Update app.py

app.py

@@ -1,4 +1,4 @@
-# app.py — Batch file transcription (up to 25 files, Icelandic forced)
+# app.py — Batch file transcription (up to 25 files, Icelandic forced, HF-safe)
 
 import os
 import gc
@@ -23,12 +23,13 @@ def transcribe_files(audio_files):
     if not audio_files:
         return None, "Hlaðið upp hljóðskrám"
 
-    audio_files = audio_files[:25]  # ✅ up to 25
+    audio_files = audio_files[:25]
 
     workdir = tempfile.mkdtemp()
     outdir = os.path.join(workdir, "transcripts")
     os.makedirs(outdir, exist_ok=True)
 
+    # Create ASR pipeline
     pipe = pipeline(
         "automatic-speech-recognition",
         model="palli23/whisper-small-sam_spjall",
@@ -36,7 +37,14 @@ def transcribe_files(audio_files):
         device=0,
     )
 
-    for idx, file in enumerate(audio_files, start=1):
+    # 🔧 PATCH generation config (critical fix)
+    gen_cfg = pipe.model.generation_config
+    gen_cfg.language = "is"
+    gen_cfg.task = "transcribe"
+    gen_cfg.forced_decoder_ids = None   # prevent conflicts
+    gen_cfg.suppress_tokens = None      # avoid tokenizer mismatch
+
+    for file in audio_files:
         audio_path = file.name
         base = os.path.splitext(os.path.basename(audio_path))[0]
         txt_path = os.path.join(outdir, f"{base}.txt")
@@ -47,8 +55,6 @@ def transcribe_files(audio_files):
             batch_size=8,
             return_timestamps=False,
             generate_kwargs={
-                "language": "is",
-                "task": "transcribe",
                 "num_beams": 5,
                 "repetition_penalty": 1.2,
                 "no_repeat_ngram_size": 3,
@@ -104,7 +110,6 @@ with gr.Blocks() as demo:
 # Launch
 # ——————————————————————————————
 demo.launch(
-    share=True,
     server_name="0.0.0.0",
     server_port=7860,
 )