老大也不加个中文语言包，可累死我了,又弄了个启动脚本

[mychenjun]

我从昨天听说ollama和lmstudio是llama套壳后就下了这里的源码编译启动,没想到还有个webui页但全是英文的我又看不懂，（真心希望老大搞个中文语言）,先把源码下到我的windows笔记本上,用vscode打开,网页上复制了在里面查找,可累死我了,一夜没睡,汉化了重新编译，再启动看，里面各个按钮点开看,还有英文，又来,总算汉化了差不多了.













那个模型我就想要把ollama下载的模型利用起来,先用ollama list列出模型列表然后找那个实际文件,做软链接到llama.cpp/models下,但是好像谷歌的‌Gemini模型llama.cpp都不能加载不知道为什么,我ollama下了三四个用llama全不能加载.希望老大看下.
还有启动老是要打很长命令，叫deepseek帮我做了个脚本来启动.
chen@mm llama.cpp % cat runllama.sh
#!/bin/bash

配置路径

LLAMA_SERVER="$HOME/llama.cpp/bin/llama-server"
MODELS_DIR="$HOME/llama.cpp/models"
DEFAULT_HOST="0.0.0.0"
DEFAULT_PORT="8080"
DEFAULT_N_GPU_LAYERS="auto"

颜色定义

RED='\033[0;31m'
GREEN='\033[0;32m'
YELLOW='\033[1;33m'
BLUE='\033[0;34m'
NC='\033[0m' # No Color

获取模型列表函数（兼容 macOS）

get_models() {
local models=()
for f in "$MODELS_DIR"/*.gguf; do
if [[ -f "$f" ]]; then
basename "$f" .gguf
fi
done
printf '%s\n' "${models[@]}"
}

显示帮助

show_help() {
echo "用法: $0 [选项]"
echo ""
echo "选项:"
echo " -h, --help 显示帮助"
echo " -m, --model MODEL 直接指定模型名称（跳过选择）"
echo " --host HOST 服务器主机地址 (默认: $DEFAULT_HOST)"
echo " -p, --port PORT 服务器端口 (默认: $DEFAULT_PORT)"
echo ""
echo "示例:"
echo " $0 # 交互式选择"
echo " $0 -m qwen3.5_9b # 直接启动指定模型"
echo " $0 --host 127.0.0.1 --port 8080 # 只允许本地访问"
}

检查 llama-server 是否存在

if [[ ! -f "$LLAMA_SERVER" ]]; then
echo -e "${RED}错误: 找不到 llama-server 在 $LLAMA_SERVER${NC}"
exit 1
fi

检查模型目录

if [[ ! -d "$MODELS_DIR" ]]; then
echo -e "${RED}错误: 模型目录不存在 $MODELS_DIR${NC}"
exit 1
fi

获取模型列表（兼容 macOS 不使用 mapfile）

models=()
for f in "$MODELS_DIR"/*.gguf; do
if [[ -f "$f" ]]; then
models+=("$(basename "$f" .gguf)")
fi
done

if [[ ${#models[@]} -eq 0 ]]; then
echo -e "${RED}错误: 在 $MODELS_DIR 中没有找到模型文件${NC}"
echo "请先创建模型软链接"
exit 1
fi

解析命令行参数

HOST="$DEFAULT_HOST"
PORT="$DEFAULT_PORT"
MODEL_NAME=""

while [[ $# -gt 0 ]]; do
case $1 in
-h|--help)
show_help
exit 0
;;
-m|--model)
MODEL_NAME="$2"
shift 2
;;
--host)
HOST="$2"
shift 2
;;
-p|--port)
PORT="$2"
shift 2
;;
*)
echo -e "${RED}未知参数: $1${NC}"
show_help
exit 1
;;
esac
done

