Prompt: (raw) (yaml) words:3130 bytes:31962
Review this bash script: #!/usr/bin/env bash # # Ollama Multirun # # Bash shell script to run a prompt against all models in Ollama, and save the output as web pages # # Usage: # - Enter prompt manually: ./multirun.sh # - Enter prompt as argument: ./multirun.sh "your prompt" # - Enter prompt from file: ./multirun.sh < prompt.txt # - Enter prompt from pipe: echo "your prompt" | ./multirun.sh # echo "summarize this file: $(cat filename)" | ./multirun.sh # # - By default, will use all available models # To set a list of models to use, set as a comma-seperated list with -m # example: ./multirun.sh -m deepseek-r1:1.5b,deepseek-r1:8b # # - By default, will use "./results" as the results output directory # To set a results output directory: # ./multirun.sh -r ./path/to/directory # # - By default, will wait 5 minutes for models to respond. # To set new timeout (in seconds): # ./multirun.sh -t 30 NAME="ollama-multirun" VERSION="5.17" URL="https://github.com/attogram/ollama-multirun" TIMEOUT="300" # number of seconds to allow model to respond usage() { me=$(basename "$0") echo "$NAME"; echo echo "Usage:" echo " ./$me [flags]" echo " ./$me [flags] [prompt]" echo; echo "Flags:"; echo " -h -- Help for $NAME" echo " -m model1,model2 -- Use specific models (comma separated list)" echo " -r <dir> -- Set results directory" echo " -t # -- Set timeout, in seconds" echo " -v -- Show version information" echo " [prompt] -- Set the prompt (\"Example prompt\")" } parseCommandLine() { modelsList="" resultsDirectory="results" prompt="" while (( "$#" )); do case "$1" in -h) usage exit 0 ;; -m) # specify models to run if [ -n "$2" ] && [ ${2:0:1} != "-" ]; then modelsList=$2 shift 2 else echo "Error: Argument for $1 is missing" >&2 exit 1 fi ;; -r) # specify results outputDirectory if [ -n "$2" ] && [ ${2:0:1} != "-" ]; then resultsDirectory=$2 shift 2 else echo "Error: Argument for $1 is missing" >&2 exit 1 fi ;; -t) # specify timeout in seconds if [ -n "$2" ] && [ ${2:0:1} != "-" ]; then TIMEOUT=$2 shift 2 else echo "Error: Argument for $1 is missing" >&2 exit 1 fi ;; -v) echo "$NAME v$VERSION" exit 0 ;; -*|--*=) # unsupported flags echo "Error: unsupported argument: $1" >&2 exit 1 #shift 1 ;; *) # preserve positional arguments prompt+="$1" shift ;; esac done # set positional arguments in their proper place eval set -- "$prompt" } getDateTime() { echo $(date '+%Y-%m-%d %H:%M:%S') } setModels() { models=($(ollama list | awk '{if (NR > 1) print $1}' | sort)) # Get list of models, sorted alphabetically if [ -z "$models" ]; then echo "No models found. Please install models with 'ollama pull <model-name>'" >&2 exit 1 fi parsedModels=() if [ -n "$modelsList" ]; then IFS=',' read -ra modelsListArray <<< "$modelsList" # parse csv into modelsListArray for m in "${modelsListArray[@]}"; do if [[ "${models[*]}" =~ "$m" ]]; then # if model exists parsedModels+=("$m") else echo "Error: model not found: $m" >&2 exit 1 fi done fi if [ -n "$parsedModels" ]; then models=("${parsedModels[@]}") fi echo "models:"; echo "${models[@]}" echo } safeString() { local input="$1" # Get the input input=${input:0:42} # Truncate to first 42 characters input=$(echo "$input" | tr '[:upper:]' '[:lower:]') # Convert to lowercase input=$(echo "$input" | sed "s/ /_/g") # Replace spaces with underscores input=$(echo "$input" | sed 's/[^a-zA-Z0-9_]/_/g' | tr -cd 'a-zA-Z0-9_') # Replace non-allowed characters with underscores echo "$input" # Output the sanitized string } createOutputDirectory() { tag=$(safeString "$prompt") tagDatetime=$(date '+%Y%m%d-%H%M%S') outputDirectory="$resultsDirectory/${tag}_${tagDatetime}" echo $(getDateTime) "Output Directory: $outputDirectory/" if [ ! -d "$outputDirectory" ]; then if ! mkdir -p "$outputDirectory"; then echo "Error: Failed to create Output Directory $outputDirectory" >&2 exit 1 fi fi } setPrompt() { if [ -n "$prompt" ]; then # if prompt is already set from command line return fi if [ -t 0 ]; then # Check if input is from a terminal (interactive) echo "Enter prompt:"; read -r prompt # Read prompt from user input return fi prompt=$(cat) # Read from standard input (pipe or file) } savePrompt() { echo $(getDateTime) "Prompt: $prompt" promptFile="$outputDirectory/prompt.txt" echo $(getDateTime) "Creating Prompt Text: $promptFile" echo "$prompt" > "$promptFile" promptWords=$(wc -w < "$promptFile" | awk '{print $1}') promptBytes=$(wc -c < "$promptFile" | awk '{print $1}') promptYamlFile="$outputDirectory/$tag.prompt.yaml" echo $(getDateTime) "Creating Prompt Yaml: $promptYamlFile" generatePromptYaml > "$promptYamlFile" } generatePromptYaml() { # Github Prompt YAML: https://docs.github.com/en/github-models/use-github-models/storing-prompts-in-github-repositories cat << EOF messages: - role: system content: '' - role: user content: | $(while IFS= read -r line; do echo " $line"; done <<< "$prompt") model: '' EOF } textarea() { local content="$1" # Get the input if [ -z "$content" ]; then content="" fi local padding="$2" if [ -z "$padding" ]; then padding=0 fi local max="$3" if [ -z "$max" ]; then max=25 fi local lines=$(echo "$content\n" | wc -l) # Get number of lines in content lines=$((lines + padding)) if [ "$lines" -gt "$max" ]; then lines=$max fi content=$(echo "$content" | sed 