Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
@@ -12,182 +12,37 @@ from datasets import load_dataset
|
|
12 |
import time
|
13 |
import fitz # PyMuPDF
|
14 |
|
15 |
-
dir_ = Path(__file__).parent
|
16 |
-
|
17 |
dataset = load_dataset("ibunescu/qa_legal_dataset_train")
|
18 |
|
19 |
-
# Load the BERT model and tokenizer
|
20 |
-
tokenizer = AutoTokenizer.from_pretrained("google-bert/bert-base-uncased")
|
21 |
-
model = AutoModelForMaskedLM.from_pretrained("google-bert/bert-base-uncased")
|
22 |
-
|
23 |
-
# Create the fill-mask pipeline
|
24 |
-
pipe = pipeline("fill-mask", model=model, tokenizer=tokenizer)
|
25 |
-
|
26 |
-
client = InferenceClient("HuggingFaceH4/zephyr-7b-beta")
|
27 |
-
|
28 |
-
def respond(
|
29 |
-
message,
|
30 |
-
history: list[tuple[str, str]],
|
31 |
-
system_message,
|
32 |
-
max_tokens,
|
33 |
-
temperature,
|
34 |
-
top_p,
|
35 |
-
):
|
36 |
-
messages = [{"role": "system", "content": system_message}]
|
37 |
-
|
38 |
-
for val in history:
|
39 |
-
if val[0]:
|
40 |
-
messages.append({"role": "user", "content": val[0]})
|
41 |
-
if val[1]:
|
42 |
-
messages.append({"role": "assistant", "content": val[1]})
|
43 |
-
|
44 |
-
messages.append({"role": "user", "content": message})
|
45 |
-
|
46 |
-
try:
|
47 |
-
response = ""
|
48 |
-
for message in client.chat_completion(
|
49 |
-
messages,
|
50 |
-
max_tokens=max_tokens,
|
51 |
-
stream=True,
|
52 |
-
temperature=temperature,
|
53 |
-
top_p=top_p,
|
54 |
-
):
|
55 |
-
token = message.choices[0].delta.content
|
56 |
-
if token is not None:
|
57 |
-
response += token
|
58 |
-
yield response, history + [(message, response)]
|
59 |
-
except Exception as e:
|
60 |
-
print(f"Error during chat completion: {e}")
|
61 |
-
yield "An error occurred during the chat completion.", history
|
62 |
-
|
63 |
-
def generate_case_outcome(prosecutor_response, defense_response):
|
64 |
-
prompt = f"Prosecutor's arguments: {prosecutor_response}\n\nDefense's arguments: {defense_response}\n\nProvide details on who won the case and why. Provide reasons for your decision and provide a link to the source of the case."
|
65 |
-
evaluation = ""
|
66 |
-
try:
|
67 |
-
for message in client.chat_completion(
|
68 |
-
[{"role": "system", "content": "You are a legal expert evaluating the details of the case presented by the prosecution and the defense."},
|
69 |
-
{"role": "user", "content": prompt}],
|
70 |
-
max_tokens=512,
|
71 |
-
stream=True,
|
72 |
-
temperature=0.6,
|
73 |
-
top_p=0.95,
|
74 |
-
):
|
75 |
-
token = message.choices[0].delta.content
|
76 |
-
if token is not None:
|
77 |
-
evaluation += token
|
78 |
-
except Exception as e:
|
79 |
-
print(f"Error during case outcome generation: {e}")
|
80 |
-
return "An error occurred during the case outcome generation."
