Update src/streamlit_app.py
Browse files- src/streamlit_app.py +39 -76
src/streamlit_app.py
CHANGED
@@ -8,28 +8,21 @@ import trafilatura
|
|
8 |
# Streamlit config
|
9 |
st.set_page_config(layout="wide", page_title="LinkBERT")
|
10 |
|
11 |
-
#
|
12 |
MODEL_ID = "dejanseo/LinkBERT-XL"
|
|
|
13 |
|
14 |
tokenizer = AutoTokenizer.from_pretrained(MODEL_ID, use_fast=True)
|
15 |
-
|
16 |
-
|
17 |
-
|
18 |
-
|
19 |
-
|
20 |
-
|
21 |
-
|
22 |
-
model = AutoModelForTokenClassification.from_pretrained(MODEL_ID)
|
23 |
-
|
24 |
-
# Explicitly move model to the determined device and dtype
|
25 |
-
if device == "cuda":
|
26 |
-
model.half().to(device) # Use .half() for float16 on GPU
|
27 |
-
else:
|
28 |
-
model.to(device) # For CPU, typically stick to float32 unless model was specifically trained with bfloat16 for CPU
|
29 |
-
|
30 |
model.eval()
|
31 |
|
32 |
-
# Functions
|
33 |
def tokenize_with_indices(text: str):
|
34 |
encoded = tokenizer.encode_plus(
|
35 |
text,
|
@@ -48,14 +41,15 @@ def fetch_and_extract_content(url: str):
|
|
48 |
return None
|
49 |
|
50 |
def process_text(inputs: str, confidence_threshold: float):
|
51 |
-
max_chunk_length = 512 - 2 #
|
52 |
words = inputs.split()
|
53 |
chunk_texts = []
|
54 |
current_chunk, current_length = [], 0
|
55 |
for word in words:
|
56 |
tok_len = len(tokenizer.tokenize(word))
|
57 |
if tok_len + current_length > max_chunk_length:
|
58 |
-
|
|
|
59 |
current_chunk = [word]
|
60 |
current_length = tok_len
|
61 |
else:
|
@@ -71,85 +65,61 @@ def process_text(inputs: str, confidence_threshold: float):
|
|
71 |
with torch.no_grad():
|
72 |
for chunk in chunk_texts:
|
73 |
input_ids, token_offsets = tokenize_with_indices(chunk)
|
74 |
-
#
|
75 |
-
input_ids_tensor = torch.tensor(input_ids).unsqueeze(0)
|
76 |
|
77 |
outputs = model(input_ids_tensor)
|
78 |
logits = outputs.logits # [1, seq_len, num_labels]
|
79 |
predictions = torch.argmax(logits, dim=-1).squeeze(0).tolist()
|
80 |
softmax_scores = F.softmax(logits, dim=-1).squeeze(0).tolist()
|
81 |
|
82 |
-
# The rest of your processing logic
|
83 |
word_info = {}
|
84 |
for idx, (start, end) in enumerate(token_offsets):
|
85 |
if idx == 0 or idx == len(token_offsets) - 1:
|
86 |
-
continue # skip
|
87 |
|
88 |
word_start = start
|
89 |
-
|
90 |
-
# This logic assumes space-separated words for the most part
|
91 |
-
while word_start > 0 and chunk[word_start - 1] not in [' ', '\n', '\t']:
|
92 |
word_start -= 1
|
93 |
-
# If a word_start maps to multiple tokens (e.g., "don't" -> ["don", "'", "t"])
|
94 |
-
# ensure we pick the earliest start for that conceptual word
|
95 |
-
while word_start > 0 and (chunk[word_start-1:word_start] == ' ' or tokenizer.decode(tokenizer.encode(chunk[word_start-1:end], add_special_tokens=False))[0] == chunk[word_start-1]):
|
96 |
-
word_start -= 1
|
97 |
|
98 |
-
# Use a tuple (word_start, actual_word_text_from_chunk) as key for more robust aggregation
|
