Spaces:
Running
Running
Update app.py
Browse files
app.py
CHANGED
@@ -5,11 +5,8 @@ from datasets import load_dataset
|
|
5 |
import random
|
6 |
import re
|
7 |
|
8 |
-
# Load model and tokenizer
|
9 |
-
model_name = "abaryan/BioXP-0.5B-MedMCQA"
|
10 |
-
|
11 |
SYSTEM_PROMPT = """
|
12 |
-
You are a medical expert. Answer the medical question with careful analysis and explain why the selected option is correct in
|
13 |
Respond in the following format:
|
14 |
<answer>
|
15 |
[correct answer]
|
@@ -19,10 +16,9 @@ Respond in the following format:
|
|
19 |
</reasoning>
|
20 |
"""
|
21 |
|
|
|
22 |
model = AutoModelForCausalLM.from_pretrained(model_name)
|
23 |
tokenizer = AutoTokenizer.from_pretrained(model_name)
|
24 |
-
|
25 |
-
# Load dataset
|
26 |
dataset = load_dataset("openlifescienceai/medmcqa")
|
27 |
|
28 |
# Move model to GPU if available
|
@@ -49,11 +45,9 @@ def predict(question: str, option_a: str = "", option_b: str = "", option_c: str
|
|
49 |
temperature: float = 0.6, top_p: float = 0.9, max_tokens: int = 256):
|
50 |
|
51 |
# Determine if this is an MCQ by checking if any option is provided
|
52 |
-
# Only treat as MCQ if at least one option is non-empty
|
53 |
is_mcq = any(opt.strip() for opt in [option_a, option_b, option_c, option_d])
|
54 |
|
55 |
if is_mcq:
|
56 |
-
# Format MCQ question with only non-empty options
|
57 |
options = []
|
58 |
if option_a.strip(): options.append(f"A. {option_a}")
|
59 |
if option_b.strip(): options.append(f"B. {option_b}")
|
@@ -67,16 +61,12 @@ def predict(question: str, option_a: str = "", option_b: str = "", option_c: str
|
|
67 |
formatted_question = f"Question: {question}"
|
68 |
system_prompt = SYSTEM_PROMPT
|
69 |
|
70 |
-
# Create chat-style prompt
|
71 |
prompt = [
|
72 |
{'role': 'system', 'content': system_prompt},
|
73 |
{'role': 'user', 'content': formatted_question}
|
74 |
]
|
75 |
|
76 |
-
|
77 |
-
text = tokenizer.apply_chat_template(prompt, tokenize=False, add_generation_prompt=True)
|
78 |
-
|
79 |
-
# Tokenize and generate
|
80 |
model_inputs = tokenizer([text], return_tensors="pt").to(device)
|
81 |
|
82 |
with torch.inference_mode():
|
@@ -87,7 +77,6 @@ def predict(question: str, option_a: str = "", option_b: str = "", option_c: str
|
|
87 |
top_p=top_p,
|
88 |
)
|
89 |
|
90 |
-
# Get only the generated response
|
91 |
generated_ids = generated_ids[0, model_inputs.input_ids.shape[1]:]
|
92 |
model_response = tokenizer.decode(generated_ids, skip_special_tokens=True)
|
93 |
|
@@ -99,22 +88,21 @@ def predict(question: str, option_a: str = "", option_b: str = "", option_c: str
|
|
99 |
# Format output with evaluation if available (only for MCQs)
|