|
81 |
return evaluation
|
82 |
|
83 |
-
def
|
84 |
-
prosecutor_count = outcome.split().count("Prosecutor")
|
85 |
-
defense_count = outcome.split().count("Defense")
|
86 |
-
if prosecutor_count > defense_count:
|
87 |
-
return "Prosecutor Wins"
|
88 |
-
elif defense_count > prosecutor_count:
|
89 |
-
return "Defense Wins"
|
90 |
-
else:
|
91 |
-
return "No clear winner"
|
92 |
|
93 |
-
|
94 |
-
|
95 |
-
|
96 |
-
background-color: #ffffff;
|
97 |
-
color: #000000;
|
98 |
-
font-family: Arial, sans-serif;
|
99 |
-
}
|
100 |
-
.gradio-container {
|
101 |
-
max-width: 1000px;
|
102 |
-
margin: 0 auto;
|
103 |
-
padding: 20px;
|
104 |
-
background-color: #ffffff;
|
105 |
-
border: 1px solid #e0e0e0;
|
106 |
-
border-radius: 8px;
|
107 |
-
box-shadow: 0 2px 5px rgba(0, 0, 0, 0.1);
|
108 |
-
}
|
109 |
-
.gr-button {
|
110 |
-
background-color: #ffffff !important;
|
111 |
-
border-color: #ffffff !important;
|
112 |
-
color: #000000 !important;
|
113 |
-
margin: 5px;
|
114 |
-
}
|
115 |
-
.gr-button:hover {
|
116 |
-
background-color: #ffffff !important;
|
117 |
-
border-color: #004085 !important;
|
118 |
-
}
|
119 |
-
.gr-input, .gr-textbox, .gr-slider, .gr-markdown, .gr-chatbox {
|
120 |
-
border-radius: 4px;
|
121 |
-
border: 1px solid #ced4da;
|
122 |
-
background-color: #ffffff !important;
|
123 |
-
color: #000000 !important;
|
124 |
-
}
|
125 |
-
.gr-input:focus, .gr-textbox:focus, .gr-slider:focus {
|
126 |
-
border-color: #ffffff;
|
127 |
-
outline: 0;
|
128 |
-
box-shadow: 0 0 0 0.2rem rgba(255, 255, 255, 1.0);
|
129 |
-
}
|
130 |
-
#flagging-button {
|
131 |
-
display: none;
|
132 |
-
}
|
133 |
-
footer {
|
134 |
-
display: none;
|
135 |
-
}
|
136 |
-
.chatbox .chat-container .chat-message {
|
137 |
-
background-color: #ffffff !important;
|
138 |
-
color: #000000 !important;
|
139 |
-
}
|
140 |
-
.chatbox .chat-container .chat-message-input {
|
141 |
-
background-color: #ffffff !important;
|
142 |
-
color: #000000 !important;
|
143 |
-
}
|
144 |
-
.gr-markdown {
|
145 |
-
background-color: #ffffff !important;
|
146 |
-
color: #000000 !important;
|
147 |
-
}
|
148 |
-
.gr-markdown h1, .gr-markdown h2, .gr-markdown h3, .gr-markdown h4, .gr-markdown h5, .gr-markdown h6, .gr-markdown p, .gr-markdown ul, .gr-markdown ol, .gr-markdown li {
|
149 |
-
color: #000000 !important;
|
150 |
-
}
|
151 |
-
.score-box {
|
152 |
-
width: 60px;
|
153 |
-
height: 60px;
|
154 |
-
display: flex;
|
155 |
-
align-items: center;
|
156 |
-
justify-content: center;
|
157 |
-
font-size: 12px;
|
158 |
-
font-weight: bold;
|
159 |
-
color: black;
|
160 |
-
margin: 5px;
|
161 |
-
}
|
162 |
-
.scroll-box {
|
163 |
-
max-height: 200px;
|
164 |
-
overflow-y: scroll;
|
165 |
-
border: 1px solid #ced4da;
|
166 |
-
padding: 10px;
|
167 |
-
border-radius: 4px;
|
168 |
}
|
169 |
"""
|
170 |
|
|
|
171 |
def chat_between_bots(system_message1, system_message2, max_tokens, temperature, top_p, history1, history2, shared_history, message):
|
172 |
response1, history1 = list(respond(message, history1, system_message1, max_tokens, temperature, top_p))[-1]
|
173 |
response2, history2 = list(respond(message, history2, system_message2, max_tokens, temperature, top_p))[-1]
|
174 |
-
shared_history.append(f"Prosecutor: {response1}")
|
175 |
-
shared_history.append(f"Defense Attorney: {response2}")
|
176 |
-
|
177 |
-
max_length = max(len(response1), len(response2))
|
178 |
-
response1 = response1[:max_length]
|
179 |
-
response2 = response2[:max_length]
|
180 |
|
181 |
-
|
182 |
-
|
183 |
-
|
184 |
-
|
|
|
|
|
|
|
|
|
185 |
|
186 |
-
def
|
187 |
-
prompt = "
|
188 |
response = ""
|
189 |
for message in client.chat_completion(
|
190 |
-
[{"role": "system", "content": "You are a legal
|
191 |
{"role": "user", "content": prompt}],
|
192 |
max_tokens=512,
|
193 |
stream=True,
|
@@ -199,86 +54,79 @@ def get_top_5_cases():
|
|
199 |
response += token
|
200 |
return response
|
201 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
202 |
def add_message(history, message):
|
203 |
for x in message["files"]:
|
204 |
-
|
205 |
-
if message["text"] is not None:
|
206 |
-
history.append((message["text"], None))
|
207 |
-
return history, gr.MultimodalTextbox(value=None, interactive=True)
|
208 |
|
209 |
-
def
|
210 |
-
|
211 |
-
|
212 |
-
|
213 |
-
|
214 |
-
|
215 |
-
|
216 |
-
|
217 |
-
|
218 |
-
def ask_about_case_outcome(shared_history, question):
|
219 |
-
prompt = f"Case Outcome: {shared_history}\n\nQuestion: {question}\n\nAnswer:"
|
220 |
response = ""
|
221 |
for message in client.chat_completion(
|
222 |
-
|
223 |
-
|
224 |
-
max_tokens=512,
|
225 |
stream=True,
|
226 |
-
temperature=0.
