Updated CUDA check to be faster on machines without CUDA capable hard…

…ware

src/gogadget/main.py

@@ -441,16 +441,6 @@ def frequency_analysis(
     )
 
 
-# @app.command(
-#     no_args_is_help=True,
-#     help=HelpText.interactive_transcript,
-#     rich_help_panel="Primary Functions",
-#     epilog=ffmpeg_warning(),
-# )
-# def interactive_transcript():
-#     return
-
-
 @app.command(
     no_args_is_help=True,
     rich_help_panel="Primary Functions",
@@ -605,15 +595,16 @@ def install(
         "Some transcriber functions may appear to freeze for a few minutes if you haven't run them before!"
     )
     CliUtils.print_status("Transcriber: Checking CUDA status")
-    transcriber = import_module(".transcriber", APP_NAME)
+    utils = import_module(".utils", APP_NAME)
 
-    cuda = transcriber.cuda_available()
+    cuda = utils.is_cuda_available()
     if cuda:
         CliUtils.print_rich("CUDA enabled, can use GPU processing")
     else:
         CliUtils.print_rich("CUDA disabled, using CPU processing")
 
     CliUtils.print_status("Transcriber: Initialising models")
+    transcriber = import_module(".transcriber", APP_NAME)
     dummy_transcribe_file = get_resources_directory() / "a.mp3"
     transcriber.transcriber(
         input_path=dummy_transcribe_file,

src/gogadget/transcriber.py

@@ -12,7 +12,7 @@
 
 from .cli_utils import CliUtils
 from .config import SUPPORTED_AUDIO_EXTS, SUPPORTED_VIDEO_EXTS
-from .utils import get_cpu_cores, list_files_with_extension
+from .utils import get_cpu_cores, is_cuda_available, list_files_with_extension
 
 
 def transcriber(
@@ -53,7 +53,7 @@ def transcriber(
     compute_type = "int8"
     device = "cpu"
     if use_gpu:
-        if cuda_available():
+        if is_cuda_available():
             device = "cuda"
             compute_type = "float16"
         else:
@@ -296,17 +296,10 @@ def reclaim_memory_gpu():
     from torch.cuda import empty_cache as empty_cuda_cache
 
     gc.collect()
-    if cuda_available():
+    if is_cuda_available():
         empty_cuda_cache()
 
 
 def reclaim_memory_cpu():
     """Force clear model from memory"""
     gc.collect()
-
-
-def cuda_available() -> bool:
-    """Check if the current system supports CUDA"""
-    from torch.cuda import is_available as is_cuda_available
-
-    return is_cuda_available()

src/gogadget/utils.py

@@ -12,6 +12,7 @@
 import pandas as pd
 
 from .cli_utils import CliUtils
+from .command_runner import get_platform, program_exists
 
 
 def get_cpu_cores(minus_one: bool = False):
@@ -173,3 +174,46 @@ def sanitise_string_html(input_string: str) -> str:
         return output_string
     else:
         return ""
+
+
+def is_cuda_available() -> bool:
+    """Check if the current system supports CUDA
+    Do a command line check first to avoid the loading times of torch"""
+
+    cli_available = check_cuda_cli_tools()
+
+    if not cli_available:
+        return False
+
+    cuda_available = check_cuda_torch()
+
+    if not cuda_available:
+        CliUtils.print_rich(
+            "CUDA not available, could not initialise using torch. If you are trying to use CUDA, do you have the CUDA specific version of torch installed?"
+        )
+
+    return cuda_available
+
+
+def check_cuda_cli_tools() -> bool:
+    """Check if the current system supports CUDA by testing for cli tools installed"""
+
+    # Macos should always return false
+    current_platform = get_platform()
+    if current_platform == "Darwin":
+        CliUtils.print_rich("CUDA not available, running on macOS")
+        return False
+
+    nvidia_smi = program_exists("nvidia-smi")
+
+    if not nvidia_smi:
+        CliUtils.print_rich("CUDA not available, nvidia-smi not found")
+
+    return nvidia_smi
+
+
+def check_cuda_torch() -> bool:
+    """Check if the current system supports CUDA using torch"""
+    from torch.cuda import is_available as is_cuda_available
+
+    return is_cuda_available()