Spaces:
Sleeping
Sleeping
Update models.py
Browse files
models.py
CHANGED
@@ -13,7 +13,7 @@ You will be tasked to classify sentences as 'J' or 'V'
|
|
13 |
|
14 |
Text: "{sentence}"
|
15 |
|
16 |
-
Please classify this text as either 'J'
|
17 |
<|assistant|>
|
18 |
"""
|
19 |
return prompt
|
@@ -33,15 +33,42 @@ Please revise this text such that it maintains the criticism in the original tex
|
|
33 |
"""
|
34 |
return prompt
|
35 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
36 |
def query_model_score(sentence, api_key, model_id, prompt_fun):
|
37 |
API_URL = f"https://api-inference.huggingface.co/models/{model_id}"
|
38 |
headers = {"Authorization": f"Bearer {api_key}"}
|
39 |
prompt = prompt_fun(sentence)
|
40 |
def query(payload):
|
41 |
-
print(payload)
|
42 |
response = requests.post(API_URL, headers=headers, json=payload)
|
43 |
return response.json()
|
44 |
-
parameters = {"max_new_tokens" :
|
45 |
options = {"wait_for_model": True}
|
46 |
data = query({"inputs": f"{prompt}", "parameters": parameters, "options": options})
|
47 |
score = data[0]['generated_text']
|
@@ -57,7 +84,7 @@ def query_model_revise(sentence, api_key, model_id, prompt_fun):
|
|
57 |
def query(payload):
|
58 |
response = requests.post(API_URL, headers=headers, json=payload)
|
59 |
return response.json()
|
60 |
-
parameters = {"max_new_tokens" : 200, "temperature": 0.
|
61 |
options = {"wait_for_model": True}
|
62 |
data = query({"inputs": f"{prompt}", "parameters": parameters, "options": options})
|
63 |
revision = data[0]['generated_text']
|
@@ -75,6 +102,13 @@ def revise_review(review, api_key, model_id, highlight_color):
|
|
75 |
"message": ""
|
76 |
}
|
77 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
78 |
try:
|
79 |
review = review.replace('"', "'")
|
80 |
sentences = parser.parse_sentences(review)
|
@@ -83,13 +117,13 @@ def revise_review(review, api_key, model_id, highlight_color):
|
|
83 |
review_revision = ""
|
84 |
for sentence in sentences:
|
85 |
if len(sentence) > 20:
|
86 |
-
score = query_model_score(sentence, api_key, model_id,
|
87 |
if score == 0:
|
88 |
review_revision += " " + sentence
|
89 |
else:
|
90 |
review_score = 1
|
91 |
revision_count +=1
|
92 |
-
revision = query_model_revise(sentence, api_key, model_id,
|
93 |
revision = revision.strip().strip('"')
|
94 |
review_revision += f"<div style='background-color: {highlight_color}; display: inline;'>{revision}</div>"
|
95 |
else:
|
|
|
13 |
|
14 |
Text: "{sentence}"
|
15 |
|
16 |
+
Please classify this text as either 'J' or 'V'. Only output 'J' or 'V' with no additional explanation.<|endoftext|>
|
17 |
<|assistant|>
|
18 |
"""
|
19 |
return prompt
|
|
|
33 |
"""
|
34 |
return prompt
|
35 |
|
36 |
+
def mistral_score(sentence):
|
37 |
+
prompt = f"""<s>[INST]
|
38 |
+
You are an assistant helping with paper reviews.
|
39 |
+
You will be tasked to classify sentences as 'J' or 'V'
|
40 |
+
|
41 |
+
'J' is positive or 'J' is encouraging.
|
42 |
+
'J' has a neutral tone or 'J' is professional.
|
43 |
+
'V' is overly blunt or 'V' contains excessive negativity and no constructive feedback.
|
44 |
+
'V' contains an accusatory tone or 'V' contains sweeping generalizations or 'V' contains personal attacks.
|
45 |
+
|
46 |
+
Text: "{sentence}"
|
47 |
+
|
48 |
+
Please classify this text as either 'J' or 'V'. Only output 'J' or 'V' with no additional explanation. [/INST]"""
|
49 |
+
return prompt
|
50 |
+
|
51 |
+
def mistral_revise(sentence):
|
52 |
+
prompt = f"""<s>[INST]
|
53 |
+
You are an assistant that helps users revise Paper Reviews.
|
54 |
+
Paper reviews exist to provide authors of academic research papers constructive critism.
|
55 |
+
|
56 |
+
This is text found in a review.
|
57 |
+
This text was classified as 'toxic':
|
58 |
+
|
59 |
+
Text: "{sentence}"
|
60 |
+
|
61 |
+
Please revise this text such that it maintains the criticism in the original text and delivers it in a friendly but professional manner. Make minimal changes to the original text. [/INST] Revised Text: """
|
62 |
+
return prompt
|
63 |
+
|
64 |
def query_model_score(sentence, api_key, model_id, prompt_fun):
|
65 |
API_URL = f"https://api-inference.huggingface.co/models/{model_id}"
|
66 |
headers = {"Authorization": f"Bearer {api_key}"}
|
67 |
prompt = prompt_fun(sentence)
|
68 |
def query(payload):
|
|
|
69 |
response = requests.post(API_URL, headers=headers, json=payload)
|
70 |
return response.json()
|
71 |
+
parameters = {"max_new_tokens" : 5, "temperature": 0.1, "return_full_text": False}
|
72 |
options = {"wait_for_model": True}
|
73 |
data = query({"inputs": f"{prompt}", "parameters": parameters, "options": options})
|
74 |
score = data[0]['generated_text']
|
|
|
84 |
def query(payload):
|
85 |
response = requests.post(API_URL, headers=headers, json=payload)
|
86 |
return response.json()
|
87 |
+
parameters = {"max_new_tokens" : 200, "temperature": 0.1, "return_full_text": False}
|
88 |
options = {"wait_for_model": True}
|
89 |
data = query({"inputs": f"{prompt}", "parameters": parameters, "options": options})
|
90 |
revision = data[0]['generated_text']
|
|
|
102 |
"message": ""
|
103 |
}
|
104 |
|
105 |
+
if 'zephyr' in model_id:
|
106 |
+
revision_prompt = zephyr_revise
|
107 |
+
score_prompt = zephyr_score
|
108 |
+
elif 'mistral' in model_id:
|
109 |
+
revision_prompt = mistral_revise
|
110 |
+
score_prompt = mistral_score
|
111 |
+
|
112 |
try:
|
113 |
review = review.replace('"', "'")
|
114 |
sentences = parser.parse_sentences(review)
|
|
|
117 |
review_revision = ""
|
118 |
for sentence in sentences:
|
119 |
if len(sentence) > 20:
|
120 |
+
score = query_model_score(sentence, api_key, model_id, score_prompt)
|
121 |
if score == 0:
|
122 |
review_revision += " " + sentence
|
123 |
else:
|
124 |
review_score = 1
|
125 |
revision_count +=1
|
126 |
+
revision = query_model_revise(sentence, api_key, model_id, revision_prompt)
|
127 |
revision = revision.strip().strip('"')
|
128 |
review_revision += f"<div style='background-color: {highlight_color}; display: inline;'>{revision}</div>"
|
129 |
else:
|