refactor model profile logic by moving rules to modelProfiles.js and updating relevant references

2026-09-08 23:53:13 +02:00
parent c471a4794d
commit da51bd88e6
5 changed files with 104 additions and 48 deletions
@@ -14,8 +14,3 @@ export const themes = {
    "beta-ai": {name: "Beta AI", short: "BA"},
    "confusity": {name: "Confusity", short: "CF"}
 };
-
-export const defaultModelProfile = {
-    badge: "General",
-    summary: "General-purpose chat model. Start here if you are unsure which model to use."
-};
@@ -0,0 +1,89 @@
+export const defaultModelProfile = {
+    badge: "General",
+    summary: "Best for general questions, drafting, and everyday conversation. Use a specialized model when you need stronger coding, reasoning, or roleplay behavior."
+};
+
+const modelProfileRules = [
+    {
+        matches: ["catgirl"],
+        badge: "Roleplay",
+        summary: "Best for playful catgirl character roleplay and expressive, in-character conversation. Not intended for neutral factual answers or technical work."
+    },
+    {
+        matches: ["femboy"],
+        badge: "Roleplay",
+        summary: "Best for personality-led roleplay, casual conversation, and maintaining a stylized character voice. Not intended for precise research or technical tasks."
+    },
+    {
+        matches: ["buenzli", "bünzli"],
+        badge: "Roleplay",
+        summary: "Best for Swiss-flavored character roleplay, local humor, and casual conversation. Choose a general model for neutral writing or factual questions."
+    },
+    {
+        matches: ["deepseek-r1", "deepseek-r1-distill"],
+        badge: "Reasoning",
+        summary: "Best for problems that benefit from deliberate reasoning, such as logic, mathematics, planning, and difficult technical questions. It can be slower and more verbose for simple chat."
+    },
+    {
+        matches: ["codellama", "code-llama", "deepseek-coder", "qwen2.5-coder", "qwen3-coder", "starcoder", "codestral", "coder"],
+        badge: "Coding",
+        summary: "Best for writing, explaining, reviewing, and debugging code. Choose a general model for creative conversation or broad non-technical questions."
+    },
+    {
+        matches: ["dolphin"],
+        badge: "Creative",
+        summary: "Best for open-ended conversation, brainstorming, creative writing, and roleplay. Its answers can be less restrained and less dependable for factual or safety-critical topics."
+    },
+    {
+        matches: ["mixtral"],
+        badge: "Versatile",
+        summary: "Best for detailed writing, summarization, multilingual tasks, and general problem-solving. It is a capable all-rounder but may respond more slowly than smaller models."
+    },
+    {
+        matches: ["mistral"],
+        badge: "Balanced",
+        summary: "Best for everyday questions, concise writing, summaries, and light coding. A strong default when you want a balance of speed and answer quality."
+    },
+    {
+        matches: ["gemma"],
+        badge: "Efficient",
+        summary: "Best for quick answers, rewriting, summarization, and lightweight general chat. Larger or reasoning-focused models are better for complex multi-step problems."
+    },
+    {
+        matches: ["qwen"],
+        badge: "Versatile",
+        summary: "Best for multilingual conversation, structured writing, instruction following, and general technical tasks. Coder variants are the better choice for code-heavy work."
+    },
+    {
+        matches: ["llama"],
+        badge: "General",
+        summary: "Best for general conversation, writing, summarization, and broad knowledge tasks. Specialized coding or reasoning models may perform better in those areas."
+    },
+    {
+        matches: ["phi"],
+        badge: "Compact",
+        summary: "Best for fast, focused questions, short summaries, and lightweight assistance. Its compact size makes it less suitable for demanding reasoning or long, nuanced responses."
+    }
+];
+
+function describeBuild(model) {
+    const parameterSize = String(model?.details?.parameter_size || "").trim();
+    const quantization = String(model?.details?.quantization_level || "").trim();
+    const details = [parameterSize, quantization].filter(Boolean);
+    return details.length ? ` Available build: ${details.join(", ")}.` : "";
+}
+
+export function getModelProfile(model = {}) {
+    const modelId = String(model.model || model.name || "");
+    const modelName = String(model.name || model.model || "");
+    const family = String(model?.details?.family || "");
+    const families = Array.isArray(model?.details?.families) ? model.details.families.join(" ") : "";
+    const haystack = `${modelId} ${modelName} ${family} ${families}`.toLowerCase();
+    const match = modelProfileRules.find((profile) => profile.matches.some((token) => haystack.includes(token)));
+    const profile = match || defaultModelProfile;
+
+    return {
+        badge: profile.badge,
+        summary: `${profile.summary}${describeBuild(model)}`
+    };
+}