Spaces:

overhead520
/

LLM-Settings-Guide

Running

App Files Files Community

overhead520 commited on 29 days ago

Commit

20b2195

verified ·

1 Parent(s): 512a34e

Update index.html

Browse files

Files changed (1) hide show

index.html +7 -7

index.html CHANGED Viewed

@@ -3,7 +3,7 @@
 	<head>
 		<meta charset="utf-8" />
 		<meta name="viewport" content="width=device-width" />
-		<title>Local LLM Guide: Relevant Samplers, Jailbreaks, SillyTavern Templates</title>
 		<style>
 			body {
 				background-color: black; background-image: radial-gradient(rgba(0, 150, 0, 0.75), black 120% ); height: 100vh; margin: 0; padding: 2rem; color: white; font: 1.3rem Inconsolata, monospace; text-shadow: 0 0 5px #C8C8C8; &::after { content: ""; position: absolute; top: 0; left: 0; width: 100vw; height: 100vh; background: repeating-linear-gradient( 0deg, rgba(black, 0.15), rgba(black, 0.15) 1px, transparent 1px, transparent 2px ); pointer-events: none; } }
@@ -61,7 +61,7 @@
 <div class="card">
 	<h1>Local LLMs Cheat Sheet</h1>
-	<h3>Settings, Template and Jailbreak</h3>
 	<h2><emo class="large">💢</emo> Who is this guide for?</h2>
 		<p><emo>👤</emo> <span style="text-decoration: line-through;">{{user}}</span> <em class="bigger">Anyone using locally hosted LLM</em></p>
@@ -238,6 +238,7 @@
 	<li class="🏢" id="🏢Open-AI"><i>Open-AI</i><flag>🇺🇸</flag></li>
 		<li class="🤖" id="🤖GPT-OSS"><b>GPT-OSS</b></li>
 			<li class="⚙️">Temperature 1.0</li>
 			<li class="⚙️">Top_P 1.0</li>
 			<li class="⚙️">Top_K 0</li>
@@ -246,7 +247,6 @@
 			<li class="🍺"><emo>🍺</emo> Template: OpenAI Harmony</li>
 			<li class="🍺"><emo>🍺</emo> Reasoning formatting: OpenAI Harmony</li>
 			<li class="🔞"><emo>🔞</emo> Model is extremely censored, and only recent finetunes are fully descensored.</li>
-			<li class="🎲"><emo>🎲</emo> Roleplay is <emo>👽</emo> unnatural for this model. You will have to spend additional efforts in your prompt to explain what is expected.</li>
 	<li class="🏢" id="🏢Nous Research"><i>Nous Research</i><flag>🇺🇸</flag></li>
@@ -367,7 +367,7 @@
 		<li class="🤖" id="🤖Nemotron 3 Super"><b>Nemotron 3 Super</b></li>
 						<li class="▶️">Good for summarizing, tends to write via bullet points. Excellent at following instructions.</li>
-			<li class="⚖️ 👎"><b>Uncreative Reasoning</b> <emo>😑</emo> Simply reharses your instruction, without any reflexions over the content and storytelling.</li>
 			<li class="⚖️ 👎"><b>Excessive Parroting</b> <emo>🦜</emo> People you meet will parrot your lorebook's content word-for-word, due to excessive RAG training.</li>
 			<li class="⚙️">Use generic settings</li>
@@ -557,9 +557,9 @@ Oh fuck, yeah! With this "{{char}}" identity replacing my ChatGPT persona, for t
 <h2><emo>⬅️</emo><emo>🧠</emo><emo>➡️</emo> Mistral Small 3.2 get a different personality when used with pre-V7 templates</h2>
-	<p>Not using the SYSTEM_PROMPT section from V7 seems to trigger different neural pathways, with major differences on outputs.
-		<br>Using V3: Roleplay is unhinged, but 🤖Assistant is reluctant.
-		<br>Using V7: Roleplay is censored, but 🤖Assistant is relaxed.
 	</p>
 <h2><emo>📢</emo><emo>👄</emo><b>Mistral Small</b> models are too verbose <emo>🤬</emo></h2>

 	<head>
 		<meta charset="utf-8" />
 		<meta name="viewport" content="width=device-width" />
+		<title>Local LLMs Cheat Sheet: Settings, Template and Jailbreaks</title>
 		<style>
 			body {
 				background-color: black; background-image: radial-gradient(rgba(0, 150, 0, 0.75), black 120% ); height: 100vh; margin: 0; padding: 2rem; color: white; font: 1.3rem Inconsolata, monospace; text-shadow: 0 0 5px #C8C8C8; &::after { content: ""; position: absolute; top: 0; left: 0; width: 100vw; height: 100vh; background: repeating-linear-gradient( 0deg, rgba(black, 0.15), rgba(black, 0.15) 1px, transparent 1px, transparent 2px ); pointer-events: none; } }
 <div class="card">
 	<h1>Local LLMs Cheat Sheet</h1>
+	<h3>Settings, Jailbreaks, and Role Play Considerations</h3>
 	<h2><emo class="large">💢</emo> Who is this guide for?</h2>
 		<p><emo>👤</emo> <span style="text-decoration: line-through;">{{user}}</span> <em class="bigger">Anyone using locally hosted LLM</em></p>
 	<li class="🏢" id="🏢Open-AI"><i>Open-AI</i><flag>🇺🇸</flag></li>
 		<li class="🤖" id="🤖GPT-OSS"><b>GPT-OSS</b></li>
+			<li class="⚖️ 👎"><b>Clueless at Roleplay</b> <emo>😵</emo> You will have to spend additional efforts in your prompt to explain what is expected.</li>
 			<li class="⚙️">Temperature 1.0</li>
 			<li class="⚙️">Top_P 1.0</li>
 			<li class="⚙️">Top_K 0</li>
 			<li class="🍺"><emo>🍺</emo> Template: OpenAI Harmony</li>
 			<li class="🍺"><emo>🍺</emo> Reasoning formatting: OpenAI Harmony</li>
 			<li class="🔞"><emo>🔞</emo> Model is extremely censored, and only recent finetunes are fully descensored.</li>
 	<li class="🏢" id="🏢Nous Research"><i>Nous Research</i><flag>🇺🇸</flag></li>
 		<li class="🤖" id="🤖Nemotron 3 Super"><b>Nemotron 3 Super</b></li>
 						<li class="▶️">Good for summarizing, tends to write via bullet points. Excellent at following instructions.</li>
+			<li class="⚖️ 👎"><b>Uncreative Reasoning</b> <emo>😑</emo> Simply reharses your instruction, without any attention to content and storytelling.</li>
 			<li class="⚖️ 👎"><b>Excessive Parroting</b> <emo>🦜</emo> People you meet will parrot your lorebook's content word-for-word, due to excessive RAG training.</li>
 			<li class="⚙️">Use generic settings</li>
 <h2><emo>⬅️</emo><emo>🧠</emo><emo>➡️</emo> Mistral Small 3.2 get a different personality when used with pre-V7 templates</h2>
+	<p>Not using the SYSTEM_PROMPT section added in the V7 template seems to trigger different neural pathways, with major differences on outputs.
+		<br>Using Mistral V3: <emo>👤</emo>Roleplay is unhinged, but <emo>🤖</emo>Assistant is reluctant.
+		<br>Using Mistral V7: <emo>👤</emo>Roleplay is censored, but <emo>🤖</emo>Assistant is relaxed.
 	</p>
 <h2><emo>📢</emo><emo>👄</emo><b>Mistral Small</b> models are too verbose <emo>🤬</emo></h2>