Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
|
@@ -4,6 +4,7 @@ from transformers import AutoTokenizer, AutoModelForSequenceClassification
|
|
| 4 |
import re
|
| 5 |
import pandas as pd
|
| 6 |
import gradio as gr
|
|
|
|
| 7 |
|
| 8 |
# -----------------------------
|
| 9 |
# MODEL INITIALIZATION
|
|
@@ -11,21 +12,30 @@ import gradio as gr
|
|
| 11 |
MODEL_NAME = "desklib/ai-text-detector-v1.01"
|
| 12 |
tokenizer = None
|
| 13 |
model = None
|
| 14 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 15 |
|
| 16 |
def get_model():
|
| 17 |
global tokenizer, model
|
| 18 |
if model is None:
|
| 19 |
print(f"Loading Specialized Model: {MODEL_NAME} on {device}")
|
|
|
|
| 20 |
tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME, use_fast=False)
|
| 21 |
model = AutoModelForSequenceClassification.from_pretrained(
|
| 22 |
MODEL_NAME,
|
| 23 |
num_labels=1,
|
| 24 |
-
ignore_mismatched_sizes=True
|
|
|
|
|
|
|
| 25 |
).to(device).eval()
|
| 26 |
return tokenizer, model
|
| 27 |
|
| 28 |
-
# UPDATED: Threshold set to 59% for visual triggers
|
| 29 |
THRESHOLD = 0.59
|
| 30 |
|
| 31 |
# -----------------------------
|
|
@@ -70,9 +80,8 @@ def analyze(text):
|
|
| 70 |
text = text.strip()
|
| 71 |
if not text:
|
| 72 |
return "—", "—", "<em>Please enter text...</em>", None
|
| 73 |
-
|
| 74 |
word_count = len(text.split())
|
| 75 |
-
# Word count requirement restored to 300
|
| 76 |
if word_count < 300:
|
| 77 |
warning_msg = f"⚠️ <b>Insufficient Text:</b> Your input has {word_count} words. Please enter at least 300 words for an accurate analysis."
|
| 78 |
return "Too Short", "N/A", f"<div style='color: #b80d0d; padding: 20px; border: 1px solid #b80d0d; border-radius: 8px;'>{warning_msg}</div>", None
|
|
@@ -104,7 +113,7 @@ def analyze(text):
|
|
| 104 |
weighted_avg = sum(p * l for p, l in zip(probs, lengths)) / total_words if total_words > 0 else 0
|
| 105 |
|
| 106 |
# -----------------------------
|
| 107 |
-
# HTML RECONSTRUCTION
|
| 108 |
# -----------------------------
|
| 109 |
highlighted_html = "<div style='font-family: sans-serif; line-height: 1.8;'>"
|
| 110 |
prob_map = {idx: probs[i] for i, idx in enumerate(pure_sents_indices)}
|
|
@@ -113,10 +122,9 @@ def analyze(text):
|
|
| 113 |
if block.startswith("\n") or block.isspace():
|
| 114 |
highlighted_html += block.replace("\n", "<br>")
|
| 115 |
continue
|
| 116 |
-
|
| 117 |
if i in prob_map:
|
| 118 |
score = prob_map[i]
|
| 119 |
-
# Color determined by the new 59% threshold
|
| 120 |
if score >= THRESHOLD:
|
| 121 |
color, bg = "#b80d0d", "rgba(184, 13, 13, 0.15)" # RED
|
| 122 |
else:
|
|
@@ -129,8 +137,8 @@ def analyze(text):
|
|
| 129 |
)
|
| 130 |
else:
|
| 131 |
highlighted_html += block
|
| 132 |
-
highlighted_html += "</div>"
|
| 133 |
|
|
|
|
| 134 |
label = f"{weighted_avg:.1%} AI Probability"
|
| 135 |
display_score = f"{weighted_avg:.2%}"
|
| 136 |
|
|
@@ -142,7 +150,7 @@ def analyze(text):
|
|
| 142 |
# -----------------------------
|
| 143 |
with gr.Blocks(theme=gr.themes.Soft()) as demo:
|
| 144 |
gr.Markdown("## 🕵️ AI Detector Pro: Raw Mode")
|
| 145 |
-
gr.Markdown(f"
|
| 146 |
|
| 147 |
with gr.Row():
|
| 148 |
with gr.Column(scale=3):
|
|
@@ -151,7 +159,7 @@ with gr.Blocks(theme=gr.themes.Soft()) as demo:
|
|
| 151 |
with gr.Column(scale=1):
|
| 152 |
verdict_out = gr.Label(label="Model Verdict (Raw)")
|
| 153 |
score_out = gr.Label(label="Exact Weighted Probability")
|
| 154 |
-
|
| 155 |
with gr.Tabs():
|
| 156 |
