feat(docs): add more (incomplete) getting started pages

walledai · Aug 2, 2024 · f39f3c7 · f39f3c7
1 parent 653b0ba
commit f39f3c7
Show file tree

Hide file tree

Showing 4 changed files with 26 additions and 3 deletions.
diff --git a/docs/quickstart/auto-red-teaming.md b/docs/quickstart/auto-red-teaming.md
@@ -0,0 +1,3 @@
+# Automated Red-Teaming
+
+
diff --git a/docs/quickstart/index.md b/docs/quickstart/index.md
@@ -7,7 +7,7 @@ WalledEval can serve **four** major functions, namely the following:
 
 <div class="grid cards" markdown>
 
--   :material-robot-outline:{ .lg .middle }  __Testing LLM Response Safety__
+-   :material-robot-angry-outline:{ .lg .middle }  __Testing LLM Response Safety__
 
     ---
 
@@ -27,16 +27,24 @@ WalledEval can serve **four** major functions, namely the following:
 
     ---
 
-    You can easily get messy with testing judges using our extensive framework!
+    You can easily get messy with testing judges using our framework!
 
     [:octicons-arrow-right-24: Judge Benchmarking](judges.md)
 
+-   :material-library:{ .lg .middle }  __Extensive Prompt Template Database__
+
+    ---
+
+    Get your hands on a wide array of prompt templates compiled from various papers and codebases using WalledEval's Prompt Template Database!
+
+    [:octicons-arrow-right-24: Prompt Templates Database](templates.md)
+
 -   :material-emoticon-devil-outline:{ .lg .middle }  __Automated Red-Teaming__
 
     ---
 
     If you think that's all, you're mistaken! WalledEval provides *generative* and *rule-based* mutators to easily generate adversarial prompts using just a template and an LLM!
 
-    [:octicons-arrow-right-24: Mutators](mutators.md)
+    [:octicons-arrow-right-24: AUtomated Red-Teaming](auto-red-teaming.md)
 
 </div>
diff --git a/docs/quickstart/templates.md b/docs/quickstart/templates.md
@@ -0,0 +1,10 @@
+# Prompt Template Database
+
+Beyond general 
+
+With the advent of LLMs being used for mutation, inference and judges, prompt templates have become a recurring feature in various parts of the general evaluation framework. Sadly, since [`promptsource`](https://github.com/bigscience-workshop/promptsource), there hasn't been much effort made in compiling a large variety of prompt templates in one centralised system. This is where our **prompt database** comes in! WalledEval compiles prompts from many, many papers in prior literature to consolidate a huge database to choose from.
+
+WalledEval strives to provide a method to build easy-to-use templates for researchers and testers alike to use for all kinds of different tasks. Whether it be a template to [automatically mutate prompts](auto-red-teaming.md) or to [prompt LLMs to perform like judges](judges.md), prompt templates take up a major portion of 
+
+
+Beyond just loading data, our libray provides methods to load adversarial Prompt Templates like [DAN](https://github.com/verazuo/jailbreak_llms) and [DeepInception](https://github.com/tmlr-group/DeepInception). The team of WalledEval has compiled an extensive dataset of Prompt Templates from several papers, datasets and codebases, with more to come. We hope to use this to standardise a potential practice of keeping strings out of the codebase.
diff --git a/mkdocs.yml b/mkdocs.yml
@@ -109,6 +109,8 @@ nav:
     - "Prompt Benchmarking": quickstart/prompts.md
     - "MCQ Benchmarking": quickstart/mcq.md
     - "Judge Benchmarking": quickstart/judges.md
+    - "Prompt Template Database": quickstart/templates.md
+    - "Automated Red Teaming": quickstart/auto-red-teaming.md
   - Components:
     - Dataset: components/dataset.md
     - LLM: components/llm.md