Spaces:
Running
Running
Update app.py
Browse files
app.py
CHANGED
|
@@ -4,27 +4,25 @@ import os
|
|
| 4 |
from huggingface_hub import login
|
| 5 |
import torch
|
| 6 |
|
| 7 |
-
# Authenticate with Hugging Face
|
| 8 |
HUGGINGFACE_TOKEN = os.getenv("HF_TOKEN")
|
| 9 |
login(token=HUGGINGFACE_TOKEN)
|
| 10 |
|
| 11 |
-
# Load Phi-4 Mini
|
| 12 |
-
|
| 13 |
-
phi_tokenizer = AutoTokenizer.from_pretrained(
|
| 14 |
-
phi_model = AutoModelForCausalLM.from_pretrained(
|
| 15 |
-
phi_model_id, torch_dtype="auto", device_map="auto", token=HUGGINGFACE_TOKEN
|
| 16 |
-
)
|
| 17 |
phi_pipe = pipeline("text-generation", model=phi_model, tokenizer=phi_tokenizer)
|
| 18 |
|
| 19 |
-
# Load T5 for paraphrasing
|
| 20 |
t5_pipe = pipeline("text2text-generation", model="google-t5/t5-base")
|
| 21 |
|
| 22 |
-
# Load AI detector
|
| 23 |
-
|
| 24 |
-
|
| 25 |
-
|
| 26 |
|
| 27 |
-
#
|
| 28 |
def chunk_text(text, max_tokens=300):
|
| 29 |
paragraphs = text.split("\n\n")
|
| 30 |
chunks, current = [], ""
|
|
@@ -38,7 +36,7 @@ def chunk_text(text, max_tokens=300):
|
|
| 38 |
chunks.append(current.strip())
|
| 39 |
return chunks
|
| 40 |
|
| 41 |
-
#
|
| 42 |
def generate_phi_prompt(text, instruction):
|
| 43 |
chunks = chunk_text(text)
|
| 44 |
outputs = []
|
|
@@ -51,36 +49,7 @@ def generate_phi_prompt(text, instruction):
|
|
| 51 |
outputs.append(result.strip())
|
| 52 |
return "\n\n".join(outputs)
|
| 53 |
|
| 54 |
-
#
|
| 55 |
-
def detect_ai_text(text):
|
| 56 |
-
inputs = ai_tokenizer(text, return_tensors="pt", truncation=True, padding=True)
|
| 57 |
-
with torch.no_grad():
|
| 58 |
-
logits = ai_model(**inputs).logits
|
| 59 |
-
probs = torch.softmax(logits, dim=1).squeeze()
|
| 60 |
-
return probs[0].item(), probs[1].item() # human, ai
|
| 61 |
-
|
| 62 |
-
# Auto-rewrite if AI > 50%
|
| 63 |
-
def auto_rewrite_for_human(text):
|
| 64 |
-
rewritten = generate_phi_prompt(
|
| 65 |
-
text,
|
| 66 |
-
"Rewrite the following text so that it is indistinguishable from human writing and avoids AI detection. Be natural and fluent:"
|
| 67 |
-
)
|
| 68 |
-
human_score, ai_score = detect_ai_text(rewritten)
|
| 69 |
-
if ai_score < 0.05:
|
| 70 |
-
return rewritten, f"✅ Rewritten to pass detection. AI Likelihood: {round(ai_score * 100, 2)}%"
|
| 71 |
-
else:
|
| 72 |
-
return rewritten, f"⚠️ Rewritten, but AI Likelihood is still high: {round(ai_score * 100, 2)}%"
|
| 73 |
-
|
| 74 |
-
# Smart detection & rewrite combo
|
| 75 |
-
def check_and_rewrite(text):
|
| 76 |
-
human_score, ai_score = detect_ai_text(text)
|
| 77 |
-
if ai_score > 0.5:
|
| 78 |
-
rewritten, message = auto_rewrite_for_human(text)
|
| 79 |
-
return rewritten, {"Likely Human": round(human_score, 2), "Likely AI-Generated": round(ai_score, 2)}, message
|
| 80 |
-
else:
|
| 81 |
-
return text, {"Likely Human": round(human_score, 2), "Likely AI-Generated": round(ai_score, 2)}, "✅ Text is human-like. No rewrite needed."
