Commit History

  1. Add REST API endpoints (/api/predict, /api/predict_legendary) via FastAPI + Gradio mount

    7c61ed6
  2. Match scikit-learn 1.8.0 to fix idf vectorizer unpickling

    32d330e
  3. Pin huggingface_hub==0.25.2 to fix HfFolder import error

    4d5c7a4
  4. Switch to Docker SDK — own environment, gradio 5.4.0, no broken HF base image

    b3e1864
  5. Fix: upgrade to Gradio 5.x, remove broken runtime pip hack, clean requirements

    40674db
  6. Add cascade + T2 model files, LFS tracking, working inference

    3f43552

  1. Trigger Space rebuild after dependency fix

    a581485
  2. Harden Space startup dependency pins

    4f5d7f8
  3. Pin huggingface_hub for Gradio compatibility

    f950fb2

  1. fix: always force-reinstall sentencepiece==0.1.99 at startup, bypass Docker cache

    0998b54
  2. debug: print full traceback on model load failure

    1802261
  3. fix: move endpoint to /run/predict_llm to avoid Gradio /api/ conflict

    b9af85a
  4. fix: force-install sentencepiece at startup to bypass Docker layer cache

    58104c0
  5. fix: pass HF_TOKEN when loading private model repo

    2a619e5
  6. fix: add sentencepiece for DebertaV2Tokenizer

    c422f24
  7. fix: add idx_to_code JSON map and fix path — model was failing to load

    8f4f96d
  8. fix: use DebertaV2Tokenizer directly from model repo to avoid Tiktoken error

    1865c70
  9. fix: use_fast=False for DeBERTa tokenizer, remove post-launch middleware

    a4b3f37
  10. fix: pin to Python 3.11 to avoid audioop removal in 3.13

    ee56c47
  11. fix: use transformers 4.46.3 for Python 3.13 tokenizers compatibility

    bbd0051
  12. fix torch version for Python 3.13

    8050127
  13. fix colorTo metadata

    38c8556
  14. initial space deployment

    e9ea65a