|
227 |
top_p=0.95,
|
228 |
):
|
229 |
token = message.choices[0].delta.content
|
230 |
if token is not None:
|
231 |
response += token
|
232 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
233 |
|
234 |
with gr.Blocks(css=custom_css) as demo:
|
235 |
history1 = gr.State([])
|
236 |
history2 = gr.State([])
|
237 |
shared_history = gr.State([])
|
238 |
-
|
239 |
-
|
|
|
240 |
with gr.Tab("Argument Evaluation"):
|
241 |
-
|
242 |
-
|
243 |
-
|
244 |
-
|
245 |
-
|
246 |
-
|
247 |
-
|
248 |
-
|
249 |
-
|
250 |
-
|
251 |
-
temperature = gr.State(0.5)
|
252 |
-
top_p = gr.State(0.95)
|
253 |
-
|
254 |
-
with gr.Row():
|
255 |
-
with gr.Column(scale=4):
|
256 |
-
prosecutor_response = gr.Textbox(label="Prosecutor's Response", interactive=True, elem_classes=["scroll-box"])
|
257 |
-
with gr.Column(scale=1):
|
258 |
-
prosecutor_score_color = gr.HTML()
|
259 |
-
|
260 |
-
with gr.Column(scale=4):
|
261 |
-
defense_response = gr.Textbox(label="Defense Attorney's Response", interactive=True, elem_classes=["scroll-box"])
|
262 |
-
with gr.Column(scale=1):
|
263 |
-
defense_score_color = gr.HTML()
|
264 |
-
|
265 |
-
outcome = gr.Textbox(label="Outcome", interactive=True, elem_classes=["scroll-box"])
|
266 |
-
|
267 |
-
with gr.Row():
|
268 |
-
submit_btn = gr.Button("Argue")
|
269 |
-
clear_btn = gr.Button("Clear and Reset")
|
270 |
-
save_btn = gr.Button("Save Conversation")
|
271 |
-
|
272 |
-
submit_btn.click(chat_between_bots, inputs=[system_message1, system_message2, max_tokens, temperature, top_p, history1, history2, shared_history, message], outputs=[prosecutor_response, defense_response, history1, history2, shared_history, outcome])
|
273 |
-
clear_btn.click(reset_conversation, outputs=[history1, history2, shared_history, prosecutor_response, defense_response, outcome])
|
274 |
-
save_btn.click(save_conversation, inputs=[history1, history2, shared_history], outputs=[history1, history2, shared_history])
|
275 |
|
276 |
-
with gr.Tab("
|
277 |
-
|
278 |
-
|
279 |
-
|
280 |
-
|
281 |
-
|
282 |
-
|
283 |
-
|
284 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
12 |
import time
|
13 |
import fitz # PyMuPDF
|
14 |
|
|
|
|
|
15 |
dataset = load_dataset("ibunescu/qa_legal_dataset_train")
|
16 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
17 |
return evaluation
|
18 |
|
19 |
+
def score_argument_from_outcome(outcome, argument):
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
20 |
|
21 |
+
if "Prosecutor" in outcome:
|
22 |
+
prosecutor_score = outcome.count("Prosecutor") * 2
|
23 |
+
if "won" in outcome and "Prosecutor" in outcome:
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
24 |
}
|
25 |
"""
|
26 |
|
27 |
+
|
28 |
def chat_between_bots(system_message1, system_message2, max_tokens, temperature, top_p, history1, history2, shared_history, message):
|
29 |
response1, history1 = list(respond(message, history1, system_message1, max_tokens, temperature, top_p))[-1]
|
30 |
response2, history2 = list(respond(message, history2, system_message2, max_tokens, temperature, top_p))[-1]
|
|
|
|
|
|
|
|
|
|
|
|
|
31 |
|
32 |
+
return response1, response2, history1, history2, shared_history, outcome, prosecutor_score_color, defense_score_color
|
33 |
+
|
34 |
+
def extract_text_from_pdf(pdf_file):
|
35 |
+
text = ""
|
36 |
+
doc = fitz.open(pdf_file)
|
37 |
+
for page in doc:
|
38 |
+
text += page.get_text()
|
39 |
+
return text
|
40 |
|
41 |
+
def ask_about_pdf(pdf_text, question):
|
42 |
+
prompt = f"PDF Content: {pdf_text}\n\nQuestion: {question}\n\nAnswer:"
|
43 |
response = ""
|
44 |
for message in client.chat_completion(
|
45 |
+
[{"role": "system", "content": "You are a legal expert answering questions based on the PDF content provided."},
|
46 |
{"role": "user", "content": prompt}],
|
47 |
max_tokens=512,
|
48 |
stream=True,
|
|
|
54 |
response += token
|
55 |
return response
|
56 |
|
57 |
+
def update_pdf_gallery_and_extract_text(pdf_files):
|
58 |
+
if len(pdf_files) > 0:
|
59 |
+
pdf_text = extract_text_from_pdf(pdf_files[0].name)
|
60 |
+
else:
|
61 |
+
pdf_text = ""
|
62 |
+
return pdf_files, pdf_text
|
63 |
+
|
64 |
def add_message(history, message):
|
65 |
for x in message["files"]:
|
66 |
+
return history, gr.MultimodalTextbox(value=None, interactive=False)
|
|
|
|
|
|
|
67 |
|
68 |
+
def bot(history):
|
69 |
+
system_message = "You are a helpful assistant."
