harp-dev commited on
Commit
3d9003b
·
verified ·
1 Parent(s): b4a11b5

Drop the length cap on the recording, keeping the reservation sized to it

Browse files

The model works through it in segments or chunks, so its length costs time rather than memory. ZeroGPU already refuses a request that the caller's tier or remaining allowance cannot cover, before reserving anything, and tells a free account how to get more. A cap set so that a free account could afford every request therefore only turned away callers who could. The only request refused here now is one whose reservation would outlast PyHARP's 15 minute timeout, which would stop it anyway.

Files changed (1) hide show
  1. app.py +8 -5
app.py CHANGED
@@ -52,8 +52,11 @@ def load_model(device: str):
52
  model.eval()
53
  return model
54
 
55
- # The whole recording is transcribed in one pass, so the cost grows with its length
56
- MAX_INPUT_SECONDS = 600
 
 
 
57
 
58
 
59
  def gpu_seconds(input_file):
@@ -103,10 +106,10 @@ def process_fn(input_audio_path: str) -> str:
103
  duration = sf.info(input_audio_path).duration
104
  except Exception as exc:
105
  raise gr.Error(f"Could not read the audio file: {exc}") from exc
106
- if duration > MAX_INPUT_SECONDS:
107
  raise gr.Error(
108
- f"The recording is {duration:.1f} s long. Trim it to {MAX_INPUT_SECONDS // 60} "
109
- "minutes or less."
110
  )
111
  midi_path = transcribe(input_audio_path)
112
  return midi_path
 
52
  model.eval()
53
  return model
54
 
55
+ # The recording is transcribed in overlapping segments, so the cost grows with its length
56
+ # PyHARP stops a job that runs past its timeout_s, 15 minutes by default, so a request whose
57
+ # reservation would exceed that is refused rather than started. A shorter one is left to
58
+ # ZeroGPU, which admits it when the caller's own allowance covers it.
59
+ MAX_GPU_SECONDS = 900
60
 
61
 
62
  def gpu_seconds(input_file):
 
106
  duration = sf.info(input_audio_path).duration
107
  except Exception as exc:
108
  raise gr.Error(f"Could not read the audio file: {exc}") from exc
109
+ if gpu_seconds(input_audio_path) > MAX_GPU_SECONDS:
110
  raise gr.Error(
111
+ f"The recording is {duration:.0f} s long, more than one request can transcribe in "
112
+ f"the {MAX_GPU_SECONDS // 60} minutes it may run. Trim it."
113
  )
114
  midi_path = transcribe(input_audio_path)
115
  return midi_path