Initial commit

InferenceServer.py

@@ -9,16 +9,12 @@ import glob
 import transformers
 from transformers import AutoTokenizer
 from transformers import AutoModelForSeq2SeqLM
-import lm_scorer
-from lm_scorer.models.auto import AutoLMScorer as LMScorer
 
 
 print("Loading models...")
 app = FastAPI()
 
 device = "cpu"
-batch_size = 1    
-scorer = LMScorer.from_pretrained("gpt2", device=device, batch_size=batch_size)    
 correction_model_tag = "prithivida/grammar_error_correcter_v2"
 correction_tokenizer = AutoTokenizer.from_pretrained(correction_model_tag)
 correction_model     = AutoModelForSeq2SeqLM.from_pretrained(correction_model_tag)
@@ -61,8 +57,5 @@ def correct(input_sentence, max_candidates=1):
         corrected.add(correction_tokenizer.decode(pred, skip_special_tokens=True).strip())
 
     corrected = list(corrected)
-    scores = scorer.sentence_score(corrected, log=True)
-    ranked_corrected = [(c,s) for c, s in zip(corrected, scores)]
-    ranked_corrected.sort(key = lambda x:x[1], reverse=True)
-    return ranked_corrected
+    return corrected[0], 0  #Corrected Sentence, Dummy score
 

app.py

@@ -1,10 +1,19 @@
 import streamlit as st
 from multiprocessing import Process
+from annotated_text import annotated_text
+from bs4 import BeautifulSoup
+import pandas as pd
+import torch
+import math
+import re
+import json
+import requests
+import spacy
+import errant
 import time
 import os
 
 def start_server():    
-    os.system("cat custom_req.txt | xargs -n 1 -L 1 pip3 install -U")
     os.system("uvicorn InferenceServer:app --port 8080 --host 0.0.0.0 --workers 1")
 
 def load_models():
@@ -29,9 +38,6 @@ if 'models_loaded' not in st.session_state:
 
 
 def show_highlights(input_text, corrected_sentence):
-    """
-        To show highlights
-    """
     try:
         strikeout = lambda x: '\u0336'.join(x) + '\u0336'
         highlight_text = highlight(input_text, corrected_sentence)
@@ -59,9 +65,6 @@ def show_highlights(input_text, corrected_sentence):
         st.stop()
 
 def show_edits(input_text, corrected_sentence):
-    """
-        To show edits
-    """
     try:
         edits = get_edits(input_text, corrected_sentence)
         df = pd.DataFrame(edits, columns=['type','original word', 'original start', 'original end', 'correct word', 'correct start', 'correct end'])
@@ -160,19 +163,7 @@ if __name__ == "__main__":
     if not st.session_state['models_loaded']:
         load_models() 
 
-    from annotated_text import annotated_text
-    from bs4 import BeautifulSoup
-    import pandas as pd
-    import torch
-    import math
-    import re
-    import json
-    import requests
-    import spacy
-    import errant
-
-
-
+    
     st.title('Gramformer')
     st.subheader('A framework for correcting english grammatical errors')
     st.markdown("Built for fun with 💙  by a quintessential foodie - Prithivi Da, The maker of [WhatTheFood](https://huggingface.co/spaces/prithivida/WhatTheFood), [Styleformer](https://github.com/PrithivirajDamodaran/Styleformer) and [Parrot paraphraser](https://github.com/PrithivirajDamodaran/Parrot_Paraphraser) | ✍️ [@prithivida](https://twitter.com/prithivida) |[[GitHub]](https://github.com/PrithivirajDamodaran)", unsafe_allow_html=True)
@@ -200,7 +191,7 @@ if __name__ == "__main__":
     )
     st.write("(or)")
     input_text = st.text_input(
-        label="Enter your own text",
+        label="Bring your own sentence",
         value=input_text
     )
 

requirements.txt

@@ -0,0 +1,13 @@
+st-annotated-text
+bs4 
+torch
+fastapi
+uvicorn
+spacy==2.3.0
+python-Levenshtein==0.12.2
+errant==2.2.0
+fsspec==2021.5.0
+tokenizers
+fuzzywuzzy==0.18.0
+sentencepiece==0.1.95
+transformers