Add files via upload

Author: BBC-Esq

settings.py

@@ -13,15 +13,13 @@ def initUI(self):
 
         hbox1_layout = QHBoxLayout()
 
-        # Replace size and quantization combo boxes with a single model combo box
         modelLabel = QLabel("Model:")
         hbox1_layout.addWidget(modelLabel)
 
         self.modelComboBox = QComboBox()
         self.modelComboBox.addItems(WHISPER_MODELS.keys())
         hbox1_layout.addWidget(self.modelComboBox)
 
-        # Keep the rest of the widgets
         computeDeviceLabel = QLabel("Device:")
         hbox1_layout.addWidget(computeDeviceLabel)
 
@@ -50,13 +48,13 @@ def initUI(self):
 
         self.beamSizeSlider = QSlider(Qt.Horizontal)
         self.beamSizeSlider.setMinimum(1)
-        self.beamSizeSlider.setMaximum(10)
-        self.beamSizeSlider.setValue(5)
+        self.beamSizeSlider.setMaximum(5)
+        self.beamSizeSlider.setValue(1)
         self.beamSizeSlider.setTickPosition(QSlider.TicksBelow)
         self.beamSizeSlider.setTickInterval(1)
         beam_size_layout.addWidget(self.beamSizeSlider)
 
-        self.beamSizeValueLabel = QLabel("5")
+        self.beamSizeValueLabel = QLabel("1")
         beam_size_layout.addWidget(self.beamSizeValueLabel)
         self.beamSizeSlider.valueChanged.connect(lambda: self.update_slider_label(self.beamSizeSlider, self.beamSizeValueLabel))
 

utilities.py

@@ -1,5 +1,6 @@
 import torch
 import ctranslate2
+import psutil
 
 def get_compute_and_platform_info():
     available_devices = ["cpu"]
@@ -9,10 +10,14 @@ def get_compute_and_platform_info():
 
     return available_devices
 
+
 def get_supported_quantizations(device_type):
     types = ctranslate2.get_supported_compute_types(device_type)
     filtered_types = [q for q in types if q != 'int16']
     desired_order = ['float32', 'float16', 'bfloat16', 'int8_float32', 'int8_float16', 'int8_bfloat16', 'int8']
     sorted_types = [q for q in desired_order if q in filtered_types]
     return sorted_types
 
+
+def get_physical_core_count():
+    return psutil.cpu_count(logical=False)

whispers2t_batch_gui.py

@@ -4,7 +4,7 @@
 from PySide6.QtWidgets import QApplication, QWidget, QVBoxLayout, QHBoxLayout, QPushButton, QFileDialog, QCheckBox, QLabel, QGroupBox, QMessageBox
 from PySide6.QtCore import Qt
 import torch    
-from utilities import get_compute_and_platform_info
+from utilities import get_compute_and_platform_info, get_supported_quantizations
 from whispers2t_batch_transcriber import Worker
 from metrics_bar import MetricsBar
 from settings import SettingsGroupBox
@@ -42,7 +42,7 @@ def initUI(self):
 
         main_layout = QVBoxLayout()
 
-        transcriberGroupBox = QGroupBox("Batch Transcriber (ctranslate2 edition - huggingface edition coming soon)")
+        transcriberGroupBox = QGroupBox("Batch Transcriber (ctranslate2 edition)")
         transcriberLayout = QVBoxLayout()
 
         self.dirLabel = QLabel("No directory selected")
@@ -121,10 +121,57 @@ def calculate_files_to_process(self):
                 total_files += len(list(directory_path.glob(pattern)))
         return total_files
 
+    def perform_checks(self):
+        model = self.settingsGroupBox.modelComboBox.currentText()
+        device = self.settingsGroupBox.computeDeviceComboBox.currentText()
+        batch_size = self.settingsGroupBox.batchSizeSlider.value()
+        beam_size = self.settingsGroupBox.beamSizeSlider.value()
+
+        # Check 1: CPU and non-float32 model
+        if "float32" not in model.lower() and device.lower() == "cpu":
+            QMessageBox.warning(self, "Invalid Configuration", 
+                                "CPU only supports Float 32 computation. Please select a different Whisper model.")
+            return False
+
+        # Check 2: CPU with high batch size
+        if device.lower() == "cpu" and batch_size > 16:
+            reply = QMessageBox.warning(self, "Performance Warning", 
+                "When using CPU it is generally recommended to use a batch size of no more than 16 "
+                "otherwise compute time will actually be worse.\n\n"
+                "Moreover, if you select a Beam Size greater than one, you should reduce the Batch Size accordingly.\n\n"
+                "For example:\n"
+                "- If you select a Beam Size of 2 (double the default value of 1) you would reduce the Batch Size (default value 16) by half.\n"
+                "- If Beam Size is set to 3 you should reduce the Batch Size to 1/3 of the default level, and so on.\n\nClick OK to proceed.",
+                QMessageBox.Ok | QMessageBox.Cancel,
+                QMessageBox.Cancel)
+            if reply == QMessageBox.Cancel:
+                return False
+
+        # Check 3: GPU compatibility
+        # Only perform this check if the device is not CPU
+        if device.lower() != "cpu":
+            supported_quantizations = get_supported_quantizations(device)
+            if "float16" in model.lower() and "float16" not in supported_quantizations:
+                QMessageBox.warning(self, "Incompatible Configuration", 
+                                    "Your GPU does not support the selected floating point value (float16). "
+                                    "Please make another selection.")
+                return False
+            if "bfloat16" in model.lower() and "bfloat16" not in supported_quantizations:
+                QMessageBox.warning(self, "Incompatible Configuration", 
+                                    "Your GPU does not support the selected floating point value (bfloat16). "
+                                    "Please make another selection.")
+                return False
+
+        return True  # All checks passed
+
     def processFiles(self):
         if hasattr(self, 'directory'):
             total_files = self.calculate_files_to_process()
 
+            # Perform checks
+            if not self.perform_checks():
+                return
+
             reply = QMessageBox.question(
                 self, 
                 'Confirm Process', 

whispers2t_batch_transcriber.py

@@ -1,13 +1,18 @@
 import os
 import sys
 import gc
-from PySide6.QtCore import QThread, Signal, QElapsedTimer
 from pathlib import Path
-import whisper_s2t
+from threading import Event
 from queue import Queue
+
+from PySide6.QtCore import QThread, Signal, QElapsedTimer
+import whisper_s2t
 import torch
-from threading import Event
+
 from constants import WHISPER_MODELS
+from utilities import get_physical_core_count
+
+CPU_THREADS = max(4, get_physical_core_count() - 1)
 
 class Worker(QThread):
     finished = Signal(str)
@@ -66,7 +71,7 @@ def run(self):
             device=self.device,
             compute_type=self.model_info['precision'],
             asr_options={'beam_size': self.beam_size},
-            cpu_threads=max(4, os.cpu_count() - 6) if self.device == "cpu" else 4,
+            cpu_threads=CPU_THREADS if self.device == "cpu" else 4,
             **model_kwargs
         )