Commit History

  1. bugfix: set device to xpu by mistake

    1242438
  2. add all qwen3 variants

    2882063
  3. user can define search timeout

    e2ee907
  4. give 5 second for web earch to gather reults

    c00d442
  5. support thinking models and streamingly display thought

    8c3c2b9
  6. do not preview prompt at error return from chat response

    c09049b
  7. inject assistant placeholder at right time

    12dd3f3
  8. disable L142 which is not needed

    3c176a1
  9. fix bug in prompt preview display

    41ee8bf
  10. add prompt preview for debug

    5fc0117
  11. fix: prevent self-talking issue by using tokenizer chat_template formatting⁠

    960db60

  1. bugfix to Error: "str" object has no attribute "pad_token_id"

    889f080
  2. add taiwan elm 1.1b & 270m instruct

    c8399e3

  1. add type in qwen3 0.6b repo id

    76d4d60
  2. add qwen3

    fe395ab

  1. Add Smollm2 360m instruct fine-tuned on TaiwanChat

    7308211

  1. keep debug message

    37f7787
  2. add debug to show web resarch result

    a2f07a4
  3. give 1 second for web search to grab data

    9ad3ffd
  4. inject web search result if web search enabled

    bc257ff
  5. refactor(app): improve streaming, background search, dtype fallback, and cleanup :contentReference[oaicite:0]{index=0}⁠

    293686e
  6. bugfixc: not using pipeline for response generation

    939895d
  7. Add original SmolLM2 135M Instruct for comparaison

    423dc1a
  8. Add SmolLM2-135M-Instruct-TaiwanChat

    38fcc03
  9. Add SmolLM2-135M TaiwanChat

    0d642b7

  1. default to gemma-3-4b

    88a6a62
  2. model repo_id typo fix

    89372fa
  3. enable web search by default

    6235e63
  4. remove tinyllama which has bad response quality

    a22cf42
  5. make streaming response

    5ea073d
  6. apply history flatten before it goint to prompt

    ef361b0
  7. better management on system prompt

    5f6306a
  8. usue chat pipeline instead of model and tokenizer individually

    ac8e9cc
  9. bugfix to padding-related issues

    f248fec
  10. add attention mask

    b6b3940
  11. Clean model description

    4731160
  12. pin torch to 2.4.0

    4c6b4c5
  13. enable zerogpu with decorator

    c5b2897
  14. Apply ZeroGPU

    d181b45
  15. switch to gradio version for stability reason

    a703203

  1. add taiwan tinyllama

    794ee70
  2. Improve responsiveness by asynchronously retrieving web search context⁠

    acda3f1
  3. add llama 3.2 taiwan 3b

    b1544e2
  4. add minicpm3 4b

    f5c0811
  5. increase xt length to max

    629495e
  6. remove all moe

    fafc8cb
  7. remove qwen 1.5 moe

    6735035
  8. adjust title style

    e9559bd
  9. use another version of qwen 1.5 moe

    96e60d6
  10. add Qwen1.5-MoE

    e17afaf