[#9] `fetch_pipeline` has been added. Fixed the bug where <pad> tokens will appear in the final output.
Browse files- idiomify/fetchers.py +15 -0
- idiomify/pipeline.py +5 -2
- main_deploy.py +5 -11
- main_infer.py +4 -12
idiomify/fetchers.py
CHANGED
|
@@ -8,6 +8,7 @@ from idiomify.paths import CONFIG_YAML, idioms_dir, literal2idiomatic, idiomifie
|
|
| 8 |
from idiomify.urls import PIE_URL
|
| 9 |
from transformers import AutoModelForSeq2SeqLM, AutoConfig, BartTokenizer
|
| 10 |
from idiomify.models import Idiomifier
|
|
|
|
| 11 |
|
| 12 |
|
| 13 |
# --- from the web --- #
|
|
@@ -75,6 +76,20 @@ def fetch_tokenizer(ver: str, run: Run = None) -> BartTokenizer:
|
|
| 75 |
return tokenizer
|
| 76 |
|
| 77 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 78 |
def fetch_config() -> dict:
|
| 79 |
with open(str(CONFIG_YAML), 'r', encoding="utf-8") as fh:
|
| 80 |
return yaml.safe_load(fh)
|
|
|
|
| 8 |
from idiomify.urls import PIE_URL
|
| 9 |
from transformers import AutoModelForSeq2SeqLM, AutoConfig, BartTokenizer
|
| 10 |
from idiomify.models import Idiomifier
|
| 11 |
+
from idiomify.pipeline import Pipeline
|
| 12 |
|
| 13 |
|
| 14 |
# --- from the web --- #
|
|
|
|
| 76 |
return tokenizer
|
| 77 |
|
| 78 |
|
| 79 |
+
def fetch_pipeline() -> Pipeline:
|
| 80 |
+
"""
|
| 81 |
+
fetch a pipeline of the version stated in config.yaml
|
| 82 |
+
"""
|
| 83 |
+
config = fetch_config()['idiomifier']
|
| 84 |
+
model = fetch_idiomifier(config['ver'])
|
| 85 |
+
tokenizer = fetch_tokenizer(config['tokenizer_ver'])
|
| 86 |
+
idioms = fetch_idioms(config['idioms_ver'])
|
| 87 |
+
model.eval() # this is crucial to obtain consistent results
|
| 88 |
+
pipeline = Pipeline(model, tokenizer, idioms)
|
| 89 |
+
return pipeline
|
| 90 |
+
|
| 91 |
+
|
| 92 |
+
# --- from local --- #
|
| 93 |
def fetch_config() -> dict:
|
| 94 |
with open(str(CONFIG_YAML), 'r', encoding="utf-8") as fh:
|
| 95 |
return yaml.safe_load(fh)
|
idiomify/pipeline.py
CHANGED
|
@@ -1,4 +1,5 @@
|
|
| 1 |
import re
|
|
|
|
| 2 |
from typing import List
|
| 3 |
from transformers import BartTokenizer
|
| 4 |
from idiomify.builders import SourcesBuilder
|
|
@@ -7,9 +8,10 @@ from idiomify.models import Idiomifier
|
|
| 7 |
|
| 8 |
class Pipeline:
|
| 9 |
|
| 10 |
-
def __init__(self, model: Idiomifier, tokenizer: BartTokenizer):
|
| 11 |
self.model = model
|
| 12 |
self.builder = SourcesBuilder(tokenizer)
|
|
|
|
| 13 |
|
| 14 |
def __call__(self, sents: List[str], max_length=100) -> List[str]:
|
| 15 |
srcs = self.builder(literal2idiomatic=[(sent, "") for sent in sents])
|
|
@@ -19,9 +21,10 @@ class Pipeline:
|
|
| 19 |
decoder_start_token_id=self.model.hparams['bos_token_id'],
|
| 20 |
max_length=max_length,
|
| 21 |
) # -> (N, L_t)
|
|
|
|
| 22 |
tgts = self.builder.tokenizer.batch_decode(pred_ids, skip_special_tokens=False)
|
| 23 |
tgts = [
|
| 24 |
-
re.sub(r"<s>|</s>", "", tgt)
|
| 25 |
for tgt in tgts
|
| 26 |
]
|
| 27 |
return tgts
|
|
|
|
| 1 |
import re
|
| 2 |
+
import pandas as pd
|
| 3 |
from typing import List
|
| 4 |
from transformers import BartTokenizer
|
| 5 |
from idiomify.builders import SourcesBuilder
|
|
|
|
| 8 |
|
| 9 |
class Pipeline:
|
| 10 |
|
| 11 |
+
def __init__(self, model: Idiomifier, tokenizer: BartTokenizer, idioms: pd.DataFrame):
|
| 12 |
self.model = model
|
| 13 |
self.builder = SourcesBuilder(tokenizer)
|
| 14 |
+
self.idioms = idioms
|
| 15 |
|
| 16 |
def __call__(self, sents: List[str], max_length=100) -> List[str]:
|
| 17 |
srcs = self.builder(literal2idiomatic=[(sent, "") for sent in sents])
|
|
|
|
| 21 |
decoder_start_token_id=self.model.hparams['bos_token_id'],
|
| 22 |
max_length=max_length,
|
| 23 |
) # -> (N, L_t)
|
| 24 |
+
# we don't skip special tokens because we have to keep <idiom> & </idiom> for highlighting idioms.
|
| 25 |
tgts = self.builder.tokenizer.batch_decode(pred_ids, skip_special_tokens=False)
|
| 26 |
tgts = [
|
| 27 |
+
re.sub(r"<s>|</s>|<pad>", "", tgt)
|
| 28 |
for tgt in tgts
|
| 29 |
]
|
| 30 |
return tgts
|
main_deploy.py
CHANGED
|
@@ -3,30 +3,24 @@ we deploy the pipeline via streamlit.
