Add settings modal to web ui

jart · jart · commit bc82424cab81 · 2024-11-24T00:03:26.000-08:00
diff --git a/llamafile/server/flagz.cpp b/llamafile/server/flagz.cpp
@@ -33,6 +33,7 @@ Client::flagz()
     json["no_display_prompt"] = FLAG_no_display_prompt;
     json["nologo"] = FLAG_nologo;
     json["temperature"] = FLAG_temperature;
+    json["top_p"] = FLAG_top_p;
     json["presence_penalty"] = FLAG_presence_penalty;
     json["frequency_penalty"] = FLAG_frequency_penalty;
     if (FLAG_seed == LLAMA_DEFAULT_SEED) {
diff --git a/llamafile/server/www/chatbot.css b/llamafile/server/www/chatbot.css
@@ -263,6 +263,158 @@ ul li:first-child {
   display: block;
 }
 
+.settings-button {
+  padding: 0.75rem 1rem;
+  background: #6c757d;
+  color: white;
+  border: none;
+  border-radius: 6px;
+  cursor: pointer;
+  font-size: 1rem;
+  transition: background-color 0.2s;
+}
+
+.settings-button:hover {
+  background: #5c636a;
+}
+
+.settings-modal {
+  position: fixed;
+  top: 0;
+  left: 0;
+  right: 0;
+  bottom: 0;
+  background: rgba(0, 0, 0, 0.5);
+  z-index: 1000;
+  display: flex;
+  align-items: center;
+  justify-content: center;
+}
+
+.settings-panel {
+  background: white;
+  padding: 20px;
+  border-radius: 8px;
+  max-width: 500px;
+  width: 90%;
+}
+
+.settings-header {
+  display: flex;
+  justify-content: space-between;
+  align-items: center;
+  margin-bottom: 20px;
+}
+
+.settings-header button {
+  background: none;
+  border: none;
+  font-size: 24px;
+  cursor: pointer;
+}
+
+.setting-item {
+  margin-bottom: 20px;
+}
+
+.setting-item label {
+  display: block;
+  margin-bottom: 8px;
+  font-weight: bold;
+}
+
+.setting-item input {
+  width: 100%;
+  margin-bottom: 8px;
+}
+
+.setting-description {
+  font-size: 0.9em;
+  color: #666;
+  margin-top: 4px;
+}
+
+.setting-item.disabled label,
+.setting-item.disabled input,
+.setting-item.disabled .setting-description {
+  opacity: 0.5;
+  cursor: not-allowed;
+}
+
+.setting-item.disabled input {
+  pointer-events: none;
+}
+
+.setting-item.disabled {
+  pointer-events: none;
+  opacity: 0.8;
+}
+
+.setting-item.disabled input[type="range"] {
+  background: #ccc;
+}
+
+.setting-item.disabled label,
+.setting-item.disabled .setting-description {
+  color: #999;
+}
+
+/* Styling for penalty range inputs */
+.penalty-range {
+  -webkit-appearance: none;
+  appearance: none;
+  height: 6px;
+  border-radius: 8px;
+}
+
+/* Chrome track styling */
+.penalty-range::-webkit-slider-runnable-track {
+  width: 100%;
+  height: 6px;
+  border-radius: 8px;
+  background: linear-gradient(to right,
+    #ff000088 0%, 
+    #00000022 30%,
+    #00000022 70%,
+    #ff000088 100%
+  );
+}
+
+/* Firefox track styling */
+.penalty-range::-moz-range-track {
+  width: 100%;
+  height: 6px;
+  border-radius: 8px;
+  background: linear-gradient(to right,
+    #ff000088 0%, 
+    #00000022 30%,
+    #00000022 70%,
+    #ff000088 100%
+  );
+}
+
+/* Chrome thumb styling */
+.penalty-range::-webkit-slider-thumb {
+  -webkit-appearance: none;
+  appearance: none;
+  width: 16px;
+  height: 16px;
+  border-radius: 50%;
+  background: #666;
+  cursor: pointer;
+  margin-top: -5px; /* Centers the thumb on the track */
+}
+
+/* Firefox thumb styling */
+.penalty-range::-moz-range-thumb {
+  width: 16px;
+  height: 16px;
+  border: none;
+  border-radius: 50%;
+  background: #666;
+  cursor: pointer;
+}
+
 @media print {
 
