-
Fix pyannote 4.x DiarizeOutput API (no itertracks)
-
Fix pyannote diarize path: use_auth_token -> token
-
Fix pyannote: use_auth_token -> token
-
Target A100 40GB runtime (AWQ Int4 thinker, no offload)
-
AWQ Int4 + Thinker-only load to fit L4 24GB
-
Upload folder using huggingface_hub
-
L4 24GB: max_memory CPU offload + sdpa default for model load
-
Clean numpy+scipy reinstall + mandatory restart; drop numpy pin
-
Fix numpy: reinstall scipy together, remove broken cell5 check
-
Fix UnboundLocalError in attn_candidates dedup (transformers_qwen.py)
-
Remove mandatory restart after cell 3; numpy check in cell 5
-
Default GPU L4 + read HF_TOKEN from Colab Secrets
-
Force-reinstall numpy to fix mixed install (_blas_supports_fpe)
-
Pin numpy>=2.3 to match Colab scipy (_blas_supports_fpe fix)
-
Cell 5: add src to sys.path so omnisub imports even if pip -e incomplete
-
Drop flash-attn from default install (too slow)
-
Restore flash-attn + add progress logs to cells 3/4/5
-
Make flash-attn optional (faster install)
-
Pin pandas <3 to avoid pyannote/Colab conflict
-
Upload OmniSub source
-
initial commit