May 18

1. 10:19

   fix: upgrade to Q6_K (600MB) better quality

   hugh007
   Verified signature

   82ad52b
2. 10:13

   feat: switch to MiniCPM-V-4.6-Thinking GGUF

   hugh007
   Verified signature

   0745c6f
3. 09:45

   fix: use hf_hub_download at runtime like OpenWolf-Agent

   hugh007
   Verified signature

   0677a3d
4. 09:31

   fix: pin llama-cpp-python==0.3.23, disable mmap

   hugh007
   Verified signature

   0bf1c9b
5. 09:25

   fix: use create_completion instead of create_chat_completion, reduce ctx

   hugh007
   Verified signature

   1d5ac04
6. 09:06

   fix: switch to MiniCPM3-4B (official openbmb GGUF)

   hugh007
   Verified signature

   ada47e2
7. 09:03

   fix: switch to Qwen2.5-1.5B for better llama.cpp compatibility

   hugh007
   Verified signature

   c558c22
8. 08:57

   Upload RESTART.txt with huggingface_hub

   hugh007
   Verified signature

   bbfd6af
9. 08:45

   fix: use pre-compiled llama-cpp-python wheel + model in image

   hugh007
   Verified signature

   ac7a783
10. 08:41

    fix: bake GGUF model into Docker image during build

    hugh007
    Verified signature

    43fe9e7
11. 08:27

    fix: use pre-compiled llama-server binary (zero compilation)

    hugh007
    Verified signature

    9d01ea4
12. 08:22

    fix: use ninja-build + CMAKE_ARGS for llama-cpp-python build

    hugh007
    Verified signature

    8067761
13. 07:53

    fix: add cmake+build-essential for llama-cpp

    hezu59158
    98775dc
14. 07:52

    trigger rebuild

    hezu59158
    25ca60d
15. 07:51

    fix: add libgomp1 for llama-cpp runtime

    hezu59158
    7aa5a3e
16. 07:48

    add README.md

    hezu59158
    bd130e6
17. 07:17

    opt: use pre-built wheel for faster build

    hezu59158
    1162144
18. 07:06

    init: MiniCPM-2B GGUF text Space

    Hugh
    37de7ca