validated models

app.js

@@ -11,12 +11,136 @@ const THETA4 = 2.045e-7;
 
 const PRESET_MODELS =
 {
-    "Falcon-RW 7B": "tiiuae/falcon-rw-7b",
-    "OPT 1.3B": "facebook/opt-1.3b",
-    "OPT 2.7B": "facebook/opt-2.7b",
-    "OPT 6.7B": "facebook/opt-6.7b",
-    "Qwen2 1.5B": "Qwen/Qwen2-1.5B",
-    "Qwen2 7B": "Qwen/Qwen2-7B"
+    // --------------------------------------------
+    // XS models (~1B)
+    // --------------------------------------------
+
+    "Llama 3.2 1B":
+    {
+        d:2048,
+        L:16,
+        nq:32,
+        nkv:8,
+        attention:"GQA (4x)"
+    },
+
+    "OPT 1.3B":
+    {
+        d:2048,
+        L:24,
+        nq:32,
+        nkv:32,
+        attention:"MHA"
+    },
+
+    "Qwen2 1.5B":
+    {
+        d:1536,
+        L:28,
+        nq:12,
+        nkv:2,
+        attention:"GQA (6x)"
+    },
+
+
+    // --------------------------------------------
+    // S models (~2B-3B)
+    // --------------------------------------------
+
+    "Gemma 2 2B":
+    {
+        d:2304,
+        L:26,
+        nq:8,
+        nkv:4,
+        attention:"GQA (2x)"
+    },
+
+    "OPT 2.7B":
+    {
+        d:2560,
+        L:32,
+        nq:32,
+        nkv:32,
+        attention:"MHA"
+    },
+
+    "Llama 3.2 3B":
+    {
+        d:3072,
+        L:28,
+        nq:24,
+        nkv:6,
+        attention:"GQA (4x)"
+    },
+
+    "Granite 3B":
+    {
+        d:2560,
+        L:32,
+        nq:32,
+        nkv:32,
+        attention:"MHA"
+    },
+
+
+    // --------------------------------------------
+    // M models (~6B-9B)
+    // --------------------------------------------
+
+    "OPT 6.7B":
+    {
+        d:4096,
+        L:32,
+        nq:32,
+        nkv:32,
+        attention:"MHA"
+    },
+
+    "Qwen2 7B":
+    {
+        d:3584,
+        L:28,
+        nq:28,
+        nkv:4,
+        attention:"GQA (7x)"
+    },
+
+    "Falcon-RW 7B":
+    {
+        d:4096,
+        L:36,
+        nq:64,
+        nkv:64,
+        attention:"MHA"
+    },
+
+    "Granite 8B":
+    {
+        d:4096,
+        L:36,
+        nq:32,
+        nkv:8,
+        attention:"GQA (4x)"
+    },
+
+    "Llama 3.1 8B":
+    {
+        d:4096,
+        L:32,
+        nq:32,
+        nkv:8,
+        attention:"GQA (4x)"
+    },
+
+    "Gemma 2 9B":
+    {
+        d:3584,
+        L:42,
+        nq:16,
+        nkv:8,
+        attention:"GQA (2x)"
+    }
 };
 
 window.onload = function()
@@ -44,14 +168,36 @@ window.onload = function()
 
 }
 
-async function loadPreset()
+function loadPreset()
 {
 
-    let repo =
+    let name =
         document.getElementById("preset").value;
 
 
-    await loadRepo(repo);
+    let model =
+        PRESET_MODELS[name];
+
+
+    document.getElementById("d").value=model.d;
+    document.getElementById("L").value=model.L;
+    document.getElementById("nq").value=model.nq;
+    document.getElementById("nkv").value=model.nkv;
+
+
+    document.getElementById("status").innerHTML =
+    `
+    ✅ Loaded validated model:
+    <b>${name}</b><br>
+    Attention: ${model.attention}<br>
+    d=${model.d},
+    L=${model.L},
+    nq=${model.nq},
+    nkv=${model.nkv}
+    `;
+
+
+    draw();
 
 }
 
@@ -433,7 +579,7 @@ async function loadRepo(repo)
 
         if(!response.ok)
             throw Error(
-                "Could not load config.json"
+                "Could not load config.json. Some models (e.g. meta-llama models require licenses)"
             );