eole/bin/main.py train \
-config eole/tests/data/data.yaml \
-src_vocab /tmp/eole.vocab.src \
-tgt_vocab /tmp/eole.vocab.tgt \
-src_vocab_size 1000 \
-tgt_vocab_size 1000 \
-model '{"architecture": "transformer", "layers": 4, "heads": 2, "hidden_size": 16, "transformer_ff": 64, "embeddings": {"word_vec_size": 16, "position_encoding_type": "Relative", "n_positions": 8}}' \
-training '{"batch_size": 10, "num_workers": 0, "bucket_size": 1024, "train_steps": 10, "valid_steps": 5}' \
-report_every 2 \
-valid_metrics "BLEU" "TER" \
-tensorboard \
-scoring_debug \
-tensorboard_log_dir /tmp/logs_dynamic-scoring_and_relative \
-dump_preds /tmp/dump_preds
python eole/tests/test_events.py --logdir /tmp/logs_dynamic-scoring_and_relative -tensorboard_checks valid_metrics
- name : Test Transformer training and validation with dynamic scoring and rotary
run: |
python3 eole/bin/main.py train \
-config eole/tests/data/data.yaml \
-src_vocab /tmp/eole.vocab.src \
-tgt_vocab /tmp/eole.vocab.tgt \
-src_vocab_size 1000 \
-tgt_vocab_size 1000 \
-model '{"architecture": "transformer", "layers": 4, "heads": 2, "hidden_size": 16, "transformer_ff": 64, "embeddings": {"word_vec_size": 16, "position_encoding_type": "Rotary"}}' \
-training '{"batch_size": 10, "num_workers": 0, "bucket_size": 1024, "train_steps": 10, "valid_steps": 5}' \
-report_every 2 \
-valid_metrics "BLEU" "TER" \
-tensorboard \
-scoring_debug \
-tensorboard_log_dir /tmp/logs_dynamic-scoring_and_rotary \
-dump_preds /tmp/dump_preds
python eole/tests/test_events.py --logdir /tmp/logs_dynamic-scoring_and_rotary -tensorboard_checks valid_metrics
- name : Test Transformer training and validation with dynamic scoring and alibi
run: |
python3 eole/bin/main.py train \
-config eole/tests/data/data.yaml \
-src_vocab /tmp/eole.vocab.src \
-tgt_vocab /tmp/eole.vocab.tgt \
-src_vocab_size 1000 \
-tgt_vocab_size 1000 \
-model '{"architecture": "transformer", "layers": 4, "heads": 2, "hidden_size": 16, "transformer_ff": 64, "embeddings": {"word_vec_size": 16, "position_encoding_type": "Alibi"}}' \
-training '{"batch_size": 10, "num_workers": 0, "bucket_size": 1024, "train_steps": 10, "valid_steps": 5}' \
-report_every 2 \
-valid_metrics "BLEU" "TER" \
-tensorboard \
-scoring_debug \
-tensorboard_log_dir /tmp/logs_dynamic-scoring_and_alibi \
-dump_preds /tmp/dump_preds
python eole/tests/test_events.py --logdir /tmp/logs_dynamic-scoring_and_alibi -tensorboard_checks valid_metrics
- name: Test LM training
run: |
python eole/bin/main.py train \
-config eole/tests/data/lm_data.yaml \
-src_vocab /tmp/eole.vocab.src \
-tgt_vocab /tmp/eole.vocab.src \
-model '{"hidden_size": 16, "transformer_ff": 64, "embeddings": {"word_vec_size": 16}, "encoder": None, "decoder": {"decoder_type": "transformer_lm", "layers": 2, "heads": 4}}' \
-training '{"batch_size": 10, "num_workers": 0, "bucket_size": 1024, "train_steps": 10}' \
-src_vocab_size 1000 \
-tgt_vocab_size 1000 \
-report_every 5
- name: Test RNN translation
run: |
head eole/tests/data/src-test.txt > /tmp/src-test.txt
python eole/bin/main.py predict \
-model_path eole/tests/test_model \
-src /tmp/src-test.txt \
-verbose
- name: Test RNN ensemble translation
run: |
head eole/tests/data/src-test.txt > /tmp/src-test.txt
python eole/bin/main.py predict \
