File size: 6,387 Bytes
83b0103
 
9c40587
83b0103
 
 
 
b64bc74
83b0103
b64bc74
 
 
83b0103
 
 
aba3d43
9c40587
 
 
b64bc74
9c40587
aba3d43
9c40587
83b0103
 
 
 
 
fdbbda2
83b0103
 
fdbbda2
83b0103
 
 
 
b64bc74
83b0103
 
 
 
 
 
b64bc74
83b0103
 
b64bc74
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
83b0103
 
 
b64bc74
83b0103
 
aba3d43
b64bc74
 
aba3d43
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
b64bc74
aba3d43
 
 
 
 
b64bc74
aba3d43
 
b64bc74
aba3d43
 
 
 
 
 
 
 
 
 
 
 
 
b64bc74
 
aba3d43
 
 
 
b64bc74
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
aba3d43
 
b64bc74
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
import gradio as gr
import torch
from transformers import RobertaTokenizerFast, RobertaForTokenClassification, AutoTokenizer
from huggingface_hub import hf_hub_download
import json
import os
from datetime import datetime
import shutil

log_path = "files/logs.jsonl"
temp_download_path = "files/download_logs.jsonl"
os.makedirs("files", exist_ok=True)

hf_model_repo = "amirghasemiveisi/surrey-nlp-pg04-best-model"
subfolder = "ner_roberta_lion_model"

model = RobertaForTokenClassification.from_pretrained(hf_model_repo, subfolder=subfolder)

tokenizer = AutoTokenizer.from_pretrained(
    "roberta-base",
    use_fast=True,
    add_prefix_space=True
)

id2label_path = hf_hub_download(repo_id=hf_model_repo, filename="id2label.json", subfolder=subfolder)
with open(id2label_path) as f:
    id2label = json.load(f)

with open("test_samples.json") as f:
    test_texts = json.load(f)

with open("test_labels.json") as f:
    test_labels = json.load(f)

options = [f"{i}: {' '.join(tokens)}" for i, tokens in enumerate(test_texts)]

def log_results(sentence_idx, sentence, words, pred_ids, true_token_labels, word_ids, feedback=None):
    timestamp = datetime.now().strftime("%Y-%m-%d %H:%M:%S")
    log = {
        "timestamp": timestamp,
        "sentence_idx": sentence_idx,
        "sentence": sentence,
        "input": words,
        "prediction": []
    }

    for i, word_id in enumerate(word_ids):
        if word_id is None or word_id >= len(words):
            continue

        try:
            true_idx = [j for j, wid in enumerate(word_ids) if wid == word_id and j < len(true_token_labels)][0]
            if true_token_labels[true_idx] == -100:
                continue

            pred_label = id2label[str(pred_ids[i])] if i < len(pred_ids) else "N/A"
            true_label = id2label[str(true_token_labels[true_idx])]
            token = words[word_id]

            log["prediction"].append({
                "token": token,
                "predicted": pred_label,
                "true": true_label
            })
        except:
            continue

    if feedback:
        log["feedback"] = feedback

    with open(log_path, "a") as f:
        f.write(json.dumps(log) + "\n")

    shutil.copy(log_path, temp_download_path)

def ner_predict_from_selection(selected_display_text):
    if not selected_display_text:
        return "Please select a test sentence", [], gr.update(value=None)

    try:
        index = int(selected_display_text.split(":")[0])
        words = test_texts[index]
        true_token_labels = test_labels[index]

        inputs = tokenizer(words, return_tensors="pt", is_split_into_words=True, truncation=True)
        word_ids = inputs.word_ids()
        with torch.no_grad():
            logits = model(**inputs).logits
        pred_ids = torch.argmax(logits, dim=2).squeeze().tolist()

        if not isinstance(pred_ids, list):
            pred_ids = [pred_ids]

        seen = set()
        highlighted_tokens = []

        for i, word_id in enumerate(word_ids):
            if word_id is None or word_id in seen:
                continue
            seen.add(word_id)

            word = words[word_id]

            true_idx = None
            for j, wid in enumerate(word_ids):
                if wid == word_id and j < len(true_token_labels):
                    true_idx = j
                    break

            if true_idx is None or true_token_labels[true_idx] == -100:
                continue

            true_id = true_token_labels[true_idx]
            pred_id = pred_ids[i] if i < len(pred_ids) else pred_ids[-1]

            true_label = id2label[str(true_id)]
            pred_label = id2label[str(pred_id)]

            if pred_label == true_label:
                highlighted_tokens.append((word, pred_label))
            else:
                highlighted_tokens.append((word, f"WRONG: {pred_label} β‰  {true_label}"))

        sentence = " ".join(words)
        log_results(index, sentence, words, pred_ids, true_token_labels, word_ids)
        return sentence, highlighted_tokens, temp_download_path

    except Exception as e:
        import traceback
        error_details = traceback.format_exc()
        print(f"Error in prediction: {error_details}")
        return f"Error processing selection: {str(e)}", [], gr.update(value=None)

def log_feedback(feedback_type, sentence_text):
    with open(log_path, "a") as f:
        f.write(json.dumps({"timestamp": datetime.now().strftime("%Y-%m-%d %H:%M:%S"), "feedback": feedback_type, "sentence": sentence_text}) + "\n")
    shutil.copy(log_path, temp_download_path)
    return

with gr.Blocks(title="RoBERTa NER with LION Optimizer - PG04 Group") as demo:
    gr.Markdown("<h1 style='text-align: center; color: #3b3b3b;'>RoBERTa NER with LION Optimizer</h1>")
    gr.Markdown("<p style='text-align: center;'>Select a test sentence and see token-level NER predictions. Feedback is logged and downloadable.</p>")

    dropdown = gr.Dropdown(choices=options, label="Select Test Sentence")
    submit_btn = gr.Button("Submit", variant="primary", elem_classes="orange-button")

    sentence_output = gr.Text(label="Original Sentence")
    prediction_output = gr.HighlightedText(label="Predicted Tags (WRONG if misclassified)")
    download_button = gr.File(label="Download Logs", interactive=True)

    with gr.Row():
        like_btn = gr.Button("πŸ‘ Like", elem_classes="green-button")
        dislike_btn = gr.Button("πŸ‘Ž Dislike", elem_classes="red-button")

    submit_btn.click(ner_predict_from_selection, inputs=dropdown, outputs=[sentence_output, prediction_output, download_button])
    like_btn.click(fn=log_feedback, inputs=[gr.Textbox(visible=False, value="like"), sentence_output], outputs=[])
    dislike_btn.click(fn=log_feedback, inputs=[gr.Textbox(visible=False, value="dislike"), sentence_output], outputs=[])

    demo.css = """
        .orange-button {
            background-color: #FFA500 !important;
            color: white !important;
            font-weight: bold;
        }
        .green-button {
            background-color: #4CAF50 !important;
            color: white !important;
            font-weight: bold;
        }
        .red-button {
            background-color: #f44336 !important;
            color: white !important;
            font-weight: bold;
        }
    """

if __name__ == "__main__":
    demo.launch(share=False)