Hugging Face's logo

Spaces:
build-small-hackathon
/
phantom-grid
Paused

App Files Community
phantom-grid / llm

Commit History

  1. Strip MiniCPM <think>...</think> reasoning tags from generated text

    unity4ar
    Verified signature
    5f5ab47

  2. Pin transformers<5.0 so KV cache works with MiniCPM bundled code; re-enable use_cache

    unity4ar
    Verified signature
    148a99f

  3. Revert to use_cache=False (eager attn also broken); hide all audio UI

    unity4ar
    Verified signature
    df99cf8

  4. Speed up: eager attn + KV cache; drop chat retries to 1; remove MiniCPM-o voice UI artifact

    unity4ar
    Verified signature
    3326f32

  5. Cap zerogpu max_new_tokens at 256 (use_cache=False makes long generations O(n^2))

    unity4ar
    Verified signature
    c788ee1

  6. Disable KV cache: openbmb modeling_minicpm.py has a cache_utils API drift bug

    unity4ar
    Verified signature
    e06599a

  7. Move .to('cuda') inside @spaces.GPU; background thread keeps model on CPU to avoid emulation bypass

    unity4ar
    Verified signature
    5a8fab5

  8. Shim is_torch_fx_available so MiniCPM trust_remote_code import works on transformers >= 5.0

    unity4ar
    Verified signature
    90e360d

  9. Use canonical .to('cuda') pattern + progress logs so container log shows what loader is doing

    unity4ar
    Verified signature
    031ce2d

  10. Load model in background thread so health/status endpoints don't block on 16GB download

    unity4ar
    Verified signature
    fbd952d

  11. Load model on cuda at module level (canonical ZeroGPU pattern)

    unity4ar
    Verified signature
    a475083

  12. Refactor: Docker+llama.cpp -> Gradio SDK + ZeroGPU transformers backend

    unity4ar
    Verified signature
    7036a02

  13. Ship Phantom Grid Docker Space

    unity4ar
    Verified signature
    d2e6f94