|
70 |
+
messages = [{"role": "system", "content": system_message}]
|
71 |
+
for val in history:
|
72 |
+
if val[0]:
|
73 |
+
messages.append({"role": "user", "content": val[0]})
|
74 |
+
if val[1]:
|
75 |
+
messages.append({"role": "assistant", "content": val[1]})
|
|
|
|
|
|
|
76 |
response = ""
|
77 |
for message in client.chat_completion(
|
78 |
+
messages,
|
79 |
+
max_tokens=150,
|
|
|
80 |
stream=True,
|
81 |
+
temperature=0.6,
|
82 |
top_p=0.95,
|
83 |
):
|
84 |
token = message.choices[0].delta.content
|
85 |
if token is not None:
|
86 |
response += token
|
87 |
+
history[-1][1] = response
|
88 |
+
yield history
|
89 |
+
|
90 |
+
def print_like_dislike(x: gr.LikeData):
|
91 |
+
print(x.index, x.value, x.liked)
|
92 |
+
|
93 |
+
def reset_conversation():
|
94 |
+
return [], [], "", "", ""
|
95 |
+
|
96 |
+
def save_conversation(history1, history2, shared_history):
|
97 |
+
return history1, history2, shared_history
|
98 |
|
99 |
with gr.Blocks(css=custom_css) as demo:
|
100 |
history1 = gr.State([])
|
101 |
history2 = gr.State([])
|
102 |
shared_history = gr.State([])
|
103 |
+
pdf_files = gr.State([])
|
104 |
+
pdf_text = gr.State("")
|
105 |
+
|
106 |
with gr.Tab("Argument Evaluation"):
|
107 |
+
message = gr.Textbox(label="Case to Argue")
|
108 |
+
|
109 |
+
shared_argument = gr.Textbox(label="Case Outcome", interactive=True)
|
110 |
+
submit_btn = gr.Button("Argue")
|
111 |
+
clear_btn = gr.Button("Clear and Reset")
|
112 |
+
save_btn = gr.Button("Save Conversation")
|
113 |
+
|
114 |
+
submit_btn.click(chat_between_bots, inputs=[system_message1, system_message2, max_tokens, temperature, top_p, history1, history2, shared_history, message], outputs=[prosecutor_response, defense_response, history1, history2, shared_argument, prosecutor_score_color, defense_score_color])
|
115 |
+
clear_btn.click(reset_conversation, outputs=[history1, history2, shared_history, prosecutor_response, defense_response, shared_argument])
|
116 |
+
save_btn.click(save_conversation, inputs=[history1, history2, shared_history], outputs=[history1, history2, shared_history])
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
117 |
|
118 |
+
with gr.Tab("PDF Management"):
|
119 |
+
pdf_upload = gr.File(label="Upload Case Files (PDF)", file_types=[".pdf"])
|
120 |
+
pdf_gallery = gr.Gallery(label="PDF Gallery")
|
121 |
+
pdf_view = gr.Textbox(label="PDF Content", interactive=False, elem_classes=["scroll-box"])
|
122 |
+
pdf_question = gr.Textbox(label="Ask a Question about the PDF")
|
123 |
+
pdf_answer = gr.Textbox(label="Answer", interactive=False, elem_classes=["scroll-box"])
|
124 |
+
pdf_upload_btn = gr.Button("Update PDF Gallery")
|
125 |
+
pdf_ask_btn = gr.Button("Ask")
|
126 |
+
|
127 |
+
pdf_upload_btn.click(update_pdf_gallery_and_extract_text, inputs=[pdf_upload], outputs=[pdf_gallery, pdf_text])
|
128 |
+
pdf_text.change(fn=lambda x: x, inputs=pdf_text, outputs=pdf_view)
|
129 |
+
pdf_ask_btn.click(ask_about_pdf, inputs=[pdf_text, pdf_question], outputs=pdf_answer)
|
130 |
+
|
131 |
+
with gr.Tab("Chatbot"):
|
132 |
+
chatbot = gr.Chatbot(
|