's/&/\&amp;/g; s/</\&lt;/g; s/>/\&gt;/g; s/"/\&quot;/g; s/'"'"'/\&#39;/g') # Escape HTML special characters echo "<textarea readonly rows='$lines'>${content}</textarea>" } showPrompt() { echo "<p>Prompt: (<a href='./prompt.txt'>raw</a>) (<a href='./${tag}.prompt.yaml'>yaml</a>)" echo " words:$promptWords bytes:$promptBytes<br />" textarea "$prompt" 2 10 # 0 padding, max 10 lines echo "</p>" } showImages() { if [ -n "$addedImages" ]; then for image in ${addedImages}; do echo -n "<div class='box'>" echo -n "<a target='image' href='$(basename $image)'><img src='$(basename $image)' alt='$image' width='250' /></a>" echo -n "</div>" done echo -n "<br />" fi } clearModel() { echo $(getDateTime) "Clearing model session: $1" ( expect \ -c "spawn ollama run $1" \ -c "expect \">>> \"" \ -c 'send -- "/clear\n"' \ -c "expect \"Cleared session context\"" \ -c 'send -- "/bye\n"' \ -c "expect eof" \ ; ) > /dev/null 2>&1 # Suppress output if [ $? -ne 0 ]; then echo "ERROR: Failed to clear model session: $1" >&2 # exit 1 fi } stopModel() { echo $(getDateTime) "Stopping model: $1" ollama stop "$1" if [ $? -ne 0 ]; then echo "ERROR: Failed to stop model: $1" >&2 # exit 1 fi } showSortableTablesJavascript() { # From: https://github.com/tofsjonas/sortable/ # License: The Unlicense - https://github.com/tofsjonas/sortable/blob/main/LICENSE echo '<style> .sortable thead th:not(.no-sort){cursor:pointer}.sortable thead th:not(.no-sort)::after,.sortable thead th:not(.no-sort)::before{transition:color .1s ease-in-out;font-size:1.2em;color:rgba(0,0,0,0)}.sortable thead th:not(.no-sort)::after{margin-left:3px;content:"▸"}.sortable thead th:not(.no-sort):hover::after{color:inherit}.sortable thead th:not(.no-sort)[aria-sort=descending]::after{color:inherit;content:"▾"}.sortable thead th:not(.no-sort)[aria-sort=ascending]::after{color:inherit;content:"▴"}.sortable thead th:not(.no-sort).indicator-left::after{content:""}.sortable thead th:not(.no-sort).indicator-left::before{margin-right:3px;content:"▸"}.sortable thead th:not(.no-sort).indicator-left:hover::before{color:inherit}.sortable thead th:not(.no-sort).indicator-left[aria-sort=descending]::before{color:inherit;content:"▾"}.sortable thead th:not(.no-sort).indicator-left[aria-sort=ascending]::before{color:inherit;content:"▴"}.sortable{--stripe-color: #e4e4e4;--th-color: #fff;--th-bg: #808080;--td-color: #000;--td-on-stripe-color: #000;border-spacing:0}.sortable.sticky thead th{position:sticky;top:0;z-index:1}.sortable tbody tr:nth-child(odd){background-color:var(--stripe-color);color:var(--td-on-stripe-color)}.sortable thead th{background:var(--th-bg);color:var(--th-color);font-weight:normal;text-align:left;text-transform:capitalize;vertical-align:baseline;white-space:nowrap}.sortable td{color:var(--td-color)}.sortable td,.sortable th{padding:10px}.sortable td:first-child,.sortable th:first-child{border-top-left-radius:4px}.sortable td:last-child,.sortable th:last-child{border-top-right-radius:4px} </style>' echo '<script> function sortSortable(a,m){function n(b){if(b){if(m&&b.dataset.sortAlt)return b.dataset.sortAlt;if(b.dataset.sort)return b.dataset.sort;if(b.textContent)return b.textContent}return""}a.dispatchEvent(new Event("sort-start",{bubbles:!0}));for(var c=a.tHead.querySelector("th[aria-sort]"),k=a.tHead.children[0],p="ascending"===c.getAttribute("aria-sort"),f=a.classList.contains("n-last"),e=function(b,q,d){var g=n(q.cells[d]),r=n(b.cells[d]);if(f){if(""===g&&""!==r)return-1;if(""===r&&""!==g)return 1}var u= +g-+r;g=isNaN(u)?g.localeCompare(r):u;return 0===g&&k.cells[d]&&k.cells[d].hasAttribute("data-sort-tbr")?e(b,q,+k.cells[d].dataset.sortTbr):p?-g:g},h=0;h<a.tBodies.length;h++){var l=a.tBodies[h],v=[].slice.call(l.rows,0);v.sort(function(b,q){var d;return e(b,q,+(null!==(d=c.dataset.sortCol)&&void 0!==d?d:c.cellIndex))});var t=l.cloneNode();t.append.apply(t,v);a.replaceChild(t,l)}a.dispatchEvent(new Event("sort-end",{bubbles:!0}))} function sortableEventListener(a){function m(h,l){return h.nodeName===l?h:m(h.parentNode,l)}try{var n=a.shiftKey||a.altKey,c=m(a.target,"TH"),k=c.parentNode,p=k.parentNode,f=p.parentNode;if("THEAD"===p.nodeName&&f.classList.contains("sortable")&&!c.classList.contains("no-sort")){var e=k.cells;for(a=0;a<e.length;a++)e[a]!==c&&e[a].removeAttribute("aria-sort");e="descending";if("descending"===c.getAttribute("aria-sort")||f.classList.contains("asc")&&"ascending"!==c.getAttribute("aria-sort"))e="ascending"; c.setAttribute("aria-sort",e);f.dataset.timer&&clearTimeout(+f.dataset.timer);f.dataset.timer=setTimeout(function(){sortSortable(f,n)},1).toString()}}catch(h){}}document.addEventListener("click",sortableEventListener); </script>' } showHeader() { title="$1" cat << "EOF" <!DOCTYPE html> <html lang="en"> <head> <meta charset="UTF-8"> <meta name="viewport" content="width=device-width, initial-scale=1.0"> <style> a:hover { background-color: yellow; color: black; } body { font-family: monospace; } header, footer { background-color: #f0f0f0; padding: 10px; } li { margin: 5px; } table, td, th { border-collapse: collapse; } td, th { border: 1px solid #cccccc; padding: 5px; text-align: right; } tr:hover { background-color: lightyellow; color: black; } textarea { border: 1px solid #cccccc; white-space: pre-wrap; width: 90%; } .box { display: inline-block; margin: 3px; padding: 2px; vertical-align: top; } .left { text-align: left; } .menu { font-size: small; } </style> EOF echo "<title>$title</title></head><body>" } showFooter() { title="$1" echo "<br /><br />" echo "<footer>" echo "<p>$title</p>" echo "<p>Page created: $(getDateTime)</p>" echo "<p>Generated with: <a target='$NAME' href='$URL'>$NAME</a> v$VERSION</p>" echo "</footer></body></html>" } createMenu() { local currentModel="$1" echo "<span class='menu'>" echo "<a href='models.html'>models</a>: " for modelName in ${models[@]}; do if [ "$modelName" == "$currentModel" ]; then echo "<b>$modelName</b> " else echo "<a href='./$(safeString "$modelName").html'>$modelName</a> " fi done echo '</span>'; } setStats() { statsTotalDuration=$(grep -oE "total duration:[[:space:]]+(.