如果没有指定模型，显示选择菜单

if [[ -z "$MODEL_NAME" ]]; then
echo -e "${GREEN}========================================${NC}"
echo -e "${GREEN} llama-server 启动脚本${NC}"
echo -e "${GREEN}========================================${NC}"
echo ""

echo -e "${YELLOW}可用模型列表:${NC}"
echo "----------------------------------------"
for i in "${!models[@]}"; do
    model_file="$MODELS_DIR/${models[$i]}.gguf"
    model_size=$(du -h "$model_file" 2>/dev/null | cut -f1)
    echo -e "$((i+1)). ${models[$i]} ${BLUE}($model_size)${NC}"
done
echo "----------------------------------------"
echo ""

# 选择模型
while true; do
    read -p "请选择模型 (1-${#models[@]}): " choice
    if [[ "$choice" =~ ^[0-9]+$ ]] && [ "$choice" -ge 1 ] && [ "$choice" -le "${#models[@]}" ]; then
        MODEL_NAME="${models[$((choice-1))]}"
        break
    else
        echo -e "${RED}无效选择，请重新输入${NC}"
    fi
done

fi

验证模型文件

MODEL_FILE="$MODELS_DIR/${MODEL_NAME}.gguf"
if [[ ! -f "$MODEL_FILE" ]]; then
echo -e "${RED}错误: 找不到模型文件 $MODEL_FILE${NC}"
exit 1
fi

echo ""
echo -e "${GREEN}已选择模型: ${MODEL_NAME}${NC}"
echo ""

从模型文件获取原生上下文大小

echo -e "${YELLOW}正在读取模型信息...${NC}"
MODEL_CTX=""
if [[ -f "$LLAMA_SERVER" ]]; then
# 尝试从模型文件获取上下文大小
MODEL_CTX=$("$HOME/llama.cpp/build/bin/llama-gguf" -f "$MODEL_FILE" 2>/dev/null | grep "llama.context_length" | awk '{print $2}')
fi

默认参数

DEFAULT_TEMP="0.80"
DEFAULT_TOP_K="40"
DEFAULT_TOP_P="0.95"
DEFAULT_PRESENCE_PENALTY="0.00"
DEFAULT_FREQUENCY_PENALTY="0.00"
DEFAULT_CTX_SIZE="${MODEL_CTX:-4096}"
DEFAULT_N_PREDICT="512"
DEFAULT_BATCH_SIZE="512"
DEFAULT_UBATCH_SIZE="512"

echo -e "${GREEN}模型原生上下文: $DEFAULT_CTX_SIZE${NC}"
echo ""

显示默认值并询问修改

echo -e "${YELLOW}========================================${NC}"
echo -e "${YELLOW}服务器参数配置${NC}"
echo -e "${YELLOW}========================================${NC}"
echo -e "${BLUE}提示: 直接回车使用默认值${NC}"
echo ""

主机地址

read -p "主机地址 [默认: $HOST]: " input
HOST=${input:-$HOST}

端口

read -p "端口 [默认: $PORT]: " input
PORT=${input:-$PORT}

echo ""
echo -e "${YELLOW}========================================${NC}"
echo -e "${YELLOW}模型参数配置${NC}"
echo -e "${YELLOW}========================================${NC}"
echo ""

Temperature

read -p "Temperature [默认: $DEFAULT_TEMP]: " input
TEMP=${input:-$DEFAULT_TEMP}

Top K

read -p "Top K [默认: $DEFAULT_TOP_K]: " input
TOP_K=${input:-$DEFAULT_TOP_K}

Top P

read -p "Top P [默认: $DEFAULT_TOP_P]: " input
TOP_P=${input:-$DEFAULT_TOP_P}

Presence Penalty

read -p "Presence Penalty [默认: $DEFAULT_PRESENCE_PENALTY]: " input
PRESENCE_PENALTY=${input:-$DEFAULT_PRESENCE_PENALTY}

Frequency Penalty

read -p "Frequency Penalty [默认: $DEFAULT_FREQUENCY_PENALTY]: " input
FREQUENCY_PENALTY=${input:-$DEFAULT_FREQUENCY_PENALTY}