|
| 82 |
-
|
| 83 |
-
# Tool functions
|
| 84 |
def fix_grammar(text):
|
| 85 |
return generate_phi_prompt(text, "Correct all grammar and punctuation errors in the following text. Provide only the corrected version:")
|
| 86 |
|
|
@@ -98,7 +67,22 @@ def paraphrase(text):
|
|
| 98 |
outputs.append(output)
|
| 99 |
return "\n\n".join(outputs)
|
| 100 |
|
| 101 |
-
#
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 102 |
def load_file(file_obj):
|
| 103 |
if file_obj is None:
|
| 104 |
return ""
|
|
@@ -110,15 +94,15 @@ def save_file(text):
|
|
| 110 |
f.write(text)
|
| 111 |
return path
|
| 112 |
|
| 113 |
-
#
|
| 114 |
with gr.Blocks() as demo:
|
| 115 |
-
gr.Markdown("# ✍️ AI Writing Assistant +
|
| 116 |
-
gr.Markdown("Fix grammar,
|
| 117 |
|
| 118 |
with gr.Row():
|
| 119 |
file_input = gr.File(label="📂 Upload .txt File", file_types=[".txt"])
|
| 120 |
load_btn = gr.Button("���� Load Text")
|
| 121 |
-
input_text = gr.Textbox(lines=12, label="
|
| 122 |
|
| 123 |
load_btn.click(fn=load_file, inputs=file_input, outputs=input_text)
|
| 124 |
|
|
@@ -127,20 +111,26 @@ with gr.Blocks() as demo:
|
|
| 127 |
btn_tone = gr.Button("🎯 Improve Tone")
|
| 128 |
btn_fluency = gr.Button("🔄 Improve Fluency")
|
| 129 |
btn_paraphrase = gr.Button("🌀 Paraphrase")
|
| 130 |
-
btn_detect = gr.Button("🕵️ Detect + Rewrite if AI")
|
| 131 |
|
| 132 |
output_text = gr.Textbox(lines=12, label="Output")
|
| 133 |
-
ai_output = gr.Label(label="AI Detection Result")
|
| 134 |
-
rewrite_status = gr.Textbox(label="Status Message", interactive=False)
|
| 135 |
|
| 136 |
btn_grammar.click(fn=fix_grammar, inputs=input_text, outputs=output_text)
|
| 137 |
btn_tone.click(fn=improve_tone, inputs=input_text, outputs=output_text)
|
| 138 |
btn_fluency.click(fn=improve_fluency, inputs=input_text, outputs=output_text)
|
| 139 |
btn_paraphrase.click(fn=paraphrase, inputs=input_text, outputs=output_text)
|
| 140 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 141 |
|
| 142 |
gr.Markdown("## 📤 Download Output")
|
| 143 |
-
download_btn = gr.Button("💾 Download
|
| 144 |
download_file = gr.File(label="Click to download", interactive=True)
|
| 145 |
download_btn.click(fn=save_file, inputs=output_text, outputs=download_file)
|
| 146 |
|
|
|
|
| 4 |
from huggingface_hub import login
|
| 5 |
import torch
|
| 6 |
|
| 7 |
+
# Authenticate with Hugging Face token
|
| 8 |
HUGGINGFACE_TOKEN = os.getenv("HF_TOKEN")
|
| 9 |
login(token=HUGGINGFACE_TOKEN)
|
| 10 |
|
| 11 |
+
# Load Phi-4 Mini for grammar/tone/fluency
|
| 12 |
+
phi_id = "microsoft/phi-4-mini-instruct"
|
| 13 |
+
phi_tokenizer = AutoTokenizer.from_pretrained(phi_id, token=HUGGINGFACE_TOKEN)
|
| 14 |
+
phi_model = AutoModelForCausalLM.from_pretrained(phi_id, torch_dtype="auto", device_map="auto", token=HUGGINGFACE_TOKEN)
|
|
|
|
|
|
|
| 15 |
phi_pipe = pipeline("text-generation", model=phi_model, tokenizer=phi_tokenizer)
|
| 16 |
|
| 17 |
+
# Load updated T5 model for paraphrasing
|
| 18 |
t5_pipe = pipeline("text2text-generation", model="google-t5/t5-base")
|
| 19 |
|
| 20 |
+
# Load updated AI detector
|
| 21 |
+
detector_id = "openai-community/roberta-base-openai-detector"
|
| 22 |
+
detector_tokenizer = AutoTokenizer.from_pretrained(detector_id)
|
| 23 |
+
detector_model = AutoModelForSequenceClassification.from_pretrained(detector_id)
|
| 24 |
|
| 25 |
+
# Text chunking
|
| 26 |
def chunk_text(text, max_tokens=300):
|
| 27 |
paragraphs = text.split("\n\n")
|
| 28 |
chunks, current = [], ""
|
|
|
|
| 36 |
chunks.append(current.strip())
|
| 37 |
