Upload 2 files

app.py

@@ -0,0 +1,88 @@
+"""
+Huggingface Models: https://huggingface.co/models
+Transformer dependency: https://pypi.org/project/transformers/
+"""
+
+import os
+os.environ['CURL_CA_BUNDLE'] = ''
+# os.environ['CUDA_VISIBLE_DEVICES'] = ''
+os.environ['TF_FORCE_GPU_ALLOW_GROWTH'] = 'true'
+
+import malaya
+import ctranslate2
+from transformers import AutoTokenizer
+import gradio as gr
+import logging
+import os
+
+logging.basicConfig(level=logging.INFO)
+
+TO_LANG = {
+    'Malay': 'ms',
+    'English': 'en',
+}
+
+TO_LANG_KEYS = list(TO_LANG.keys())
+
+model = None
+map_lang = {
+    'en': 'Inggeris',
+    'ms': 'Melayu',
+    'pasar ms': 'pasar Melayu',
+    'manglish': 'Manglish',
+}
+
+converter = ctranslate2.converters.TransformersConverter(
+    'mesolitica/translation-t5-small-standard-bahasa-cased-v2'
+    )
+try:
+    converter.convert('t5-small-ct2', quantization='int8')
+except BaseException:
+    pass
+
+tokenizer = AutoTokenizer.from_pretrained(
+    'mesolitica/translation-t5-small-standard-bahasa-cased-v2',
+    use_fast=False,
+)
+
+
+def translate(text, to_lang):
+    global model
+    to_lang = TO_LANG[to_lang]
+    if model is None:
+        translator = ctranslate2.Translator('t5-small-ct2')
+
+    prefix = f'terjemah ke {map_lang[to_lang]}: {text}'
+    input_tokens = tokenizer.convert_ids_to_tokens(tokenizer.encode(prefix))
+    outputs = translator.translate_batch(
+        [input_tokens],
+        max_input_length=6144,
+        max_decoding_length=6144,
+        disable_unk=True,
+    )
+    results = []
+    for o in outputs:
+        o = o.hypotheses[0]
+        o = tokenizer.convert_tokens_to_ids(o)
+        results.append(o)
+    results = tokenizer.batch_decode(
+        [[i for i in o if i not in [0, 1, 2]] for o in results],
+        spaces_between_special_tokens=False,
+    )
+    return results[0]
+
+demo = gr.Interface(
+    fn=translate,
+    inputs=[
+        gr.components.Textbox(label='Input Text'),
+        gr.components.Dropdown(label='Output language', choices=TO_LANG_KEYS, value='Malay'),
+    ],
+    outputs=[
+        gr.components.Textbox(label='Output Text')
+    ],
+    cache_examples=False,
+    title='Malaysian NMT',
+    description='This translation is model able to translate malay, english, manglish to a target language. It is also able to maintain the text structure as it is and only translate necessary texts, eg, programming code.'
+)
+
+demo.launch(server_name='0.0.0.0',share=True)

requirements.txt

@@ -0,0 +1,5 @@
+malaya @ git+https://github.com/huseinzol05/malaya@5.1
+torch
+malaya-boilerplate
+ctranslate2
+gradio