Spaces:
Runtime error
Runtime error
Update app.py
Browse files
app.py
CHANGED
@@ -19,14 +19,13 @@ model.eval()
|
|
19 |
clip_model = CLIPModel.from_pretrained("openai/clip-vit-base-patch32")
|
20 |
clip_processor = CLIPProcessor.from_pretrained("openai/clip-vit-base-patch32")
|
21 |
|
22 |
-
# ====
|
23 |
explainer = LimeTextExplainer(class_names=["Real", "Fake"])
|
24 |
|
25 |
-
# ==== Groq Client
|
26 |
-
|
27 |
-
groq_client = Groq(api_key=groq_api_key)
|
28 |
|
29 |
-
# ====
|
30 |
def classify_news(text):
|
31 |
inputs = tokenizer(text, return_tensors="pt", truncation=True, padding=True)
|
32 |
with torch.no_grad():
|
@@ -34,17 +33,17 @@ def classify_news(text):
|
|
34 |
probs = torch.nn.functional.softmax(outputs.logits, dim=-1)
|
35 |
return probs[0].tolist()
|
36 |
|
37 |
-
# ====
|
38 |
def explain_prediction(text):
|
39 |
def predictor(texts):
|
40 |
return [classify_news(t) for t in texts]
|
41 |
explanation = explainer.explain_instance(text, predictor, num_features=6)
|
42 |
return explanation.as_list()
|
43 |
|
44 |
-
# ====
|
45 |
def get_llm_opinion(text):
|
46 |
try:
|
47 |
-
response =
|
48 |
model="mixtral-8x7b-32768",
|
49 |
messages=[
|
50 |
{"role": "system", "content": "You are a helpful assistant that detects fake news based on user input."},
|
@@ -55,7 +54,7 @@ def get_llm_opinion(text):
|
|
55 |
except Exception as e:
|
56 |
return f"LLM Error: {str(e)}"
|
57 |
|
58 |
-
# ====
|
59 |
def verify_image_with_text(image, text):
|
60 |
if image is None:
|
61 |
return "No image uploaded."
|
@@ -77,7 +76,6 @@ def analyze_news(text, image=None):
|
|
77 |
llm_verdict = get_llm_opinion(text)
|
78 |
img_verification = verify_image_with_text(image, text)
|
79 |
|
80 |
-
# JSON Report
|
81 |
report = {
|
82 |
"model_prediction": model_label,
|
83 |
"bert_probs": {"Real": round(prediction[0], 3), "Fake": round(prediction[1], 3)},
|
|
|
19 |
clip_model = CLIPModel.from_pretrained("openai/clip-vit-base-patch32")
|
20 |
clip_processor = CLIPProcessor.from_pretrained("openai/clip-vit-base-patch32")
|
21 |
|
22 |
+
# ==== LIME Explainer ====
|
23 |
explainer = LimeTextExplainer(class_names=["Real", "Fake"])
|
24 |
|
25 |
+
# ==== Groq Client ====
|
26 |
+
client = Groq(api_key=os.environ.get("GROQ_API_KEY"))
|
|
|
27 |
|
28 |
+
# ==== News Classification ====
|
29 |
def classify_news(text):
|
30 |
inputs = tokenizer(text, return_tensors="pt", truncation=True, padding=True)
|
31 |
with torch.no_grad():
|
|
|
33 |
probs = torch.nn.functional.softmax(outputs.logits, dim=-1)
|
34 |
return probs[0].tolist()
|
35 |
|
36 |
+
# ==== LIME Explanation ====
|
37 |
def explain_prediction(text):
|
38 |
def predictor(texts):
|
39 |
return [classify_news(t) for t in texts]
|
40 |
explanation = explainer.explain_instance(text, predictor, num_features=6)
|
41 |
return explanation.as_list()
|
42 |
|
43 |
+
# ==== LLM Second Opinion ====
|
44 |
def get_llm_opinion(text):
|
45 |
try:
|
46 |
+
response = client.chat.completions.create(
|
47 |
model="mixtral-8x7b-32768",
|
48 |
messages=[
|
49 |
{"role": "system", "content": "You are a helpful assistant that detects fake news based on user input."},
|
|
|
54 |
except Exception as e:
|
55 |
return f"LLM Error: {str(e)}"
|
56 |
|
57 |
+
# ==== CLIP Image-Text Similarity ====
|
58 |
def verify_image_with_text(image, text):
|
59 |
if image is None:
|
60 |
return "No image uploaded."
|
|
|
76 |
llm_verdict = get_llm_opinion(text)
|
77 |
img_verification = verify_image_with_text(image, text)
|
78 |
|
|
|
79 |
report = {
|
80 |
"model_prediction": model_label,
|
81 |
"bert_probs": {"Real": round(prediction[0], 3), "Fake": round(prediction[1], 3)},
|