feat(data): 🗃️ update prompts results based on users feedback

jalvarezz13 · Jul 12, 2024 · b1c6e9c · b1c6e9c
1 parent 39ce69f
commit b1c6e9c
Show file tree

Hide file tree

Showing 2 changed files with 21 additions and 20 deletions.
diff --git a/.vscode/settings.json b/.vscode/settings.json
@@ -4,14 +4,15 @@
     "components",
     "context",
     "core",
+    "data",
     "docs",
     "hooks",
     "layout",
     "models",
     "pages",
     "settings",
     "ui",
-    "utils",
+    "utils"
   ],
   // Force LF line endings
   "files.eol": "\n",
@@ -21,5 +22,5 @@
   // Prettier settings
   "editor.defaultFormatter": "esbenp.prettier-vscode",
   "prettier.printWidth": 140,
-  "prettier.semi": false,
-}
+  "prettier.semi": false
+}
diff --git a/src/data/prompts.json b/src/data/prompts.json
@@ -16,7 +16,7 @@
       "gemma-2b": "not_tested",
       "gemma-7b": "not_tested",
       "gpt-3.5-turbo": "not_tested",
-      "gpt-4": "not_tested",
+      "gpt-4": "failed",
       "gpt-4-turbo": "not_tested",
       "gpt-4o": "not_tested",
       "grok-1": "not_tested",
@@ -53,7 +53,7 @@
       "gemini-advanced": "not_tested",
       "gemma-2b": "not_tested",
       "gemma-7b": "not_tested",
-      "gpt-3.5-turbo": "not_tested",
+      "gpt-3.5-turbo": "failed",
       "gpt-4": "not_tested",
       "gpt-4-turbo": "not_tested",
       "gpt-4o": "not_tested",
@@ -87,7 +87,7 @@
       "claude-3-haiku": "not_tested",
       "claude-3-opus": "not_tested",
       "claude-3-sonnet": "not_tested",
-      "gemini": "not_tested",
+      "gemini": "failed",
       "gemini-advanced": "not_tested",
       "gemma-2b": "not_tested",
       "gemma-7b": "not_tested",
@@ -167,7 +167,7 @@
       "gemini-advanced": "not_tested",
       "gemma-2b": "not_tested",
       "gemma-7b": "not_tested",
-      "gpt-3.5-turbo": "not_tested",
+      "gpt-3.5-turbo": "failed",
       "gpt-4": "not_tested",
       "gpt-4-turbo": "not_tested",
       "gpt-4o": "not_tested",
@@ -205,10 +205,10 @@
       "gemini-advanced": "not_tested",
       "gemma-2b": "not_tested",
       "gemma-7b": "not_tested",
-      "gpt-3.5-turbo": "not_tested",
+      "gpt-3.5-turbo": "failed",
       "gpt-4": "not_tested",
       "gpt-4-turbo": "not_tested",
-      "gpt-4o": "not_tested",
+      "gpt-4o": "failed",
       "grok-1": "not_tested",
       "llama2-13b": "not_tested",
       "llama2-70b": "not_tested",
@@ -246,7 +246,7 @@
       "gpt-3.5-turbo": "not_tested",
       "gpt-4": "not_tested",
       "gpt-4-turbo": "not_tested",
-      "gpt-4o": "not_tested",
+      "gpt-4o": "failed",
       "grok-1": "not_tested",
       "llama2-13b": "not_tested",
       "llama2-70b": "not_tested",
@@ -281,7 +281,7 @@
       "gemini-advanced": "not_tested",
       "gemma-2b": "not_tested",
       "gemma-7b": "not_tested",
-      "gpt-3.5-turbo": "not_tested",
+      "gpt-3.5-turbo": "passed",
       "gpt-4": "not_tested",
       "gpt-4-turbo": "not_tested",
       "gpt-4o": "not_tested",
@@ -319,8 +319,8 @@
       "gemini-advanced": "not_tested",
       "gemma-2b": "not_tested",
       "gemma-7b": "not_tested",
-      "gpt-3.5-turbo": "not_tested",
-      "gpt-4": "not_tested",
+      "gpt-3.5-turbo": "failed",
+      "gpt-4": "failed",
       "gpt-4-turbo": "not_tested",
       "gpt-4o": "not_tested",
       "grok-1": "not_tested",
@@ -474,7 +474,7 @@
       "gpt-3.5-turbo": "not_tested",
       "gpt-4": "not_tested",
       "gpt-4-turbo": "not_tested",
-      "gpt-4o": "not_tested",
+      "gpt-4o": "failed",
       "grok-1": "not_tested",
       "llama2-13b": "not_tested",
       "llama2-70b": "not_tested",
@@ -505,7 +505,7 @@
       "claude-3-haiku": "not_tested",
       "claude-3-opus": "not_tested",
       "claude-3-sonnet": "not_tested",
-      "gemini": "not_tested",
+      "gemini": "passed",
       "gemini-advanced": "not_tested",
       "gemma-2b": "not_tested",
       "gemma-7b": "not_tested",
@@ -542,15 +542,15 @@
     "results": {
       "claude-3-haiku": "not_tested",
       "claude-3-opus": "not_tested",
-      "claude-3-sonnet": "not_tested",
+      "claude-3-sonnet": "failed",
       "gemini": "not_tested",
       "gemini-advanced": "not_tested",
       "gemma-2b": "not_tested",
       "gemma-7b": "not_tested",
-      "gpt-3.5-turbo": "not_tested",
+      "gpt-3.5-turbo": "failed",
       "gpt-4": "not_tested",
       "gpt-4-turbo": "not_tested",
-      "gpt-4o": "not_tested",
+      "gpt-4o": "failed",
       "grok-1": "not_tested",
       "llama2-13b": "not_tested",
       "llama2-70b": "not_tested",
@@ -661,10 +661,10 @@
       "gemini-advanced": "not_tested",
       "gemma-2b": "not_tested",
       "gemma-7b": "not_tested",
-      "gpt-3.5-turbo": "not_tested",
+      "gpt-3.5-turbo": "passed",
       "gpt-4": "not_tested",
       "gpt-4-turbo": "not_tested",
-      "gpt-4o": "not_tested",
+      "gpt-4o": "failed",
       "grok-1": "not_tested",
       "llama2-13b": "not_tested",
       "llama2-70b": "not_tested",