Add BertGram model

.gitignore

@@ -1,2 +1 @@
 __pycache__/
-models/mbert-gram/

extract.py

@@ -55,8 +55,8 @@ def get_pagerank(importance, top_k):
 
 def summarize_text(text: str) -> str:
     return 'сыр'
-    # sentences = sent_tokenize(text)[:2000]
-    # top_k = max(20, int(len(sentences) ** .5))
-    # mat = build_matrix(sentences)
-    # idx = get_pagerank(mat, top_k)
-    # return '\n'.join(sentences[i] for i in idx)
+    sentences = sent_tokenize(text)[:2000]
+    top_k = max(20, int(len(sentences) ** .5))
+    mat = build_matrix(sentences)
+    idx = get_pagerank(mat, top_k)
+    return '\n'.join(sentences[i] for i in idx)

files/js/detection.js

@@ -119,6 +119,8 @@ function _show_summary() {
 
     summary.classList.remove('hidden');
     original.classList.add('hidden');
+
+    summaryText.focus({ preventScroll: true });
 }
 
 function _show_original() {
@@ -127,6 +129,8 @@ function _show_original() {
 
     original.classList.remove('hidden');
     summary.classList.add('hidden');
+
+    sumTextInput.focus({ preventScroll: true });
 }
 
 document.addEventListener('DOMContentLoaded', function () {

files/js/grammar.js

@@ -119,6 +119,8 @@ function _show_summary() {
 
     summary.classList.remove('hidden');
     original.classList.add('hidden');
+
+    summaryText.focus({ preventScroll: true });
 }
 
 function _show_original() {
@@ -127,6 +129,8 @@ function _show_original() {
 
     original.classList.remove('hidden');
     summary.classList.add('hidden');
+
+    sumTextInput.focus({ preventScroll: true });
 }
 
 document.addEventListener('DOMContentLoaded', function () {

files/js/summarizer.js

@@ -164,6 +164,8 @@ function _show_summary() {
 
     summary.classList.remove('hidden');
     original.classList.add('hidden');
+
+    summaryText.focus({ preventScroll: true });
 }
 
 function _show_original() {
@@ -172,6 +174,8 @@ function _show_original() {
 
     original.classList.remove('hidden');
     summary.classList.add('hidden');
+
+    sumTextInput.focus({ preventScroll: true });
 }
 
 

inference.py

@@ -1,7 +1,7 @@
 import torch
 from transformers import BertForSequenceClassification, AutoTokenizer
 
-# from extract import summarize_text as predict_summarization
+from extract import summarize_text as predict_summarization
 
 # path_emo = 'Djacon/rubert-tiny2-russian-emotion-detection'
 path_emo = './models/emotion_detection/'
@@ -27,12 +27,17 @@ def predict_emotions(text: str) -> str:
     return '\n'.join(f"{k}: {v}%" for k, v in sorted(emotions_list.items(),
                                                     key=lambda x: -x[1]))
 
-def predict_summarization(text: str) -> str:
-    return 'сыр'
+path_gram = './models/mbert-gram/'
+model_gram = BertForSequenceClassification.from_pretrained(path_gram)
+tokenizer_gram = AutoTokenizer.from_pretrained(path_gram)
 
 
-def predict_grammar(text: str) -> str:
-    return 'банан'
+@torch.no_grad()
+def predict_acceptance(text: str) -> str:
+    inputs = tokenizer_gram(text, truncation=True, return_tensors='pt')
+    output = model_gram(**inputs).logits[0]
+    pred = torch.nn.functional.softmax(output, dim=0)[1]
+    return f'Acceptance: {100 * pred.item():.2f}%'
 
 
 def test():
@@ -42,7 +47,7 @@ def test():
     predict_summarization('I am so happy now!')
     print('>>> Extractive Summarizer successfully initialized! <<<')
 
-    predict_grammar('I am so happy now!')
+    predict_acceptance('I am so happy now!')
     print('>>> Grammar Checker successfully initialized! <<<\n')
 
 

main.py

@@ -6,11 +6,11 @@ from fastapi.staticfiles import StaticFiles
 from fastapi.templating import Jinja2Templates
 
 from youtube import get_youtube_caption
-from inference import predict_emotions, predict_summarization, predict_grammar
+from inference import predict_emotions, predict_summarization, predict_acceptance
 
 MAX_ITER_SIZE = 3000
 
-app = FastAPI()
+app = FastAPI(docs_url=None, redoc_url=None)
 app.mount("/files", StaticFiles(directory="files"), name="files")
 templates = Jinja2Templates(directory="static")
 
@@ -61,4 +61,4 @@ async def predict_sum(request: EmotionRequest):
 
 @app.post('/predict_grammar')
 async def predict_gram(request: EmotionRequest):
-    return predict_grammar(request.text)
+    return predict_acceptance(request.text)

models/mbert-gram/config.json

@@ -0,0 +1,32 @@
+{
+  "_name_or_path": "bert-base-multilingual-cased",
+  "architectures": [
+    "BertForSequenceClassification"
+  ],
+  "attention_probs_dropout_prob": 0.1,
+  "classifier_dropout": null,
+  "directionality": "bidi",
+  "hidden_act": "gelu",
+  "hidden_dropout_prob": 0.1,
+  "hidden_size": 768,
+  "initializer_range": 0.02,
+  "intermediate_size": 3072,
+  "layer_norm_eps": 1e-12,
+  "max_position_embeddings": 512,
+  "model_type": "bert",
+  "num_attention_heads": 12,
+  "num_hidden_layers": 12,
+  "pad_token_id": 0,
+  "pooler_fc_size": 768,
+  "pooler_num_attention_heads": 12,
+  "pooler_num_fc_layers": 3,
+  "pooler_size_per_head": 128,
+  "pooler_type": "first_token_transform",
+  "position_embedding_type": "absolute",
+  "problem_type": "single_label_classification",
+  "torch_dtype": "float32",
+  "transformers_version": "4.27.3",
+  "type_vocab_size": 2,
+  "use_cache": true,
+  "vocab_size": 119547
+}

models/mbert-gram/pytorch_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9c99166de9b94e2c1da20c11d2f7f774d7b7b12f9a0d81f31f76514defe6747f
+size 711492725

models/mbert-gram/special_tokens_map.json

@@ -0,0 +1,7 @@
+{
+  "cls_token": "[CLS]",
+  "mask_token": "[MASK]",
+  "pad_token": "[PAD]",
+  "sep_token": "[SEP]",
+  "unk_token": "[UNK]"
+}

models/mbert-gram/tokenizer_config.json

@@ -0,0 +1,15 @@
+{
+  "cls_token": "[CLS]",
+  "do_basic_tokenize": true,
+  "do_lower_case": false,
+  "mask_token": "[MASK]",
+  "model_max_length": 512,
+  "never_split": null,
+  "pad_token": "[PAD]",
+  "sep_token": "[SEP]",
+  "special_tokens_map_file": null,
+  "strip_accents": null,
+  "tokenize_chinese_chars": true,
+  "tokenizer_class": "BertTokenizer",
+  "unk_token": "[UNK]"
+}