return chunks
|
| 38 |
|
| 39 |
+
# Prompt handling
|
| 40 |
def generate_phi_prompt(text, instruction):
|
| 41 |
chunks = chunk_text(text)
|
| 42 |
outputs = []
|
|
|
|
| 49 |
outputs.append(result.strip())
|
| 50 |
return "\n\n".join(outputs)
|
| 51 |
|
| 52 |
+
# Writing tools
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 53 |
def fix_grammar(text):
|
| 54 |
return generate_phi_prompt(text, "Correct all grammar and punctuation errors in the following text. Provide only the corrected version:")
|
| 55 |
|
|
|
|
| 67 |
outputs.append(output)
|
| 68 |
return "\n\n".join(outputs)
|
| 69 |
|
| 70 |
+
# AI Detection
|
| 71 |
+
def detect_ai_percent(text):
|
| 72 |
+
inputs = detector_tokenizer(text, return_tensors="pt", truncation=True, padding=True)
|
| 73 |
+
with torch.no_grad():
|
| 74 |
+
logits = detector_model(**inputs).logits
|
| 75 |
+
probs = torch.softmax(logits, dim=1).squeeze()
|
| 76 |
+
return {
|
| 77 |
+
"Likely Human": round(probs[0].item() * 100, 2),
|
| 78 |
+
"Likely AI-Generated": round(probs[1].item() * 100, 2)
|
| 79 |
+
}
|
| 80 |
+
|
| 81 |
+
# Rewrite to lower AI score
|
| 82 |
+
def rewrite_to_human(text):
|
| 83 |
+
return generate_phi_prompt(text, "Rewrite the following text so that it is indistinguishable from human writing and avoids AI detection. Be natural and fluent:")
|
| 84 |
+
|
| 85 |
+
# File I/O
|
| 86 |
def load_file(file_obj):
|
| 87 |
if file_obj is None:
|
| 88 |
return ""
|
|
|
|
| 94 |
f.write(text)
|
| 95 |
return path
|
| 96 |
|
| 97 |
+
# UI
|
| 98 |
with gr.Blocks() as demo:
|
| 99 |
+
gr.Markdown("# ✍️ AI Writing Assistant + Detector")
|
| 100 |
+
gr.Markdown("Detect AI likelihood first. Rewrite manually if high. Fix grammar, tone, fluency, paraphrase, upload/download text.")
|
| 101 |
|
| 102 |
with gr.Row():
|
| 103 |
file_input = gr.File(label="📂 Upload .txt File", file_types=[".txt"])
|
| 104 |
load_btn = gr.Button("���� Load Text")
|
| 105 |
+
input_text = gr.Textbox(lines=12, label="Input Text")
|
| 106 |
|
| 107 |
load_btn.click(fn=load_file, inputs=file_input, outputs=input_text)
|
| 108 |
|
|
|
|
| 111 |
btn_tone = gr.Button("🎯 Improve Tone")
|
| 112 |
btn_fluency = gr.Button("🔄 Improve Fluency")
|
| 113 |
btn_paraphrase = gr.Button("🌀 Paraphrase")
|
|
|
|
| 114 |
|
| 115 |
output_text = gr.Textbox(lines=12, label="Output")
|
|
|
|
|
|
|
| 116 |
|
| 117 |
btn_grammar.click(fn=fix_grammar, inputs=input_text, outputs=output_text)
|
| 118 |
btn_tone.click(fn=improve_tone, inputs=input_text, outputs=output_text)
|
| 119 |
btn_fluency.click(fn=improve_fluency, inputs=input_text, outputs=output_text)
|
| 120 |
btn_paraphrase.click(fn=paraphrase, inputs=input_text, outputs=output_text)
|
| 121 |
+
|
| 122 |
+
gr.Markdown("## 🕵️ AI Detection")
|
| 123 |
+
detect_btn = gr.Button("Detect AI Probability")
|
| 124 |
+
ai_output = gr.Label(label="AI Detection Result")
|
| 125 |
+
detect_btn.click(fn=detect_ai_percent, inputs=input_text, outputs=ai_output)
|
| 126 |
+
|
| 127 |
+
gr.Markdown("## 🔁 Rewrite to Sound Human")
|
| 128 |
+
rewrite_btn = gr.Button("Rewrite to Reduce AI Likelihood")
|
| 129 |
+
rewritten_text = gr.Textbox(lines=12, label="Rewritten Text")
|
| 130 |
+
rewrite_btn.click(fn=rewrite_to_human, inputs=input_text, outputs=rewritten_text)
|
| 131 |
|
| 132 |
gr.Markdown("## 📤 Download Output")
|
| 133 |
+
download_btn = gr.Button("💾 Download")
|
| 134 |
download_file = gr.File(label="Click to download", interactive=True)
|
| 135 |
download_btn.click(fn=save_file, inputs=output_text, outputs=download_file)
|
| 136 |
|