[llama.cpp] AI answer does not stop automatically when inference is launched on Android phone

[zhouwg]

CPU-only reference on Xiaomi 14.

here is a screenshot to demonstrate this issue(the input question is: where is China's capital city?):

same issue here with following models on Xiaomi 14 (Xiaomi 14 is available since Oct 2023, Xiaomi 14 contains a very very very powerful mobile SoC ------ Qualcomm SM8650-AB Snapdragon 8 Gen 3 (4 nm) ------ and was used for personal device-side AI PoC development activity).

• qwen1_5-1_8b-chat-q4_0.gguf (https://huggingface.co/Qwen/Qwen1.5-1.8B-Chat-GGUF/resolve/main/qwen1_5-1_8b-chat-q4_0.gguf , 1.1 GB)
• llama-2-7b.Q4_K_M.gguf (https://huggingface.co/TheBloke/Llama-2-7B-Chat-GGUF/resolve/main/llama-2-7b-chat.Q4_K_M.gguf, 4.08 GB)
• blossom-v3-baichuan2-7b.Q4_K_M.gguf (https://huggingface.co/TheBloke/blossom-v3-baichuan2-7B-GGUF/resolve/main/blossom-v3-baichuan2-7b.Q4_K_M.gguf, 4.61 GB)

---

original similar issue report could be found at upstream llama.cpp:

• Baichuan2-7B-Chat model converted to ggml-model-q4_0.gguf, AI answer does not stop automatically when inference is made ggerganov/llama.cpp#5034

• Infinite loop of "context shift" ggerganov/llama.cpp#3969

[zhouwg]

a workaround method has been used for fix this issue.so close it accordingly.