上下文大小

read -p "上下文大小 [默认: $DEFAULT_CTX_SIZE]: " input
CTX_SIZE=${input:-$DEFAULT_CTX_SIZE}

最大预测 token 数

read -p "最大预测 Token 数 [默认: $DEFAULT_N_PREDICT, -1=无限制]: " input
N_PREDICT=${input:-$DEFAULT_N_PREDICT}

Batch Size

read -p "Batch Size [默认: $DEFAULT_BATCH_SIZE]: " input
BATCH_SIZE=${input:-$DEFAULT_BATCH_SIZE}

GPU 层数

read -p "GPU 层数 [默认: $DEFAULT_N_GPU_LAYERS, auto=自动, all=全部, 数字=指定层数]: " input
N_GPU_LAYERS=${input:-$DEFAULT_N_GPU_LAYERS}

echo ""
echo -e "${GREEN}========================================${NC}"
echo -e "${GREEN}配置总结${NC}"
echo -e "${GREEN}========================================${NC}"
echo -e "模型: ${BLUE}$MODEL_NAME${NC}"
echo -e "主机: ${BLUE}http://$HOST:$PORT${NC}"
echo -e "Temperature: ${BLUE}$TEMP${NC}"
echo -e "Top K: ${BLUE}$TOP_K${NC}"
echo -e "Top P: ${BLUE}$TOP_P${NC}"
echo -e "Presence Penalty: ${BLUE}$PRESENCE_PENALTY${NC}"
echo -e "Frequency Penalty:${BLUE}$FREQUENCY_PENALTY${NC}"
echo -e "上下文大小: ${BLUE}$CTX_SIZE${NC}"
echo -e "最大 Token: ${BLUE}$N_PREDICT${NC}"
echo -e "Batch Size: ${BLUE}$BATCH_SIZE${NC}"
echo -e "GPU 层数: ${BLUE}$N_GPU_LAYERS${NC}"
echo ""

read -p "启动服务器? (y/n): " confirm
if [[ ! "$confirm" =~ ^[Yy]$ ]]; then
echo "已取消"
exit 0
fi

echo ""
echo -e "${GREEN}正在启动 llama-server...${NC}"
echo "================================================"

构建命令

CMD="$LLAMA_SERVER
-m "$MODEL_FILE"
--host $HOST
--port $PORT
--temp $TEMP
--top-k $TOP_K
--top-p $TOP_P
--presence-penalty $PRESENCE_PENALTY
--frequency-penalty $FREQUENCY_PENALTY
--ctx-size $CTX_SIZE
--n-predict $N_PREDICT
--batch-size $BATCH_SIZE"

添加 GPU 层数参数

if [[ "$N_GPU_LAYERS" != "auto" ]]; then
if [[ "$N_GPU_LAYERS" == "all" ]]; then
CMD="$CMD -ngl 999"
else
CMD="$CMD -ngl $N_GPU_LAYERS"
fi
fi

echo -e "${BLUE}执行命令:${NC}"
echo "$CMD"
echo "================================================"
echo ""

执行命令

eval "$CMD"
chen@mm llama.cpp %
最后希望llama.cpp越来越好.

[mychenjun]

