ollama 加速

ollama pull的时候，速度是一个先增加后下降的趋势，而ollama支持断点续传，我们可以每隔一段时间停掉重新拉取就行，下面是示例脚本，拉取千问32b

下面脚本保存为pull.sh

#!/bin/bash

# Check if model parameter is provided
if [ -z "$1" ]; then
    echo "Usage: $0 <model_name>"
    echo "Example: $0 qwen2.5:32b"
    exit 1
fi

MODEL="$1"

while true; do
    # Check if model already exists
    modelExists=$(ollama list | grep "$MODEL")
    
    if [[ -n "$modelExists" ]]; then
        echo "Model $MODEL is ready"
        break
    fi
    
    echo "Downloading $MODEL..."
    # Start model download
    ollama pull "$MODEL" &
    process_pid=$!
    
    # Wait for 60 seconds
    sleep 60
    
    # Try to stop the download process and restart
    if kill -0 $process_pid 2>/dev/null; then
        echo "Killing and restarting download of $MODEL..."
        kill -9 $process_pid
    else
        echo "Error with download process"
    fi
done

echo "Download of $MODEL completed successfully"

执行拉取：

chmod +x pull.sh 
./pull.sh qwq:32b
Downloading qwq:32b...
pulling manifest 
pulling c62ccde5630c...   3% ▕████                                                                                                                                                                    ▏ 557 MB/ 19 GB  6.4 MB/s  50m27sKilling and restarting download of qwq:32b...
Downloading qwq:32b...
pulling manifest