   html,
diff --git a/llamafile/server/www/chatbot.js b/llamafile/server/www/chatbot.js
@@ -27,13 +27,17 @@ const DEFAULT_FLAGZ = {
   "frequency_penalty": 0,
   "presence_penalty": 0,
   "temperature": 0.8,
+  "top_p": 0.95,
   "seed": null
 };
 
 const chatMessages = document.getElementById("chat-messages");
 const chatInput = document.getElementById("chat-input");
 const sendButton = document.getElementById("send-button");
 const stopButton = document.getElementById("stop-button");
+const settingsButton = document.getElementById("settings-button");
+const settingsModal = document.getElementById("settings-modal");
+const closeSettings = document.getElementById("close-settings");
 
 let abortController = null;
 let disableAutoScroll = false;
@@ -168,6 +172,7 @@ async function sendMessage() {
   // update chat history
   chatHistory.push({ role: "user", content: message });
 
+  const settings = loadSettings();
   try {
     const response = await fetch(API_ENDPOINT, {
       method: "POST",
@@ -178,7 +183,10 @@ async function sendMessage() {
       body: JSON.stringify({
         model: "gpt-3.5-turbo",
         messages: chatHistory,
-        temperature: 0.0,
+        temperature: settings.temperature,
+        top_p: settings.top_p,
+        presence_penalty: settings.presence_penalty,
+        frequency_penalty: settings.frequency_penalty,
         stream: true
       }),
       signal: abortController.signal
@@ -361,9 +369,93 @@ function startChat(history) {
   scrollToBottom();
 }
 