99 |
-
# For simplicity here, we stick to word_start
|
100 |
if word_start not in word_info:
|
101 |
-
# Initialize with default for "not link"
|
102 |
word_info[word_start] = {"prediction": 0, "confidence": 0.0, "subtokens": []}
|
103 |
|
104 |
conf_pct = softmax_scores[idx][predictions[idx]] * 100.0
|
105 |
-
|
106 |
-
# Only mark as 1 if the current token's prediction is 1 AND confidence meets threshold
|
107 |
if predictions[idx] == 1 and conf_pct >= confidence_threshold:
|
108 |
-
word_info[word_start]["prediction"] = 1
|
109 |
-
|
110 |
-
# Keep the max confidence for any token within the 'word'
|
111 |
word_info[word_start]["confidence"] = max(word_info[word_start]["confidence"], conf_pct)
|
112 |
word_info[word_start]["subtokens"].append((start, end, chunk[start:end]))
|
113 |
|
114 |
last_end = 0
|
115 |
-
# Sort by word_start to maintain order
|
116 |
for word_start in sorted(word_info.keys()):
|
117 |
word_data = word_info[word_start]
|
118 |
-
|
119 |
-
for subtoken_start, subtoken_end, subtoken_text in sorted(word_data["subtokens"], key=lambda x: x[0]):
|
120 |
escaped = subtoken_text.replace("$", "\\$")
|
121 |
-
# Add any text between the last processed token and the current one
|
122 |
if last_end < subtoken_start:
|
123 |
reconstructed_text += chunk[last_end:subtoken_start]
|
124 |
-
|
125 |
if word_data["prediction"] == 1:
|
126 |
-
# Apply highlight to the subtoken
|
127 |
reconstructed_text += (
|
128 |
-
f"<span style='background-color: rgba(0, 255, 0
|
129 |
)
|
130 |
else:
|
131 |
-
reconstructed_text += escaped
|
132 |
-
|
133 |
last_end = subtoken_end
|
134 |
|
135 |
-
# For DataFrame, append the info for each *subtoken*
|
136 |
df_data["Word"].append(escaped)
|
137 |
-
df_data["Prediction"].append(word_data["prediction"])
|
138 |
-
df_data["Confidence"].append(
|
139 |
df_data["Start"].append(subtoken_start + original_position_offset)
|
140 |
df_data["End"].append(subtoken_end + original_position_offset)
|
141 |
|
142 |
-
|
143 |
-
|
144 |
-
|
145 |
-
|
146 |
-
# Update offset for the next chunk. Add 1 for the space that was implicitly there.
|
147 |
-
original_position_offset += len(chunk) + 1
|
148 |
|
149 |
df_tokens = pd.DataFrame(df_data)
|
150 |
return reconstructed_text, df_tokens
|
151 |
|
152 |
-
# UI
|
153 |
st.title("LinkBERT")
|
154 |
st.markdown("""
|
155 |
LinkBERT predicts natural link placement within web content. Enter text or a URL for extraction. Increase the threshold to reduce link predictions.
|
@@ -160,29 +130,22 @@ confidence_threshold = st.slider("Confidence Threshold", 50, 100, 50)
|
|
160 |
tab1, tab2 = st.tabs(["Text Input", "URL Input"])
|
161 |
|
162 |
with tab1:
|
163 |
-
user_input = st.text_area("Enter text to process:"
|
164 |
if st.button("Process Text"):
|
165 |
-
|
166 |
-
|
167 |
-
|
168 |
-
st.dataframe(df_tokens)
|
169 |
-
else:
|
170 |
-
st.warning("Please enter some text to process.")
