Update app.py

app.py

@@ -1,75 +1,32 @@
-import gradio as gr
-import warnings
-import torch
-from transformers import AutoProcessor, AutoModelForSpeechSeq2Seq
-import soundfile as sf
-
-warnings.filterwarnings("ignore")
-
-# Load tokenizer + model
-processor = AutoProcessor.from_pretrained("NbAiLab/nb-whisper-large-verbatim")
-model = AutoModelForSpeechSeq2Seq.from_pretrained("NbAiLab/nb-whisper-large-verbatim")
-
-# set up device
-device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
-torch_dtype = torch.float32
-
-# move model to device
-model.to(device)
-
-def transcribe_audio(audio_file, batch_size=4):
-    audio_input, sample_rate = sf.read(audio_file)
-    chunk_size = 16000 * 30 
-    chunks = [audio_input[i:i + chunk_size] for i in range(0, len(audio_input), chunk_size)]
-
-    transcription = ""
-    for i in range(0, len(chunks), batch_size):
-        batch_chunks = chunks[i:i + batch_size]
-        inputs = processor(batch_chunks, sampling_rate=16000, return_tensors="pt", padding=False)
-        inputs = inputs.to(device)
-
-        # Manually define the attention mask
-        attention_mask = torch.ones(inputs.input_features.shape[:2], dtype=torch.long)
-        attention_mask = attention_mask.to(device)
-
-        tokenizer.pad_token != tokenizer.eos_token
-
-        with torch.no_grad():
-            output = model.generate(
-                inputs.input_features,
-                max_length=1024,  # Increase max_length for longer outputs
-                num_beams=7,
-                task="transcribe",
-                attention_mask=attention_mask,
-                forced_decoder_ids=None,  # forced_decoder_ids must not be set
-                language="no"
-            )
-        transcription += " ".join(processor.batch_decode(output, skip_special_tokens=True)) + " "
-
-    return transcription.strip()
-
-
-
-
-# HTML |banner image
-banner_html = """
-<div style="text-align: center;">
-    <img src="https://huggingface.co/spaces/camparchimedes/ola_s-audioshop/resolve/main/Olas_AudioSwitch_Shop.png" width="87%" height="auto"/>
-</div>
-"""
-
-# Gradio interface
-iface = gr.Blocks()
-
-with iface:
-    gr.HTML(banner_html)
-    gr.Markdown("# 𝐍𝐯𝐢𝐝𝐢𝐚 𝐀𝟏𝟎𝟎 👋🏼👾🦾⚡ @{NbAiLab/whisper-norwegian-medium}\nUpload audio file:☕")
-    audio_input = gr.Audio(type="filepath")
-    batch_size_input = gr.Slider(minimum=1, maximum=16, step=1, label="Batch Size")
-    transcription_output = gr.Textbox()
-    transcribe_button = gr.Button("Transcribe")
-
-    transcribe_button.click(fn=transcribe_audio, inputs=[audio_input, batch_size_input], outputs=transcription_output)
-
-# Launch interface
-iface.launch(share=True, debug=True)
+def test_eos_pad():
+    from datasets import load_dataset
+    import torch
+    from transformers import GPT2Tokenizer, GPT2LMHeadModel
+
+    raw_text_batch = 'a'
+
+    tokenizer = GPT2Tokenizer.from_pretrained("gpt2")
+    # print(f'{tokenizer.eos_token=}')
+    # print(f'{tokenizer.eos_token_id=}')
+    # print(f'{tokenizer.pad_token=}')
+    # print(f'{tokenizer.pad_token_id=}')
+
+    # print(f'{raw_text_batch=}')
+    # tokenize_batch = tokenizer(raw_text_batch, padding="max_length", max_length=5, truncation=True, return_tensors="pt")
+    # print(f'{tokenize_batch=}')
+
+    if tokenizer.pad_token_id is None:
+        tokenizer.pad_token = tokenizer.eos_token
+    probe_network = GPT2LMHeadModel.from_pretrained("gpt2")
+    device = torch.device(f"cuda:{0}" if torch.cuda.is_available() else "cpu")
+    probe_network = probe_network.to(device)
+
+    print(f'{tokenizer.eos_token=}')
+    print(f'{tokenizer.eos_token_id=}')
+    print(f'{tokenizer.pad_token=}')
+    print(f'{tokenizer.pad_token_id=}')
+
+    print(f'{raw_text_batch=}')
+    tokenize_batch = tokenizer(raw_text_batch, padding="max_length", max_length=5, truncation=True, return_tensors="pt")
+    print(f'{tokenize_batch=}')
+    print('Done')