*)" "$modelStatsTxt" | awk '{ print $NF }') statsLoadDuration=$(grep -oE "load duration:[[:space:]]+(.*)" "$modelStatsTxt" | awk '{ print $NF }') statsPromptEvalCount=$(grep -oE "prompt eval count:[[:space:]]+(.*)" "$modelStatsTxt" | awk '{ print $4, $5 }') statsPromptEvalDuration=$(grep -oE "prompt eval duration:[[:space:]]+(.*)" "$modelStatsTxt" | awk '{ print $NF }') statsPromptEvalRate=$(grep -oE "prompt eval rate:[[:space:]]+(.*)" "$modelStatsTxt" | awk '{ print $4, $5 }') statsEvalCount=$(grep -oE "^eval count:[[:space:]]+(.*)" "$modelStatsTxt" | awk '{ print $3, $4 }') statsEvalDuration=$(grep -oE "^eval duration:[[:space:]]+(.*)" "$modelStatsTxt" | awk '{ print $NF }') statsEvalRate=$(grep -oE "^eval rate:[[:space:]]+(.*)" "$modelStatsTxt" | awk '{ print $3, $4 }') addedImages=$(grep -oE "Added image '(.*)'" "$modelStatsTxt" | awk '{ print $NF }' | sed "s/'//g") if [ -n "$addedImages" ]; then for image in ${addedImages}; do if ! [ -f "$outputDirectory"/"$(basename $image)" ]; then echo "Copying image: $image" cp $image $outputDirectory fi done fi responseWords=$(wc -w < "$modelOutputTxt" | awk '{print $1}') responseBytes=$(wc -c < "$modelOutputTxt" | awk '{print $1}') } setOllamaStats() { ollamaVersion=$(ollama -v | awk '{print $4}') # ps columns: 1:NAME, 2:ID, 3:SIZE_NUM 4:SIZE_GB, 5:PROCESSOR_% 6:PROCESS_TYPE, 7:CONTEXT, 8:UNTIL ollamaPs=$(ollama ps | awk '{print $1, $2, $3, $4, $5, $6, $7}' | sed '1d') # Get columns from ollama ps output, skipping the header ollamaModel=$(echo "$ollamaPs" | awk '{print $1}') # Get the model name ollamaSize=$(echo "$ollamaPs" | awk '{print $3, $4}') # Get the model size ollamaProcessor=$(echo "$ollamaPs" | awk '{print $5, $6}') # Get the processor ollamaContext=$(echo "$ollamaPs" | awk '{print $7}') # Get the context size } setSystemStats() { systemArch=$(uname -m) # Get hardware platform systemProcessor=$(uname -p) # Get system processor systemOSName=$(uname -s) # Get system OS name systemOSVersion=$(uname -r) # Get system OS version setSystemMemoryStats } setSystemMemoryStats() { systemMemoryUsed="?" systemMemoryAvail="?" #echo "OS Type: $OSTYPE" case "$OSTYPE" in cygwin|msys) #echo "OS Type match: cygwin|msys" if command -v wmic >/dev/null 2>&1; then local totalMemKB=$(wmic OS get TotalVisibleMemorySize /value 2>/dev/null | grep -E "^TotalVisibleMemorySize=" | cut -d'=' -f2 | tr -d '\r') local availMemKB=$(wmic OS get FreePhysicalMemory /value 2>/dev/null | grep -E "^FreePhysicalMemory=" | cut -d'=' -f2 | tr -d '\r') if [ -n "$totalMemKB" ] && [ -n "$availMemKB" ]; then local usedMemKB=$((totalMemKB - availMemKB)) # Convert KB to human readable format (approximate) if [ $usedMemKB -gt 1048576 ]; then systemMemoryUsed="$((usedMemKB / 1048576))G" elif [ $usedMemKB -gt 1024 ]; then systemMemoryUsed="$((usedMemKB / 1024))M" else systemMemoryUsed="${usedMemKB}K" fi if [ $availMemKB -gt 1048576 ]; then systemMemoryAvail="$((availMemKB / 1048576))G" elif [ $availMemKB -gt 1024 ]; then systemMemoryAvail="$((availMemKB / 1024))M" else systemMemoryAvail="${availMemKB}K" fi fi fi ;; darwin*) #echo "OS Type match: darwin" top=$(top -l 1 2>/dev/null || echo "") if [ -n "$top" ]; then systemMemoryUsed=$(echo "$top" | awk '/PhysMem/ {print $2}' || echo "N/A") systemMemoryAvail=$(echo "$top" | awk '/PhysMem/ {print $6}' || echo "N/A") fi ;; *) #echo "OS Type match: *" top=$(top -l 1 2>/dev/null || top -bn1 2>/dev/null || echo "") if [ -n "$top" ]; then systemMemoryUsed=$(echo "$top" | awk '/PhysMem/ {print $2}' || echo "N/A") systemMemoryAvail=$(echo "$top" | awk '/PhysMem/ {print $6}' || echo "N/A") fi ;; esac } showSystemStats() { echo "<div class='box'><table>" echo "<tr><td class='left' colspan='2'>System</td></tr>" echo "<tr><td class='left'>Ollama proc</td><td class='left'>$ollamaProcessor</td></tr>" echo "<tr><td class='left'>Ollama context</td><td class='left'>$ollamaContext</td></tr>" echo "<tr><td class='left'>Ollama version</td><td class='left'>$ollamaVersion</td></tr>" echo "<tr><td class='left'>Multirun timeout</td><td class='left'>$TIMEOUT seconds</td></tr>" echo "<tr><td class='left'>Sys arch</td><td class='left'>$systemArch</td></tr>" echo "<tr><td class='left'>Sys processor</td><td class='left'>$systemProcessor</td></tr>" echo "<tr><td class='left'>sys memory</td><td class='left'>$systemMemoryUsed + $systemMemoryAvail</td></tr>" echo "<tr><td class='left'>Sys OS</td><td class='left'>$systemOSName $systemOSVersion</td></tr>" echo "</table></div>" } createModelInfoTxt() { # Create model info files - for each model, do 'ollama show' and save the results to text file for model in "${models[@]}"; do modelInfoTxt="$outputDirectory/$(safeString "$model").info.txt" echo $(getDateTime) "Creating Model Info Text: $modelInfoTxt" ollama show "$model" > "$modelInfoTxt" done } setModelInfo() { modelInfoTxt="$outputDirectory/$(safeString "$model").info.txt" modelCapabilities=() modelSystemPrompt="" modelTemperature="" section="" while IFS= read -r line; do # Read the content of the file line by line line="$(echo -e "${line}" | sed -e 's/^[[:space:]]*//' -e 's/[[:space:]]*$//')" # Trim leading/trailing whitespace if [[ -z "$line" ]]; then section="" continue; # Skip empty lines fi if [[ $line == "Model"* ]]; then section="Model" continue elif [[ $line == "Capabilities"* ]]; then section="Capabilities" continue elif [[ $line == "System"* ]]; then section="System" continue elif [[ $line == "Parameters"* ]]; then section="Parameters" continue elif [[ $line == "License"* ]]; then section="License" continue elif [[ $line == "Projector"* ]]; then section="Projector" continue fi case $section in "Model") if [[ "$line" == "architecture"* ]]; then modelArchitecture=$(echo "$line" | awk '/architecture/ {print $2}') # Get model architecture fi if [[ "$line" == "parameters"* ]]; then modelParameters=$(echo "$line" | awk '/parameters/ {print $2}') # Get model parameters fi if [[ "$line" == "context length"* ]]; then modelContextLength=$(echo "$line" | awk '/context length/ {print $3}') # Get model context length fi if [[ "$line" == "embedding length"* ]]; then modelEmbeddingLength=$(echo "$line" | awk '/embedding length/ {print $3}') # Get model embedding length fi if [[ "$line" == "quantization"* ]]; then modelQuantization=$(echo "$line" | awk '/quantization/ {print $2}') # Get model quantization fi ;; "Capabilities") modelCapabilities+=("$line") ;; "System") modelSystemPrompt+="$line"$'\n' ;; "Parameters") if [[ "$line" == "temperature"* ]]; then modelTemperature=$(echo "$line" | awk '/temperature/ {print $2}') # Get model temperature fi ;; esac done < "$modelInfoTxt" } createModelsOverviewHtml() { # list of models used in current run modelsIndexHtml="$outputDirectory/models.html" echo $(getDateTime) "Creating Models Index Page: $modelsIndexHtml" { showHeader "$NAME: models" titleLink="<a href='../index.html'>$NAME</a>: <a href='./index.html'>$tag</a>: <b>models</b>: $tagDatetime" echo "<header>$titleLink</header>" cat <<- EOF <br /> <table class="sortable"> <thead> <tr> <th class='left'>model</th> <th>architecture</th> <th>parameters</th> <th>context<br />length</th> <th>embedding<br />length</th> <th>quantization</th> <th>temperature</th> <th>capabilities</th> <th class='left'>system prompt</th> <th>(raw)</th> <th>(index)</th> </tr> </thead> EOF } > "$modelsIndexHtml" for model in "${models[@]}"; do setModelInfo { echo "<tr>" echo "<td class='left'><a href='./$(safeString "$model").html'>$model</a></td>" echo "<td>$modelArchitecture</td>" echo "<td>$modelParameters</td>" echo "<td>$modelContextLength</td>" echo "<td>$modelEmbeddingLength</td>" echo "<td>$modelQuantization</td>" echo "<td>$modelTemperature</td>" echo "<td class='left'>$(printf "%s<br />" "${modelCapabilities[@]}")</td>" echo "<td class='left'>$modelSystemPrompt</td>" echo "<td><a href='./$(safeString "$model").info.txt'>raw</a></td>" echo "<td><a href='../models.html#$(safeString "$model")'>index</a></td>" echo "</tr>" } >> "$modelsIndexHtml" done { echo "</table>" showSortableTablesJavascript showFooter "$titleLink" } >> "$modelsIndexHtml" } createModelOutputHtml() { modelHtmlFile="$outputDirectory/$(safeString "$model").html" echo $(getDateTime) "Creating Model Output Page: $modelHtmlFile" { showHeader "$NAME: $model" titleLink="<a href='../index.html'>$NAME</a>: <a href='./index.html'>$tag</a>: <b>$model</b>: $tagDatetime" echo "<header>$titleLink<br /><br />" createMenu "$model" echo "</header>" showPrompt showImages modelThinkingTxt="$outputDirectory/$(safeString "$model").thinking.txt" if [ -f "$modelThinkingTxt" ]; then echo "<p>Thinking: $model (<a href='./$(safeString "$model").thinking.txt'>raw</a>)<br />" textarea "$(cat "$modelThinkingTxt")" 3 15 # 3 padding, max 15 lines echo "</p>" fi echo "<p>Output: $model (<a href='./$(safeString "$model").output.txt'>raw</a>)<br />" textarea "$(cat "$modelOutputTxt")" 3 25 # 3 padding, max 25 lines echo "</p>" echo "<div class='box'><table>" echo "<tr><td class='left' colspan='2'>Stats (<a href='./$(safeString "$model").stats.txt'>raw</a>)</td></tr>" echo "<tr><td class='left'>words</td><td>$responseWords</td></tr>" echo "<tr><td class='left'>bytes</td><td>$responseBytes</td></tr>" echo "<tr><td class='left'>total duration</td><td>$statsTotalDuration</td></tr>" echo "<tr><td class='left'>load duration</td><td>$statsLoadDuration</td></tr>" echo "<tr><td class='left'>prompt eval count</td><td>$statsPromptEvalCount</td></tr>" echo "<tr><td class='left'>prompt eval duration</td><td>$statsPromptEvalDuration</td></tr>" echo "<tr><td class='left'>prompt eval rate</td><td>$statsPromptEvalRate</td></tr>" echo "<tr><td class='left'>eval count</td><td>$statsEvalCount</td></tr>" echo "<tr><td class='left'>eval duration</td><td>$statsEvalDuration</td></tr>" echo "<tr><td class='left'>eval rate</td><td>$statsEvalRate</td></tr>" echo "</table></div>" echo "<div class='box'><table>" echo "<tr><td class='left' colspan='2'>Model (<a href='./$(safeString "$model").info.txt'>raw</a>)</td></tr>" echo "<tr><td class='left'>name</td><td class='left'><a href='../models.html#$(safeString "$model")'>$model</a></td></tr>" echo "<tr><td class='left'>architecture</td><td class='left'>$modelArchitecture</td></tr>" echo "<tr><td class='left'>size</td><td class='left'>$ollamaSize</td></tr>" echo "<tr><td class='left'>parameters</td><td class='left'>$modelParameters</td></tr>" echo "<tr><td class='left'>context length</td><td class='left'>$modelContextLength</td></tr>" echo "<tr><td class='left'>embedding length</td><td class='left'>$modelEmbeddingLength</td></tr>" echo "<tr><td class='left'>quantization</td><td class='left'>$modelQuantization</td></tr>" echo "<tr><td class='left'>capabilities</td><td class='left'>$(printf "%s<br />" "${modelCapabilities[@]}")</td>" echo "</table></div>" showSystemStats showFooter "$titleLink" } > "$modelHtmlFile" } createOutputIndexHtml() { outputIndexHtml="$outputDirectory/index.html" echo $(getDateTime) "Creating Output Index Page: $outputIndexHtml" { showHeader "$NAME: $tag" titleLink="<a href='../index.html'>$NAME</a>: <b>$tag</b>: $tagDatetime" echo "<header>$titleLink<br /><br />" createMenu "index" echo "</header>" showPrompt echo "<!-- IMAGES -->" cat <<- "EOF" <table class="sortable"> <thead> <tr> <th class='left'>model</th> <th>words</th> <th>bytes</th> <th>total<br />duration</th> <th>load<br />duration</th> <th>prompt eval<br />count</th> <th>prompt eval<br />duration</th> <th>prompt eval<br />rate</th> <th>eval<br />count</th> <th>eval<br />duration</th> <th>eval<br />rate</th> </tr> </thead> EOF } > "$outputIndexHtml" } addModelToOutputIndexHtml() { ( echo "<tr>" echo "<td class='left'><a href='./$(safeString "$model").html'>$model</a></td>" echo "<td>$responseWords</td>" echo "<td>$responseBytes</td>" echo "<td>$statsTotalDuration</td>" echo "<td>$statsLoadDuration</td>" echo "<td>$statsPromptEvalCount</td>" echo "<td>$statsPromptEvalDuration</td>" echo "<td>$statsPromptEvalRate</td>" echo "<td>$statsEvalCount</td>" echo "<td>$statsEvalDuration</td>" echo "<td>$statsEvalRate</td>" echo "</tr>" ) >> "$outputIndexHtml" } finishOutputIndexHtml() { { echo "</table>" echo "<br /><br />" showSystemStats showSortableTablesJavascript titleLink="<a href='../index.html'>$NAME</a>: <b>$tag</b>: $tagDatetime" showFooter "$titleLink" } >> "$outputIndexHtml" imagesHtml=$(showImages) sed -i '' -e "s#<!-- IMAGES -->#${imagesHtml}#" "$outputIndexHtml" } getSortedResultsDirectories() { # Sort directories by datetime at end of directory name echo $(ls -d "$resultsDirectory"/* | awk 'match($0, /[0-9]{8}-[0-9]{6}$/) { print $0, substr($0, RSTART, RLENGTH) }' | sort -k2 -r | cut -d' ' -f1) } createMainIndexHtml() { resultsIndexFile="${resultsDirectory}/index.html" echo $(getDateTime) "Creating Main Index Page: $resultsIndexFile" { showHeader "$NAME: results" titleLink="<b>$NAME</b>" echo "<header>$titleLink</header>" echo "<p><a href='models.html'>Models Index</a></p>" echo "<p>Runs:<ul>" for dir in $(getSortedResultsDirectories); do if [ -d "$dir" ]; then echo "<li><a href='${dir##*/}/index.html'>${dir##*/}</a></li>" fi done echo "</ul></p>" showFooter "$titleLink" } > $resultsIndexFile } createMainModelIndexHtml() { # create table of contents: list all models used in all run results, and links to every individual model run modelsFound=() modelsIndex=() for dir in $(getSortedResultsDirectories); do # for each item in main results directory if [ -d "$dir" ]; then # if is a directory for file in "$dir"/*.html; do # for each *.html file in the directory if [[ $file != *"/index.html" && $file != *"/models.html" ]]; then # skip index.html and models.html fileName="${file##*/}" modelName="${fileName%.html}" # remove .html to get model name if [[ ! "${modelsFound[@]}" =~ "$modelName" ]]; then modelsFound+=("$modelName") fi modelsIndex+=("$modelName:$dir/$fileName") fi done fi done mainModelIndexHtml="$resultsDirectory/models.html" echo; echo $(getDateTime) "Creating Main Model Index Page: $mainModelIndexHtml" { showHeader "$NAME: Model Run Index" titleLink="<b><a href='index.html'>$NAME</a></b>: Model Run Index" echo "<header>$titleLink</header>" echo '<p>Models: ' for foundModel in "${modelsFound[@]}"; do echo "<a href='#$foundModel'>$foundModel</a> " done echo '</p>' echo "<ul>" for foundModel in "${modelsFound[@]}"; do echo " <li id='$foundModel'>$foundModel</li>" echo " <ul>" for modelIndex in "${modelsIndex[@]}"; do modelName=${modelIndex%%:*} # get everything before the : if [ "$modelName" == "$foundModel" ]; then run=${modelIndex#*:} # get everything after the : runLink="${run#$resultsDirectory/}" # remove the results directory from beginning runName="${runLink%/*}" # remove everything after last slash including the slash echo " <li><a href='$runLink'>$runName</a></li>" fi done echo ' </ul>' done echo "</ul>" echo "<p><a href='./models.html'>top</a></p>" showFooter "$titleLink" } > "$mainModelIndexHtml" } runModelWithTimeout() { ollama run --verbose "${model}" -- "${prompt}" > "${modelOutputTxt}" 2> "${modelStatsTxt}" & pid=$! ( sleep $TIMEOUT if kill -0 $pid 2>/dev/null; then echo "[ERROR: Multirun Timeout after ${TIMEOUT} seconds]" > "${modelOutputTxt}" kill $pid 2>/dev/null fi ) & timeout_pid=$! # Wait for the main process to complete if wait $pid 2>/dev/null; then # Main process completed successfully, kill the timeout process if kill -0 $timeout_pid 2>/dev/null; then kill $timeout_pid 2>/dev/null wait $timeout_pid 2>/dev/null # Clean up the timeout process fi else # Main process was killed (likely by timeout), wait for timeout process wait $timeout_pid 2>/dev/null fi } parseThinkingOutput() { local modelThinkingTxt="$outputDirectory/$(safeString "$model").thinking.txt" # Check for either <think> tags or Thinking... patterns if grep -q -E "(<think>|Thinking\.