Init

utils/translator.py

@@ -0,0 +1,15 @@
+from transformers import AutoProcessor, SeamlessM4Tv2Model
+
+processor = AutoProcessor.from_pretrained("facebook/seamless-m4t-v2-large")
+model = SeamlessM4Tv2Model.from_pretrained("facebook/seamless-m4t-v2-large")
+
+def translator(subtitle_line: str, target_language: str) -> str:
+    # Ensure the model and processor are loaded to the GPU
+    model.to('cuda')
+    # Move input tensors to GPU
+    text_inputs = processor(text=subtitle_line, src_lang="eng", return_tensors="pt")
+    text_inputs = {key: value.to('cuda') for key, value in text_inputs.items()}
+    # Generate output tokens on GPU
+    output_tokens = model.generate(**text_inputs, tgt_lang=target_language, num_beams=5, generate_speech=False)
+    # Decode the result
+    return processor.decode(output_tokens[0].tolist()[0], skip_special_tokens=True)