Commit History

  1. feat: increase default max_new_tokens to 384 (max 1024) and enforce sentence completion Running

    3ace22d
  2. feat: organize catalog by vendor/size and activate gemma4-e4b and kanana-2-3b

    ed0b0be
  3. style: update markdown heading for summarizers output

    203c07b
  4. feat: prioritize Side-by-Side SLM tab and generalize input parser methods

    e78f1af
  5. feat: separate SLM execution trigger and add automatic zero-failure CPU fallback

    f06c5d9
  6. refactor: consolidate SLM summary executions into single @spaces.GPU call to prevent quota throttle

    af7ed1f
  7. feat: restore ZeroGPU CUDA execution mode with grace fallback messaging

    e3d0009
  8. feat: add ZeroGPU platform probe and route lightweight SLMs directly to CPU

    461b5d3
  9. feat: integrate ultra-lightweight hybrid SLM models for instant CPU/GPU benchmarking

    af52cc7
  10. fix: guard against CPU fallback during ZeroGPU allocation drops and reduce default tokens

    afbaaaf
  11. fix: restore missing parenthesis in AutoModelForCausalLM call

    3bbbf5d
  12. fix: eliminate module-level cuda calls and align with ZeroGPU default python runtime

    e47379e
  13. fix: attach @spaces.GPU directly to Gradio event handlers and chain execution

    1b8b3a1
  14. fix: resolve sumy ObjectDocumentModel and ZeroGPU CUDA wrapper initialization

    9cc8c20
  15. fix: remove invalid ssr argument from demo.launch

    32312b0
  16. fix: disable api_name, ssr, show_api and remove complex type hints to prevent Gradio 5.20 get_api_info crash

    5f03cb1
  17. fix: isolate error handling, fix chat history streaming, and switch default active to native bfloat16 models

    3e6e80a
  18. chore: ignore .env

    2de5589
  19. feat: implement Frontier SLM Comparison & Text Summarizers

    97d08f3