-model_path eole/tests/test_model \
eole/tests/test_model \
-src /tmp/src-test.txt \
-verbose
- name: Test RNN translation with beam search
run: |
python eole/bin/main.py predict \
-model_path eole/tests/test_model2 \
-src eole/tests/data/morph/src.valid \
-verbose \
-batch_size 10 \
-beam_size 10 \
-tgt eole/tests/data/morph/tgt.valid \
-out /tmp/trans
diff eole/tests/data/morph/tgt.valid /tmp/trans && rm /tmp/trans
- name: Test RNN translation with random sampling
run: |
python eole/bin/main.py predict \
-model_path eole/tests/test_model2 \
-src eole/tests/data/morph/src.valid \
-verbose \
-batch_size 10 \
-beam_size 1 \
-seed 1 \
-top_k "-1" \
-temperature 0.0001 \
-tgt eole/tests/data/morph/tgt.valid \
-out /tmp/trans
diff eole/tests/data/morph/tgt.valid /tmp/trans && rm /tmp/trans
- name: Test LM generation
run: |
head eole/tests/data/src-test.txt > /tmp/src-test.txt
python eole/bin/main.py predict \
-model_path eole/tests/test_model_lm \
-src /tmp/src-test.txt \
-verbose
- name: Test LM generation with beam search
run: |
python eole/bin/main.py predict \
-model_path eole/tests/test_model_lm \
-src eole/tests/data/data_lm/src-gen.txt \
-verbose -batch_size 1 \
-beam_size 10 \
-ban_unk_token \
-length_penalty none \
-out /tmp/gen
diff eole/tests/data/data_lm/gen-beam-sol.txt /tmp/gen && rm /tmp/gen
- name: Test LM generation with random sampling
run: |
python eole/bin/main.py predict -model_path eole/tests/test_model_lm \
-src eole/tests/data/data_lm/src-gen.txt \
-verbose -batch_size 1 \
-beam_size 1 \
-seed 1 \
-top_k -1 \
-temperature 0.0001 \
-ban_unk_token \
-length_penalty none \
-out /tmp/gen
diff eole/tests/data/data_lm/gen-sampling-sol.txt /tmp/gen && rm /tmp/gen
- name: Test LM generation with random top-k/nucleus sampling
run: |
python eole/bin/main.py predict -model_path eole/tests/test_model_lm \
-src eole/tests/data/data_lm/src-gen.txt \
-verbose -batch_size 1 \
-beam_size 1 \
-seed 3 \
-top_k -1 \
-top_p 0.95 \
-temperature 1 \
-ban_unk_token \
-length_penalty none \
-out /tmp/gen
diff eole/tests/data/data_lm/gen-nucleus-sampling-sol$(python -c "import torch; print(torch.__version__[0])").txt /tmp/gen && rm /tmp/gen
- name: Test LM generation with random sampling multi-beams
run: |
python eole/bin/main.py predict -model_path eole/tests/test_model_lm \
-src eole/tests/data/data_lm/src-gen.txt \
-verbose -batch_size 1 \
-beam_size 10 \
-seed 2 \
-top_k 50 \
-top_p 0.95 \
-temperature 1 \
-length_penalty avg \
-ban_unk_token \
-min_length 5 \
-out /tmp/gen
diff eole/tests/data/data_lm/gen-sampling-beams-sol$(python -c "import torch; print(torch.__version__[0])").txt /tmp/gen && rm /tmp/gen
- name: Test py-LM inference engine
run: |
head eole/tests/data/src-test.txt > /tmp/src-test.txt
python eole/tests/test_inference_engines.py \
-model eole/tests/test_model_lm \
-model_type decoder \
-input_file /tmp/src-test.txt \
-inference_config_file eole/tests/data/inference-engine_py.yaml \
-out /tmp/inference_engine_lm_py_outputs
- name: Test ct2-LM inference engine
run: |
head eole/tests/data/src-test.txt > /tmp/src-test.txt
python eole/tests/test_inference_engines.py \
-model eole/tests \
-model_type decoder \
-input_file /tmp/src-test.txt \
-inference_config_file eole/tests/data/inference-engine_py.yaml \
-engine ct2 \