100 |
output = cleaned_response
|
101 |
|
102 |
-
if is_mcq and correct_option is not None:
|
103 |
-
|
104 |
-
|
105 |
-
|
106 |
|
107 |
-
|
108 |
-
|
109 |
-
|
110 |
-
|
111 |
-
|
112 |
-
|
113 |
-
|
114 |
|
115 |
return output
|
116 |
|
117 |
-
# Create Gradio interface with mobile-optimized design
|
118 |
with gr.Blocks(
|
119 |
title="BioXP Medical MCQ Assistant",
|
120 |
theme=gr.themes.Soft(
|
@@ -132,7 +120,6 @@ with gr.Blocks(
|
|
132 |
|
133 |
with gr.Row():
|
134 |
with gr.Column(scale=1):
|
135 |
-
# Input fields with mobile-friendly spacing
|
136 |
question = gr.Textbox(
|
137 |
label="Medical Question",
|
138 |
placeholder="Enter your medical question here...",
|
@@ -141,7 +128,6 @@ with gr.Blocks(
|
|
141 |
elem_classes=["mobile-input"]
|
142 |
)
|
143 |
|
144 |
-
# Options in a mobile-friendly accordion
|
145 |
with gr.Accordion("Options", open=True):
|
146 |
option_a = gr.Textbox(
|
147 |
label="Option A",
|
@@ -168,7 +154,6 @@ with gr.Blocks(
|
|
168 |
elem_classes=["mobile-input"]
|
169 |
)
|
170 |
|
171 |
-
# Generation parameters in a collapsible section
|
172 |
with gr.Accordion("Advanced Settings", open=False):
|
173 |
with gr.Row():
|
174 |
with gr.Column(scale=1):
|
@@ -202,13 +187,11 @@ with gr.Blocks(
|
|
202 |
correct_option = gr.Number(visible=False)
|
203 |
expert_explanation = gr.Textbox(visible=False)
|
204 |
|
205 |
-
# Buttons with mobile-friendly spacing
|
206 |
with gr.Row():
|
207 |
predict_btn = gr.Button("Get Answer", variant="primary", size="lg", elem_classes=["mobile-button"])
|
208 |
random_btn = gr.Button("Random Question", variant="secondary", size="lg", elem_classes=["mobile-button"])
|
209 |
|
210 |
with gr.Column(scale=1):
|
211 |
-
# Output with mobile-friendly styling
|
212 |
output = gr.Textbox(
|
213 |
label="Model's Response",
|
214 |
lines=12,
|
@@ -232,10 +215,8 @@ with gr.Blocks(
|
|
232 |
outputs=[question, option_a, option_b, option_c, option_d, correct_option, expert_explanation]
|
233 |
)
|
234 |
|
235 |
-
# Add mobile-optimized CSS
|
236 |
gr.HTML("""
|
237 |
<style>
|
238 |
-
/* Mobile-friendly base styles */
|
239 |
.container {
|
240 |
max-width: 100%;
|
241 |
padding: 0.5rem;
|
@@ -258,7 +239,6 @@ with gr.Blocks(
|
|
258 |
font-weight: 500;
|
259 |
}
|
260 |
|
261 |
-
/* Response box styling */
|
262 |
.response-box {
|
263 |
font-family: 'Inter', sans-serif;
|
264 |
line-height: 1.6;
|
|
|
5 |
import random
|
6 |
import re
|
7 |
|
|
|
|
|
|
|
8 |
SYSTEM_PROMPT = """
|
9 |
+
You are a medical expert. Answer the medical question with careful analysis and explain why the selected option is correct in 2 sentences without repeating.