|
|
| 3 |
"""
|
| 4 |
import re
|
| 5 |
import streamlit as st
|
| 6 |
-
from idiomify.fetchers import
|
| 7 |
from idiomify.pipeline import Pipeline
|
| 8 |
|
| 9 |
|
| 10 |
@st.cache(allow_output_mutation=True)
|
| 11 |
-
def
|
| 12 |
-
|
| 13 |
-
model = fetch_idiomifier(config['ver'])
|
| 14 |
-
tokenizer = fetch_tokenizer(config['tokenizer_ver'])
|
| 15 |
-
idioms = fetch_idioms(config['idioms_ver'])
|
| 16 |
-
return config, model, tokenizer, idioms
|
| 17 |
|
| 18 |
|
| 19 |
def main():
|
| 20 |
# fetch a pre-trained model
|
| 21 |
-
|
| 22 |
-
model.eval()
|
| 23 |
-
pipeline = Pipeline(model, tokenizer)
|
| 24 |
st.title("Idiomify Demo")
|
| 25 |
text = st.text_area("Type sentences here",
|
| 26 |
value="Just remember that there will always be a hope even when things look hopeless")
|
| 27 |
with st.sidebar:
|
| 28 |
st.subheader("Supported idioms")
|
| 29 |
-
idioms = [row["Idiom"] for _, row in idioms.iterrows()]
|
| 30 |
st.write(" / ".join(idioms))
|
| 31 |
|
| 32 |
if st.button(label="Idiomify"):
|
|
|
|
| 3 |
"""
|
| 4 |
import re
|
| 5 |
import streamlit as st
|
| 6 |
+
from idiomify.fetchers import fetch_pipeline
|
| 7 |
from idiomify.pipeline import Pipeline
|
| 8 |
|
| 9 |
|
| 10 |
@st.cache(allow_output_mutation=True)
|
| 11 |
+
def cache_pipeline() -> Pipeline:
|
| 12 |
+
return fetch_pipeline()
|
|
|
|
|
|
|
|
|
|
|
|
|
| 13 |
|
| 14 |
|
| 15 |
def main():
|
| 16 |
# fetch a pre-trained model
|
| 17 |
+
pipeline = cache_pipeline()
|
|
|
|
|
|
|
| 18 |
st.title("Idiomify Demo")
|
| 19 |
text = st.text_area("Type sentences here",
|
| 20 |
value="Just remember that there will always be a hope even when things look hopeless")
|
| 21 |
with st.sidebar:
|
| 22 |
st.subheader("Supported idioms")
|
| 23 |
+
idioms = [row["Idiom"] for _, row in pipeline.idioms.iterrows()]
|
| 24 |
st.write(" / ".join(idioms))
|
| 25 |
|
| 26 |
if st.button(label="Idiomify"):
|
main_infer.py
CHANGED
|
@@ -2,9 +2,7 @@
|
|
| 2 |
This is for just a simple sanity check on the inference.
|
| 3 |
"""
|
| 4 |
import argparse
|
| 5 |
-
from idiomify.
|
| 6 |
-
from idiomify.fetchers import fetch_config, fetch_idiomifier, fetch_tokenizer
|
| 7 |
-
from transformers import BartTokenizer
|
| 8 |
|
| 9 |
|
| 10 |
def main():
|
|
@@ -12,15 +10,9 @@ def main():
|
|
| 12 |
parser.add_argument("--sent", type=str,
|
| 13 |
default="Just remember that there will always be a hope even when things look hopeless")
|
| 14 |
args = parser.parse_args()
|
| 15 |
-
|
| 16 |
-
|
| 17 |
-
|
| 18 |
-
tokenizer = fetch_tokenizer(config['tokenizer_ver'])
|
| 19 |
-
model.eval() # this is crucial
|
| 20 |
-
pipeline = Pipeline(model, tokenizer)
|
| 21 |
-
src = config['sent']
|
| 22 |
-
tgts = pipeline(sents=[src])
|
| 23 |
-
print(src, "\n->", tgts[0])
|
| 24 |
|
| 25 |
|
| 26 |
if __name__ == '__main__':
|
|
|
|
| 2 |
This is for just a simple sanity check on the inference.
|
| 3 |
"""
|
| 4 |
import argparse
|
| 5 |
+
from idiomify.fetchers import fetch_pipeline
|
|
|
|
|
|
|
| 6 |
|
| 7 |
|
| 8 |
def main():
|
|
|
|
| 10 |
parser.add_argument("--sent", type=str,
|
| 11 |
default="Just remember that there will always be a hope even when things look hopeless")
|
| 12 |
args = parser.parse_args()
|
| 13 |
+
pipeline = fetch_pipeline()
|
| 14 |
+
tgts = pipeline(sents=[args.sent])
|
| 15 |
+
print(args.sent, "\n->", tgts[0])
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 16 |
|
| 17 |
|
| 18 |
if __name__ == '__main__':
|