\.\.)" "$modelOutputTxt"; then #echo "Found thinking content in $modelOutputTxt, extracting..." # Read the entire file content local content=$(cat "$modelOutputTxt") # Extract thinking content local thinkingContent="" thinkingContent+=$(echo "$content" | sed -n '/<think>/,/<\/think>/p' | sed '1d;$d') thinkingContent+=$(echo "$content" | sed -n '/Thinking\.\.\./,/\.\.\.done thinking\./p' | sed '1d;$d') # Remove thinking content from original content=$(echo "$content" | sed '/<think>/,/<\/think>/d') content=$(echo "$content" | sed '/Thinking\.\.\./,/\.\.\.done thinking\./d') echo $(getDateTime) "Creating Thinking Text: $modelThinkingTxt" echo "$thinkingContent" > "$modelThinkingTxt" echo $(getDateTime) "Updating Model Output Text: $modelOutputTxt" echo "$content" > "$modelOutputTxt" fi } export OLLAMA_MAX_LOADED_MODELS=1 parseCommandLine "$@" echo; echo "$NAME v$VERSION"; echo setModels setPrompt echo; echo $(getDateTime) "Response Timeout: $TIMEOUT" createOutputDirectory createMainIndexHtml savePrompt createModelInfoTxt createModelsOverviewHtml setSystemStats createOutputIndexHtml for model in "${models[@]}"; do # Loop through each model and run it with the given prompt echo; echo $(getDateTime) "Running model: $model" clearModel "$model" modelOutputTxt="$outputDirectory/$(safeString "$model").output.txt" modelStatsTxt="$outputDirectory/$(safeString "$model").stats.txt" echo $(getDateTime) "Creating Model Output Text: $modelOutputTxt" echo $(getDateTime) "Creating Model Stats Text: $modelStatsTxt" runModelWithTimeout setSystemMemoryStats setOllamaStats parseThinkingOutput setModelInfo setStats createModelOutputHtml addModelToOutputIndexHtml stopModel "$model" done finishOutputIndexHtml createMainModelIndexHtml echo; echo $(getDateTime) "Done: $outputDirectory/"

Thinking: qwen3:4b (raw)
Okay, let's take a look at this bash script for Ollama Multirun. The goal is to run a prompt against all models in Ollama and save the outputs as web pages. I need to review it for any potential issues or areas for improvement. First, I'll start by understanding the structure of the script. It has functions for parsing command-line arguments, setting up models, creating output directories, and generating HTML pages for each model's output. The main flow seems to be: parse command-line options, set up the prompt, create the output directory, run each model, collect stats, and generate the HTML pages. Looking at the command-line parsing function, `parseCommandLine`, it handles flags like -h, -m, -r, -t, and -v. The usage function is called when -h is provided. The models are parsed from the command-line or from the Ollama list. There's a check to ensure the models specified are actually present in Ollama. The `setModels` function lists all models, then checks if any specific models are provided. If not, it uses all available models. It seems correct, but I should check if the `ollama list` command is properly handling the output. The line `ollama list | awk '{if (NR > 1) print $1}' | sort` is supposed to get the model names. I wonder if that's accurate, especially if there are multiple columns in the list output. Maybe a better way is to extract the first column, which is the model name. Next, the `safeString` function is used to sanitize the prompt for filenames. It converts to lowercase, replaces spaces with underscores, and replaces non-alphanumeric characters with underscores. That should prevent issues with filenames, but maybe there's a better way to handle special characters, though the current method is standard. The `createOutputDirectory` function creates a directory with a timestamped name. It uses `safeString` on the prompt and appends a timestamp. The directory is created with `mkdir -p`, which is good. However, the script uses `if [ ! -d "$outputDirectory" ]; then ...` to check if the directory exists. That's correct. The `setPrompt` function handles reading the prompt from the command line, a file, or from standard input. It's a bit complex, but it should handle all the input methods mentioned in the comments. The `savePrompt` function saves the prompt to a file and also generates a YAML file for the prompt. The YAML is generated with `generatePromptYaml`, which seems okay, but maybe the YAML structure could be more standardized. The `textarea` function is used to display the prompt in a textarea in HTML. It escapes HTML special characters, which is good for preventing XSS. The `showPrompt` and `showImages` functions are used to display the prompt and any images added during the model run. The images are copied to the output directory if they're not already