Commit History

  1. fix: switch to Hatsunama/Qwen3-4B-OBLITERATED-repaired with proper ZeroGPU module loading, reasoning mode, and optimal sampling

    3dd6c1e
  2. feat: load directly from OBLITERATUS/Qwen3-4B-OBLITERATED for instant parallel downloading

    92dd059
  3. fix: simplify predict signature, explicitly move model to cuda, enforce clean non-thinking chat

    ade5a0c
  4. feat: add default system prompt, enable_thinking toggle, optimal Qwen3 sampling params

    2e03127
  5. fix: restore corrupted embed_tokens from base Qwen3-4B and remove InfNanRemoveLogitsProcessor

    f0ff220

  1. fix(stability): load in bfloat16 and add InfNanRemoveLogitsProcessor

    deepu2135
    c911d54
  2. fix: remove unsupported type kwarg from ChatInterface and check cache

    deepu2135
    e571fd1
  3. feat: setup Qwen3-4B-OBLITERATED Gradio chatbot with ZeroGPU

    deepu2135
    53a2882
  4. initial commit

    Verified signature
    ec1fe1a