我链接的ollama模型
chen@mm llama.cpp % cd models
chen@mm models % ls
all-minilm_latest.gguf deepseek-r1_8b.gguf llama3.2_3b.gguf qwen2.5-coder_14b.gguf
deepseek-coder_6.7b.gguf glm4_9b.gguf nomic-embed-text_latest.gguf qwen2.5_14b.gguf
deepseek-r1_14b.gguf llama3.1_8b.gguf phi4-reasoning_latest.gguf translategemma_latest.gguf
chen@mm models % ls -la
total 0
drwxr-xr-x 14 chen staff 448 May 2 13:26 .
drwxr-xr-x 5 chen staff 160 May 2 22:25 ..
lrwxr-xr-x 1 chen staff 104 May 2 11:19 all-minilm_latest.gguf -> /Users/chen/.ollama/models/blobs/sha256-797b70c4edf85907fe0a49eb85811256f65fa0f7bf52166b147fd16be2be4662
lrwxr-xr-x 1 chen staff 104 May 2 11:19 deepseek-coder_6.7b.gguf -> /Users/chen/.ollama/models/blobs/sha256-59bb50d8116b6a1f9bfbb940d6bb946a05554e591e30c8c2429ed6c854867ecb
lrwxr-xr-x 1 chen staff 104 May 2 11:19 deepseek-r1_14b.gguf -> /Users/chen/.ollama/models/blobs/sha256-6e9f90f02bb3b39b59e81916e8cfce9deb45aeaeb9a54a5be4414486b907dc1e
lrwxr-xr-x 1 chen staff 104 May 2 11:19 deepseek-r1_8b.gguf -> /Users/chen/.ollama/models/blobs/sha256-e6a7edc1a4d7d9b2de136a221a57336b76316cfe53a252aeba814496c5ae439d
lrwxr-xr-x 1 chen staff 104 May 2 11:19 glm4_9b.gguf -> /Users/chen/.ollama/models/blobs/sha256-b506a070d1152798d435ec4e7687336567ae653b3106f73b7b4ac7be1cbc4449
lrwxr-xr-x 1 chen staff 104 May 2 11:19 llama3.1_8b.gguf -> /Users/chen/.ollama/models/blobs/sha256-667b0c1932bc6ffc593ed1d03f895bf2dc8dc6df21db3042284a6f4416b06a29
lrwxr-xr-x 1 chen staff 104 May 2 11:19 llama3.2_3b.gguf -> /Users/chen/.ollama/models/blobs/sha256-dde5aa3fc5ffc17176b5e8bdc82f587b24b2678c6c66101bf7da77af9f7ccdff
lrwxr-xr-x 1 chen staff 104 May 2 11:19 nomic-embed-text_latest.gguf -> /Users/chen/.ollama/models/blobs/sha256-970aa74c0a90ef7482477cf803618e776e173c007bf957f635f1015bfcfef0e6
lrwxr-xr-x 1 chen staff 104 May 2 11:19 phi4-reasoning_latest.gguf -> /Users/chen/.ollama/models/blobs/sha256-89f38c5f78de6092c203b63dd62ff1285276a6f8ee140dd1eb2f6e196a631f40
lrwxr-xr-x 1 chen staff 104 May 2 11:19 qwen2.5-coder_14b.gguf -> /Users/chen/.ollama/models/blobs/sha256-ac9bc7a69dab38da1c790838955f1293420b55ab555ef6b4615efa1c1507b1ed
lrwxr-xr-x 1 chen staff 104 May 2 11:19 qwen2.5_14b.gguf -> /Users/chen/.ollama/models/blobs/sha256-2049f5674b1e92b4464e5729975c9689fcfbf0b0e4443ccf10b5339f370f9a54
lrwxr-xr-x 1 chen staff 104 May 2 11:19 translategemma_latest.gguf -> /Users/chen/.ollama/models/blobs/sha256-bdbf939b402e2f88fbe3e918beb777813009335756b4c17be7fe008dfe4815d4

[wjddythhk]

第一次用cpp，B站教程也比较少，可以请教一下大佬要怎么操作吗？

[mychenjun]

忘了说我的系统了.
Model Name: Mac mini
Model Identifier: Mac16,10
Chip: Apple M4
Total Number of Cores: 10 (4 Performance and 6 Efficiency)
Memory: 24 GB

[Pauliehedron]

Try for Mac <3 : https://github.com/ggml-org/LlamaBarn

How llama.cpp/llama-server: https://unsloth.ai/docs/zh/mo-xing/gemma-4#llama.cpp-zhi-nan