edugp commited on
Commit
5640d38
1 Parent(s): 2388248

Switch PAWS by XNLI

Browse files
Files changed (1) hide show
  1. app.py +8 -15
app.py CHANGED
@@ -34,20 +34,11 @@ PROMPT_LIST = [
34
 
35
  PAWS_X_PROMPT_LIST = [
36
  "Te amo.</s>Te adoro.",
37
- "Te odio.</s>Te detesto.",
38
- "Me gusta montar en bicicleta.</s>París es una ciudad francesa."
39
  ]
40
 
41
 
42
- @st.cache(show_spinner=False, persist=True)
43
- def load_model(masked_text, model_url):
44
- model = AutoModelForMaskedLM.from_pretrained(model_url)
45
- tokenizer = AutoTokenizer.from_pretrained(model_url)
46
- nlp = pipeline("fill-mask", model=model, tokenizer=tokenizer)
47
- result = nlp(masked_text)
48
- return result
49
-
50
-
51
  @st.cache(show_spinner=False, persist=True)
52
  def load_model(masked_text, model_url):
53
  model = AutoModelForMaskedLM.from_pretrained(model_url)
@@ -64,8 +55,10 @@ def load_model_pair_classification(text, model_url_pair_classification):
64
  nlp = pipeline("text-classification", model=model, tokenizer=tokenizer)
65
  result = nlp(f"{text}</s>")
66
  if result[0]["label"] == "LABEL_0":
67
- return f"Different meaning: {result[0]['score']:02f}"
68
- return f"Paraphrase: {result[0]['score']:02f}"
 
 
69
 
70
 
71
  # Page
@@ -141,11 +134,11 @@ if st.button("Fill the mask"):
141
  st.markdown(
142
  """
143
  ### Fine-tuning to PAWS-X for paraphrase identification
144
- Here you can play with the RoBERTa Base Gaussian Seq Len 512 model fine-tuned to PAWS-X.
145
  """
146
  )
147
 
148
- pawsx_model_url = "bertin-project/bertin-base-paws-x-es"
149
  paraphrase_prompt = st.selectbox("Paraphrase Prompt", ["Random", "Custom"])
150
  if paraphrase_prompt == "Custom":
151
  paraphrase_prompt_box = "Enter two sentences separated by </s> here..."
 
34
 
35
  PAWS_X_PROMPT_LIST = [
36
  "Te amo.</s>Te adoro.",
37
+ "Te amo.</s>Te detesto.",
38
+ "Te amo.</s>Voy a caminar al campo."
39
  ]
40
 
41
 
 
 
 
 
 
 
 
 
 
42
  @st.cache(show_spinner=False, persist=True)
43
  def load_model(masked_text, model_url):
44
  model = AutoModelForMaskedLM.from_pretrained(model_url)
 
55
  nlp = pipeline("text-classification", model=model, tokenizer=tokenizer)
56
  result = nlp(f"{text}</s>")
57
  if result[0]["label"] == "LABEL_0":
58
+ return f"Entailment: {result[0]['score']:02f}"
59
+ if result[0]["label"] == "LABEL_1":
60
+ return f"Neutral: {result[0]['score']:02f}"
61
+ return f"Contradiction: {result[0]['score']:02f}"
62
 
63
 
64
  # Page
 
134
  st.markdown(
135
  """
136
  ### Fine-tuning to PAWS-X for paraphrase identification
137
+ Here you can play with the RoBERTa Base Gaussian Seq Len 512 model fine-tuned to XNLI.
138
  """
139
  )
140
 
141
+ pawsx_model_url = "bertin-project/bertin-base-xnli-es"
142
  paraphrase_prompt = st.selectbox("Paraphrase Prompt", ["Random", "Custom"])
143
  if paraphrase_prompt == "Custom":
144
  paraphrase_prompt_box = "Enter two sentences separated by </s> here..."