-out /tmp/inference_engine_lm_py_outputs
- name: Test py-SEQ2SEQ inference engine
run: |
head eole/tests/data/src-test.txt > /tmp/src-test.txt
python eole/tests/test_inference_engines.py \
-model eole/tests/test_model \
-model_type encoder_decoder \
-input_file /tmp/src-test.txt \
-inference_config_file eole/tests/data/inference-engine_py.yaml \
-out /tmp/inference_engine_lm_py_outputs
- name: Test embeddings_to_torch tool
run: |
python eole/bin/main.py tools embeddings_to_torch \
-emb_file_enc eole/tests/sample_glove.txt \
-emb_file_dec eole/tests/sample_glove.txt \
-model_path eole/tests/test_model \
-output_file /tmp/q_gloveembeddings \
&& rm /tmp/q_gloveembeddings*
- name: Test extract_embeddings tool
run: |
python eole/bin/main.py model extract_embeddings \
-model eole/tests/test_model
- name: Test checkpoint vocabulary update
run: |
python eole/bin/main.py train \
-config eole/tests/data/data.yaml \
-src_vocab /tmp/eole.vocab.src \
-tgt_vocab /tmp/eole.vocab.tgt \
-src_vocab_size 1000 \
-tgt_vocab_size 1000 \
-model '{"architecture": "rnn", "hidden_size": 10, "embeddings": {"word_vec_size": 5, "position_encoding_type": None}}' \
-training '{"batch_size": 10, "num_workers": 0, "bucket_size": 1024, "train_steps": 10, "model_path": "/tmp/eole.model", "save_checkpoint_steps": 10}' \
-report_every 5
sed -i '1s/^/new_tok\t100000000\n/' /tmp/eole.vocab.src
python eole/bin/main.py train \
-config eole/tests/data/data.yaml \
-src_vocab /tmp/eole.vocab.src \
-tgt_vocab /tmp/eole.vocab.tgt \
-src_vocab_size 1000 \
-tgt_vocab_size 1000 \
-training '{"batch_size": 10, "num_workers": 0, "bucket_size": 1024, "train_steps": 20, "train_from": "/tmp/eole.model/step_10", "save_checkpoint_steps": 10, "update_vocab": True, "reset_optim": "states"}' \
-report_every 5
- name: Test checkpoint vocabulary update with LM
run: |
python eole/bin/main.py train \
-config eole/tests/data/lm_data.yaml \
-src_vocab /tmp/eole.vocab.src \
-tgt_vocab /tmp/eole.vocab.src \
-model '{"layers": 2, "hidden_size": 16, "transformer_ff": 64, "embeddings": {"word_vec_size": 16}, "encoder": None, "decoder": {"decoder_type": "transformer_lm", "heads": 4}}' \
-training '{"batch_size": 10, "num_workers": 0, "bucket_size": 1024, "train_steps": 10, "model_path": "/tmp/lm.eole.model", "save_checkpoint_steps": 10}' \
-src_vocab_size 1000 \
-tgt_vocab_size 1000 \
-report_every 5
sed -i '1s/^/new_tok2\t100000000\n/' /tmp/eole.vocab.src
python eole/bin/main.py train \
-config eole/tests/data/lm_data.yaml \
-src_vocab /tmp/eole.vocab.src \
-tgt_vocab /tmp/eole.vocab.src \
-training '{"batch_size": 10, "num_workers": 0, "bucket_size": 1024, "train_steps": 20, "train_from": "/tmp/lm.eole.model/step_10", "save_checkpoint_steps": 10, "update_vocab": True, "reset_optim": "states"}' \
-src_vocab_size 1000 \
-tgt_vocab_size 1000 \
-report_every 5
build-docs:
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v2
- name: Set up Python 3.10
uses: actions/setup-python@v2
with:
python-version: '3.10'
- name: Install dependencies
run: |
python -m pip install --upgrade pip
pip install --upgrade setuptools
pip install -e .
pip install -r docs/requirements.txt
pip install -r requirements.opt.txt
pip install rich
- name: Build docs
run: |
set -e
# Check that docs are built without errors
cd docs/ && make html && cd ..