+function loadSettings() {
+  const stored = localStorage.getItem('v1.modelSettings');
+  if (stored) {
+    return JSON.parse(stored);
+  }
+  return {
+    temperature: flagz.temperature,
+    top_p: flagz.top_p,
+    presence_penalty: flagz.presence_penalty,
+    frequency_penalty: flagz.frequency_penalty,
+  };
+}
+
+function saveSettings(settings) {
+  localStorage.setItem('v1.modelSettings', JSON.stringify(settings));
+}
+
+function formatDoubleWithPlus(x) {
+  return (x >= 0 ? "+" : "") + x.toFixed(2);
+}
+
+function updateSettingsDisplay(settings) {
+  document.getElementById("temp-value").textContent = settings.temperature ? settings.temperature.toFixed(2) : "0.00 (deterministic)";
+  document.getElementById("top-p-value").textContent = settings.top_p.toFixed(2);
+  document.getElementById("presence-value").textContent = formatDoubleWithPlus(settings.presence_penalty);
+  document.getElementById("frequency-value").textContent = formatDoubleWithPlus(settings.frequency_penalty);
+  document.getElementById("temperature").value = settings.temperature;
+  document.getElementById("top-p").value = settings.top_p;
+  document.getElementById("presence-penalty").value = settings.presence_penalty;
+  document.getElementById("frequency-penalty").value = settings.frequency_penalty;
+
+  // Handle top-p disabling - using a more reliable selector
+  const topPSettingItem = document.querySelector('.setting-item:has(#top-p)');
+  if (settings.temperature === 0) {
+    topPSettingItem.classList.add('disabled');
+  } else {
+    topPSettingItem.classList.remove('disabled');
+  }
+
+  // Update top-p description with percentage
+  const topPDescription = topPSettingItem.querySelector('.setting-description');
+  if (settings.top_p >= 1) {
+    topPDescription.textContent = "Disabled. All tokens will be considered by the sampler.";
+  } else if (settings.top_p > .5) {
+    const percentage = Math.round((1 - settings.top_p) * 100);
+    topPDescription.textContent = `The bottom ${percentage}% tokens will be ignored by the sampler.`;
+  } else {
+    const percentage = Math.round(settings.top_p * 100);
+    topPDescription.textContent = `Only the top ${percentage}% tokens will be considered by the sampler.`;
+  }
+}
+
+function setupSettings() {    
+  settingsButton.addEventListener("click", () => {
+    settingsModal.style.display = "flex";
+    updateSettingsDisplay(loadSettings());
+  });
+  closeSettings.addEventListener("click", () => {
+    settingsModal.style.display = "none";
+  });
+  ["temperature", "top-p", "presence-penalty", "frequency-penalty"].forEach(id => {
+    const element = document.getElementById(id);
+    element.addEventListener("input", (e) => {
+      const settings = loadSettings();
+      const value = parseFloat(e.target.value);
+      const key = id.replace(/-/g, '_');
+      settings[key] = value;
+      saveSettings(settings);
+      updateSettingsDisplay(settings);
+    });
+  });
+  settingsModal.addEventListener("mousedown", (e) => {
+    if (e.target === settingsModal) {
+      settingsModal.style.display = "none";
+    }
+  });
+  document.addEventListener("keydown", (e) => {
+    if (e.key === "Escape") {
+      settingsModal.style.display = "none";
+    }
+  });
+}
+
 async function chatbot() {
   flagz = await fetchFlagz();
   updateModelInfo();
+  setupSettings();
   startChat([{ role: "system", content: getSystemPrompt() }]);
   sendButton.addEventListener("click", sendMessage);
   stopButton.addEventListener("click", stopMessage);
diff --git a/llamafile/server/www/highlight.js b/llamafile/server/www/highlight.js
@@ -46,7 +46,7 @@ class HighlightDom extends Highlight {
   feed(s) {
     for (let i = 0; i < s.length; ++i) {
       this.text += s[i];
-      if (isspace(s[i])) {
+      if (isspace(s[i]) || this.text.length > 50) {
         this.flushText();
       }
     }
diff --git a/llamafile/server/www/index.html b/llamafile/server/www/index.html
@@ -22,6 +22,36 @@ <h1>
     <textarea class="chat-input" id="chat-input" placeholder="Type your message..." rows="1" autocomplete="off"></textarea>
     <button class="send-button" id="send-button">Send</button>
     <button class="stop-button" id="stop-button" style="display:none">Stop</button>
+    <button class="settings-button" id="settings-button" title="Settings">⚙️</button>
+  </div>
+</div>
+
+<div id="settings-modal" class="settings-modal" style="display:none">
+  <div class="settings-panel">
+    <div class="settings-header">
+      <h2>Model Settings</h2>
+      <button id="close-settings">×</button>
+    </div>
+    <div class="setting-item">
+      <label>temperature = <span id="temp-value">0.8</span></label>
+      <input type="range" id="temperature" min="0" max="2" step="0.01" value="0.8">
+      <p class="setting-description">How much randomness to use during sampling. Higher values improve improvisation, and lower values result in determinism.</p>
+    </div>
+    <div class="setting-item">
+      <label>top_p = <span id="top-p-value">0.95</span></label>
+      <input type="range" id="top-p" min="0.01" max="1" step="0.01" value="0.95">
+      <p class="setting-description">Controls diversity via nucleus sampling. Reduces the probability mass from which to sample.</p>
+    </div>
+    <div class="setting-item">
+      <label>presence_penalty = <span id="presence-value">0</span></label>
+      <input type="range" id="presence-penalty" class="penalty-range" min="-2" max="2" step="0.1" value="0">
+      <p class="setting-description">Positive values discourage repetition; negative values encourage it.</p>
+    </div>
+    <div class="setting-item">
+      <label>frequency_penalty = <span id="frequency-value">0</span></label>
+      <input type="range" id="frequency-penalty" class="penalty-range" min="-2" max="2" step="0.1" value="0">
+      <p class="setting-description">Penalizes new tokens based on their frequency in the text so far.</p>
+    </div>
   </div>
 </div>
 <div id="bottom"></div>

Original file line number	Diff line number	Diff line change
`@@ -46,7 +46,7 @@ class HighlightDom extends Highlight {`
`46`	`46`	`feed(s) {`
`47`	`47`	`for (let i = 0; i < s.length; ++i) {`
`48`	`48`	`this.text += s[i];`
`49`		`- if (isspace(s[i])) {`
	`49`	`+ if (isspace(s[i]) \|\| this.text.length > 50) {`
`50`	`50`	`this.flushText();`
`51`	`51`	`}`
`52`	`52`	`}`