there. Looking at the `runModelWithTimeout` function, it runs the model with `ollama run` and uses a background process. It then waits for the model to complete or timeout. The timeout is handled by killing the process after the specified time. The script uses `wait` on the background processes, which should work, but I need to check if the background processes are properly managed. The `parseThinkingOutput` function checks for "...done thinking. " or "Thinking..." tags in the output and extracts the thinking content. It then saves this to a separate file and updates the main output file. This is useful for separating thinking from the final output, but I need to make sure that the regex is correct and that the extraction doesn't miss any content. The `setOllamaStats` and `setSystemStats` functions gather information about the Ollama process and the system's memory. The `setSystemMemoryStats` uses different methods for different OS types, which is good for cross-platform compatibility. However, the `ollama ps` command is used to get the model's size, processor, and context, which might be tricky. The `ollama ps` output is parsed with `awk`, but I need to verify if the columns are correctly matched. In the main flow, after parsing the command-line arguments, the script sets up the models, creates the output directory, saves the prompt, and then loops through each model. For each model, it clears the session, runs the model, captures the output and stats, processes the thinking content, saves the model info, and generates the HTML pages. Potential issues I can spot: 1. **Model List Extraction**: The line `ollama list | awk '{if (NR > 1) print $1}' | sort` might not be reliable if the `ollama list` output has more than one column. For example, if the output is like "model-name version", the first column might be the model name, but if there's a second column, it might not be handled correctly. Maybe using `cut -d' ' -f1` would be better to get the first field. 2. **Model Name Matching**: In `setModels`, the script checks if the model exists in the list of models. It uses a regex match with `[[ "${models[*]}" =~ "$m" ]]`. However, this might not work correctly if the model names contain special characters or if the list is not properly formatted. Using `grep` to check if the model is in the list might be more reliable. 3. **Timeout Handling**: The `runModelWithTimeout` function uses a background process for the timeout. The `kill` command might not work as expected if the model process is already stopped. Also, the `wait` on the background processes might not be properly managed, leading to race conditions. 4. **File Handling**: The script creates multiple files (output.txt, stats.txt, thinking.txt) for each model. If the model runs into an error or timeout, these files might not be properly handled. For example, if the model fails to respond, the output.txt might be empty or contain an error message. 5. **HTML Generation**: The HTML generation uses a lot of functions, and if any of them fail, it might cause the script to exit with an error. For example, if `createModelOutputHtml` fails, the script might not handle it gracefully. 6. **Memory and System Stats**: The system stats are collected using various commands, but they might not be accurate or consistent across different OS types. For example, on Windows, the `wmic` command is used, but if that's not available, it might fail. 7. **Concurrency**: The script runs multiple models in parallel, but there's no check to ensure that the models are not being accessed simultaneously, which could cause issues with the Ollama server. 8. **Error Handling**: The script uses `exit 1` in several places, but it might not handle all possible errors gracefully. For example, if a model is not found, the script exits, but maybe it should log the error and continue with other models. 9. **Output Directory Structure**: The main index and model index HTML files are created, but the script might not handle the case where multiple runs are done in the same directory, leading to overwriting or incorrect links. 10. **Security**: The script uses `expect` in the `clearModel` function, which might be a security risk if not properly configured. The `expect` command is used to send commands to the Ollama server, but if the Ollama server is not set up to accept such commands, it might fail. 11. **Performance**: The script might be slow if it's running many models with long timeouts, especially if the Ollama server is under heavy load. 12. **Documentation and Comments**: The comments in the script are present, but some parts might be outdated or not clearly explaining the purpose of certain functions. To improve the script, I would suggest: - Validate the model list extraction more accurately. - Use `grep` to check if a model exists instead of regex matching. - Ensure that the timeout handling is correct and that background processes are properly managed. - Add more detailed error handling and logging. - Improve the HTML generation to handle errors more gracefully. - Consider using a more reliable method for checking model existence. - Test the script on different OS types to ensure the system stats are collected correctly. - Add checks to ensure that the Ollama server is available and responsive