|
171 |
|
172 |
with tab2:
|
173 |
-
url_input = st.text_input("Enter URL to process:"
|
174 |
if st.button("Fetch and Process"):
|
175 |
-
|
176 |
-
|
177 |
-
|
178 |
-
|
179 |
-
|
180 |
-
st.markdown(highlighted_text, unsafe_allow_html=True)
|
181 |
-
st.dataframe(df_tokens)
|
182 |
-
else:
|
183 |
-
st.error("Could not fetch content from the URL. Please check the URL and try again.")
|
184 |
else:
|
185 |
-
st.
|
186 |
|
187 |
st.divider()
|
188 |
st.markdown("""
|
@@ -202,4 +165,4 @@ LinkBERT was fine-tuned on a dataset of organic web content and editorial links.
|
|
202 |
Interested in using this in an automated pipeline for bulk link prediction?
|
203 |
|
204 |
Please [book an appointment](https://dejanmarketing.com/conference/).
|
205 |
-
""")
|
|
|
8 |
# Streamlit config
|
9 |
st.set_page_config(layout="wide", page_title="LinkBERT")
|
10 |
|
11 |
+
# Model setup (load fully to avoid meta tensors)
|
12 |
MODEL_ID = "dejanseo/LinkBERT-XL"
|
13 |
+
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
|
14 |
|
15 |
tokenizer = AutoTokenizer.from_pretrained(MODEL_ID, use_fast=True)
|
16 |
+
# Force materialized tensors on CPU, then move — avoids meta tensors
|
17 |
+
model = AutoModelForTokenClassification.from_pretrained(
|
18 |
+
MODEL_ID,
|
19 |
+
low_cpu_mem_usage=False, # important: materialize weights
|
20 |
+
torch_dtype=torch.float16 if torch.cuda.is_available() else torch.float32
|
21 |
+
)
|
22 |
+
model.to(device)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
23 |
model.eval()
|
24 |
|
25 |
+
# Functions
|
26 |
def tokenize_with_indices(text: str):
|
27 |
encoded = tokenizer.encode_plus(
|
28 |
text,
|
|
|
41 |
return None
|
42 |
|
43 |
def process_text(inputs: str, confidence_threshold: float):
|
44 |
+
max_chunk_length = 512 - 2 # leave room for specials
|
45 |
words = inputs.split()
|
46 |
chunk_texts = []
|
47 |
current_chunk, current_length = [], 0
|
48 |
for word in words:
|
49 |
tok_len = len(tokenizer.tokenize(word))
|
50 |
if tok_len + current_length > max_chunk_length:
|
51 |
+
if current_chunk:
|
52 |
+
chunk_texts.append(" ".join(current_chunk))
|
53 |
current_chunk = [word]
|
54 |
current_length = tok_len
|
55 |
else:
|
|
|
65 |
with torch.no_grad():
|
66 |
for chunk in chunk_texts:
|
67 |
input_ids, token_offsets = tokenize_with_indices(chunk)
|
68 |
+
# Build tensors on correct device; no meta usage
|
69 |
+
input_ids_tensor = torch.tensor(input_ids, dtype=torch.long, device=device).unsqueeze(0)
|
70 |
|
71 |
outputs = model(input_ids_tensor)
|
72 |
logits = outputs.logits # [1, seq_len, num_labels]
|
73 |
predictions = torch.argmax(logits, dim=-1).squeeze(0).tolist()
|
74 |
softmax_scores = F.softmax(logits, dim=-1).squeeze(0).tolist()
|
75 |
|
|
|
76 |
word_info = {}
|
77 |
for idx, (start, end) in enumerate(token_offsets):
|
78 |
if idx == 0 or idx == len(token_offsets) - 1:
|
79 |
+
continue # skip special tokens
|
80 |
|
81 |
word_start = start
|
82 |
+
while word_start > 0 and chunk[word_start - 1] != ' ':
|
|
|
|
|
83 |
word_start -= 1
|
|
|
|
|
|
|
|
|
84 |
|
|
|
|
|
85 |
if word_start not in word_info:
|
|
|
86 |
word_info[word_start] = {"prediction": 0, "confidence": 0.0, "subtokens": []}
|
87 |
|
88 |
conf_pct = softmax_scores[idx][predictions[idx]] * 100.0
|
|
|
|
|
89 |
if predictions[idx] == 1 and conf_pct >= confidence_threshold:
|
90 |
+
word_info[word_start]["prediction"] = 1
|
|
|
|
|
91 |
word_info[word_start]["confidence"] = max(word_info[word_start]["confidence"], conf_pct)
|
92 |
word_info[word_start]["subtokens"].append((start, end, chunk[start:end]))
|
93 |
|
94 |
last_end = 0
|
|
|
95 |
for word_start in sorted(word_info.keys()):
|
96 |
word_data = word_info[word_start]
|
97 |
+
for subtoken_start, subtoken_end, subtoken_text in word_data["subtokens"]:
|
|
|
98 |
escaped = subtoken_text.replace("$", "\\$")
|
|
|
99 |
if last_end < subtoken_start:
|
100 |
reconstructed_text += chunk[last_end:subtoken_start]
|
|
|
101 |
if word_data["prediction"] == 1:
|
|
|
102 |
reconstructed_text += (
|
103 |
+
f"<span style='background-color: rgba(0, 255, 0); display: inline;'>{escaped}</span>"
|
104 |
)
|
105 |
else:
|
106 |
+
reconstructed_text += escaped
|
|
|
107 |
last_end = subtoken_end
|
108 |
|
|
|
109 |
df_data["Word"].append(escaped)
|
110 |
+
df_data["Prediction"].append(word_data["prediction"])
|
111 |
+
df_data["Confidence"].append(word_info[word_start]["confidence"])
|
112 |
df_data["Start"].append(subtoken_start + original_position_offset)
|
113 |
df_data["End"].append(subtoken_end + original_position_offset)
|
114 |
|
115 |
+
original_position_offset += len(chunk) + 1
|
116 |
+
|
117 |
+
reconstructed_text += chunk[last_end:].replace("$", "\\$")
|
|
|
|
|
|
|
118 |
|
119 |
df_tokens = pd.DataFrame(df_data)
|
120 |
return reconstructed_text, df_tokens
|
121 |
|
122 |
+
# UI
|
123 |
st.title("LinkBERT")
|
124 |
st.markdown("""
|
125 |
LinkBERT predicts natural link placement within web content. Enter text or a URL for extraction. Increase the threshold to reduce link predictions.
|
|
|
130 |
tab1, tab2 = st.tabs(["Text Input", "URL Input"])
|
131 |
|
132 |
with tab1:
|
133 |
+
user_input = st.text_area("Enter text to process:")
|
134 |
if st.button("Process Text"):
|
135 |
+
highlighted_text, df_tokens = process_text(user_input, confidence_threshold)
|
136 |
+
st.markdown(highlighted_text, unsafe_allow_html=True)
|
137 |
+
st.dataframe(df_tokens)
|
|
|
|
|
|
|
138 |
|
139 |
with tab2:
|
140 |
+
url_input = st.text_input("Enter URL to process:")
|
141 |
if st.button("Fetch and Process"):
|
142 |
+
content = fetch_and_extract_content(url_input)
|
143 |
+
if content:
|
144 |
+
highlighted_text, df_tokens = process_text(content, confidence_threshold)
|
145 |
+
st.markdown(highlighted_text, unsafe_allow_html=True)
|
146 |
+
st.dataframe(df_tokens)
|
|
|
|
|
|
|
|
|
147 |
else:
|
148 |
+
st.error("Could not fetch content from the URL. Please check the URL and try again.")
|
149 |
|
150 |
st.divider()
|
151 |
st.markdown("""
|
|
|
165 |
Interested in using this in an automated pipeline for bulk link prediction?
|
166 |
|
167 |
Please [book an appointment](https://dejanmarketing.com/conference/).
|
168 |
+
""")
|