|
10 |
Respond in the following format:
|
11 |
<answer>
|
12 |
[correct answer]
|
|
|
16 |
</reasoning>
|
17 |
"""
|
18 |
|
19 |
+
model_name = "abaryan/BioXP-0.5B-MedMCQA"
|
20 |
model = AutoModelForCausalLM.from_pretrained(model_name)
|
21 |
tokenizer = AutoTokenizer.from_pretrained(model_name)
|
|
|
|
|
22 |
dataset = load_dataset("openlifescienceai/medmcqa")
|
23 |
|
24 |
# Move model to GPU if available
|
|
|
45 |
temperature: float = 0.6, top_p: float = 0.9, max_tokens: int = 256):
|
46 |
|
47 |
# Determine if this is an MCQ by checking if any option is provided
|
|
|
48 |
is_mcq = any(opt.strip() for opt in [option_a, option_b, option_c, option_d])
|
49 |
|
50 |
if is_mcq:
|
|
|
51 |
options = []
|
52 |
if option_a.strip(): options.append(f"A. {option_a}")
|
53 |
if option_b.strip(): options.append(f"B. {option_b}")
|
|
|
61 |
formatted_question = f"Question: {question}"
|
62 |
system_prompt = SYSTEM_PROMPT
|
63 |
|
|
|
64 |
prompt = [
|
65 |
{'role': 'system', 'content': system_prompt},
|
66 |
{'role': 'user', 'content': formatted_question}
|
67 |
]
|
68 |
|
69 |
+
text = tokenizer.apply_chat_template(prompt, tokenize=False, add_generation_prompt=True)
|
|
|
|
|
|
|
70 |
model_inputs = tokenizer([text], return_tensors="pt").to(device)
|
71 |
|
72 |
with torch.inference_mode():
|
|
|
77 |
top_p=top_p,
|
78 |
)
|
79 |
|
|
|
80 |
generated_ids = generated_ids[0, model_inputs.input_ids.shape[1]:]
|
81 |
model_response = tokenizer.decode(generated_ids, skip_special_tokens=True)
|
82 |
|
|
|
88 |
# Format output with evaluation if available (only for MCQs)
|
89 |
output = cleaned_response
|
90 |
|
91 |
+
# if is_mcq and correct_option is not None:
|
92 |
+
# correct_letter = chr(65 + correct_option)
|
93 |
+
# answer_match = re.search(r"Answer:\s*([A-D])", cleaned_response, re.IGNORECASE)
|
94 |
+
# model_answer = answer_match.group(1).upper() if answer_match else "Not found"
|
95 |
|
96 |
+
# is_correct = model_answer == correct_letter
|
97 |
+
# output += f"\n\n---\nEvaluation:\n"
|
98 |
+
# output += f"Correct Answer: {correct_letter}\n"
|
99 |
+
# output += f"Model's Answer: {model_answer}\n"
|
100 |
+
# output += f"Result: {'✅ Correct' if is_correct else '❌ Incorrect'}\n"
|
101 |
+
# if explanation:
|
102 |
+
# output += f"\nExpert Explanation:\n{explanation}"
|
103 |
|
104 |
return output
|
105 |
|
|
|
106 |
with gr.Blocks(
|
107 |
title="BioXP Medical MCQ Assistant",
|
108 |
theme=gr.themes.Soft(
|
|
|
120 |
|
121 |
with gr.Row():
|
122 |
with gr.Column(scale=1):
|
|
|
123 |
question = gr.Textbox(
|
124 |
label="Medical Question",
|
125 |
placeholder="Enter your medical question here...",
|
|
|
128 |
elem_classes=["mobile-input"]
|
129 |
)
|
130 |
|
|
|
131 |
with gr.Accordion("Options", open=True):
|
132 |
option_a = gr.Textbox(
|
133 |
label="Option A",
|
|
|
154 |
elem_classes=["mobile-input"]
|
155 |
)
|
156 |
|
|
|
157 |
with gr.Accordion("Advanced Settings", open=False):
|
158 |
with gr.Row():
|
159 |
with gr.Column(scale=1):
|
|
|
187 |
correct_option = gr.Number(visible=False)
|
188 |
expert_explanation = gr.Textbox(visible=False)
|
189 |
|
|
|
190 |
with gr.Row():
|
191 |
predict_btn = gr.Button("Get Answer", variant="primary", size="lg", elem_classes=["mobile-button"])
|
192 |
random_btn = gr.Button("Random Question", variant="secondary", size="lg", elem_classes=["mobile-button"])
|
193 |
|
194 |
with gr.Column(scale=1):
|
|
|
195 |
output = gr.Textbox(
|
196 |
label="Model's Response",
|
197 |
lines=12,
|
|
|
215 |
outputs=[question, option_a, option_b, option_c, option_d, correct_option, expert_explanation]
|
216 |
)
|
217 |
|
|
|
218 |
gr.HTML("""
|
219 |
<style>
|
|
|
220 |
.container {
|
221 |
max-width: 100%;
|
222 |
padding: 0.5rem;
|
|
|
239 |
font-weight: 500;
|
240 |
}
|
241 |
|
|
|
242 |
.response-box {
|
243 |
font-family: 'Inter', sans-serif;
|
244 |
line-height: 1.6;
|