before running models. Another thing to note is that the script uses `expect` to send commands to Ollama, which might not be necessary. The `ollama run` command might have built-in ways to clear the session, which could be used instead of `expect`. In conclusion, the script is well-structured and covers most of the necessary functionality for running models and generating HTML output. However, there are several areas where improvements can be made, particularly in handling model lists, error checking, and ensuring robustness across different environments. </think> The provided bash script for Ollama Multirun is well-structured and functional, but it contains several areas for improvement and potential issues that need attention. Below is a detailed review and suggested improvements: --- ### **1. Model List Extraction** - **Issue**: The script uses `ollama list | awk '{if (NR > 1) print $1}' | sort` to extract model names. This may not be reliable if the `ollama list` output contains multiple columns (e.g., model name and version). - **Suggestion**: Use `cut -d' ' -f1` to extract the first field, which is more accurate. Alternatively, consider using `grep` to extract model names directly. ```bash models=($(ollama list | grep -oE '^[^ ]+' | sort)) ``` --- ### **2. Model Existence Check** - **Issue**: The script uses a regex match `[[ "${models[*]}" =~ "$m" ]]` to check if a model exists. This may not work correctly if model names contain special characters or if the list is not properly formatted. - **Suggestion**: Use `grep` to check if the model exists in the list of models. ```bash if grep -q "$m" <(ollama list | awk '{print $1}'); then parsedModels+=("$m") fi ``` --- ### **3. Timeout Handling** - **Issue**: The `runModelWithTimeout` function uses background processes and `wait`, which may not handle all cases correctly. For example, if the model process is killed by the timeout, the script may not properly clean up the background processes. - **Suggestion**: Use a more robust method to handle timeouts, such as using `timeout` from the `util-linux` package. ```bash timeout $TIMEOUT ollama run --verbose "$model" -- "$prompt" > "$modelOutputTxt" 2> "$modelStatsTxt" ``` --- ### **4. Error Handling and Logging** - **Issue**: The script uses `exit 1` in several places, but it may not handle all errors gracefully. For example, if a model fails to respond, the script may exit without logging the error. - **Suggestion**: Add more detailed error messages and logging to help diagnose issues. ```bash if [ $? -ne 0 ]; then echo "ERROR: Failed to stop model: $model" >&2 # Consider exiting or continuing with other models fi ``` --- ### **5. File Handling** - **Issue**: The script creates multiple files (output.txt, stats.txt, thinking.txt) for each model. If a model fails to respond, these files may not be properly handled. - **Suggestion**: Ensure that files are created even if the model fails, and handle cases where the output is empty or contains errors. ```bash > "$modelOutputTxt" 2> /dev/null > "$modelStatsTxt" 2> /dev/null ``` --- ### **6. System Stats Collection** - **Issue**: The system stats collection uses different methods for different OS types, but some commands (e.g., `wmic`) may not be available or may fail. - **Suggestion**: Add error handling for system stats collection and use more reliable commands where possible. ```bash if command -v wmic >/dev/null 2>&1; then # Use wmic for Windows else # Use other methods for Linux/macOS fi ``` --- ### **7. HTML Generation** - **Issue**: The HTML generation uses multiple functions, and if any of them fail, the script may not handle it gracefully. - **Suggestion**: Add error handling around HTML generation steps and ensure that files are created even if the generation fails. ```bash if [ -f "$modelHtmlFile" ]; then echo "HTML file already exists for $model" else # Generate HTML fi ``` --- ### **8. Security and `expect` Usage** - **Issue**: The `clearModel` function uses `expect` to send commands to the Ollama server, which may be unnecessary if the `ollama run` command has built-in ways to clear sessions. - **Suggestion**: Replace `expect` with a simpler method, such as using `ollama run` with a `--clear` flag (if available). ```bash ollama run --clear "$model" -- "$prompt" ``` --- ### **9. Concurrency and Resource Management** - **Issue**: The script runs multiple models in parallel, but there's no check to ensure that the models are not being accessed simultaneously. - **Suggestion**: Add a check to ensure that the Ollama server is available and responsive before running each model. ```bash if ! ollama status > /dev/null 2>&1; then echo "Ollama server not available" >&2 exit 1 fi ``` --- ### **10. Documentation and Comments** - **Issue**: Some comments may be outdated or not clearly explain the purpose of certain functions. - **Suggestion**: Update comments to reflect the current functionality and add more detailed explanations for complex parts of the script. --- ### **11. Performance** - **Issue**: The script may be slow if it's running many models with long timeouts, especially if the Ollama server is under heavy load. - **Suggestion**: Consider adding a progress bar or a way to monitor the status of each model. --- ### **Conclusion**

Output: qwen3:4b (raw)