with gr.TabItem("Visual Heatmap"):
|
| 157 |
html_out = gr.HTML()
|
|
|
|
| 4 |
import re
|
| 5 |
import pandas as pd
|
| 6 |
import gradio as gr
|
| 7 |
+
import os
|
| 8 |
|
| 9 |
# -----------------------------
|
| 10 |
# MODEL INITIALIZATION
|
|
|
|
| 12 |
MODEL_NAME = "desklib/ai-text-detector-v1.01"
|
| 13 |
tokenizer = None
|
| 14 |
model = None
|
| 15 |
+
|
| 16 |
+
# Force CPU if CUDA is not properly initialized to prevent crash
|
| 17 |
+
if torch.cuda.is_available():
|
| 18 |
+
device = torch.device("cuda")
|
| 19 |
+
dtype = torch.float16 # Half precision for GPU speed/memory
|
| 20 |
+
else:
|
| 21 |
+
device = torch.device("cpu")
|
| 22 |
+
dtype = torch.float32 # Full precision for CPU stability
|
| 23 |
|
| 24 |
def get_model():
|
| 25 |
global tokenizer, model
|
| 26 |
if model is None:
|
| 27 |
print(f"Loading Specialized Model: {MODEL_NAME} on {device}")
|
| 28 |
+
# Added low_cpu_mem_usage to prevent Build Exit Code 1 (OOM)
|
| 29 |
tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME, use_fast=False)
|
| 30 |
model = AutoModelForSequenceClassification.from_pretrained(
|
| 31 |
MODEL_NAME,
|
| 32 |
num_labels=1,
|
| 33 |
+
ignore_mismatched_sizes=True,
|
| 34 |
+
low_cpu_mem_usage=True,
|
| 35 |
+
torch_dtype=dtype
|
| 36 |
).to(device).eval()
|
| 37 |
return tokenizer, model
|
| 38 |
|
|
|
|
| 39 |
THRESHOLD = 0.59
|
| 40 |
|
| 41 |
# -----------------------------
|
|
|
|
| 80 |
text = text.strip()
|
| 81 |
if not text:
|
| 82 |
return "—", "—", "<em>Please enter text...</em>", None
|
| 83 |
+
|
| 84 |
word_count = len(text.split())
|
|
|
|
| 85 |
if word_count < 300:
|
| 86 |
warning_msg = f"⚠️ <b>Insufficient Text:</b> Your input has {word_count} words. Please enter at least 300 words for an accurate analysis."
|
| 87 |
return "Too Short", "N/A", f"<div style='color: #b80d0d; padding: 20px; border: 1px solid #b80d0d; border-radius: 8px;'>{warning_msg}</div>", None
|
|
|
|
| 113 |
weighted_avg = sum(p * l for p, l in zip(probs, lengths)) / total_words if total_words > 0 else 0
|
| 114 |
|
| 115 |
# -----------------------------
|
| 116 |
+
# HTML RECONSTRUCTION
|
| 117 |
# -----------------------------
|
| 118 |
highlighted_html = "<div style='font-family: sans-serif; line-height: 1.8;'>"
|
| 119 |
prob_map = {idx: probs[i] for i, idx in enumerate(pure_sents_indices)}
|
|
|
|
| 122 |
if block.startswith("\n") or block.isspace():
|
| 123 |
highlighted_html += block.replace("\n", "<br>")
|
| 124 |
continue
|
| 125 |
+
|
| 126 |
if i in prob_map:
|
| 127 |
score = prob_map[i]
|
|
|
|
| 128 |
if score >= THRESHOLD:
|
| 129 |
color, bg = "#b80d0d", "rgba(184, 13, 13, 0.15)" # RED
|
| 130 |
else:
|
|
|
|
| 137 |
)
|
| 138 |
else:
|
| 139 |
highlighted_html += block
|
|
|
|
| 140 |
|
| 141 |
+
highlighted_html += "</div>"
|
| 142 |
label = f"{weighted_avg:.1%} AI Probability"
|
| 143 |
display_score = f"{weighted_avg:.2%}"
|
| 144 |
|
|
|
|
| 150 |
# -----------------------------
|
| 151 |
with gr.Blocks(theme=gr.themes.Soft()) as demo:
|
| 152 |
gr.Markdown("## 🕵️ AI Detector Pro: Raw Mode")
|
| 153 |
+
gr.Markdown(f"Visual highlight triggers at **{THRESHOLD*100:.0f}%**.")
|
| 154 |
|
| 155 |
with gr.Row():
|
| 156 |
with gr.Column(scale=3):
|
|
|
|
| 159 |
with gr.Column(scale=1):
|
| 160 |
verdict_out = gr.Label(label="Model Verdict (Raw)")
|
| 161 |
score_out = gr.Label(label="Exact Weighted Probability")
|
| 162 |
+
|
| 163 |
with gr.Tabs():
|
| 164 |
with gr.TabItem("Visual Heatmap"):
|
| 165 |
html_out = gr.HTML()
|