Upload models.json with huggingface_hub

models.json

@@ -3,6 +3,12 @@
   "updated_at": "2026-04-10",
   "min_app_version": "1.0",
   "categories": [
+    {
+      "id": "llm",
+      "name": "Large Language Models",
+      "icon": "bubble.left.and.text.bubble.right",
+      "order": 0
+    },
     {
       "id": "segmentation",
       "name": "Segmentation",
@@ -65,6 +71,48 @@
     }
   ],
   "models": [
+    {
+      "id": "gemma4_e2b",
+      "name": "Gemma 4 E2B",
+      "subtitle": "Google DeepMind, 2025",
+      "category_id": "llm",
+      "description_md": "Google's latest on-device multimodal LLM. 2.3B effective parameters with Per-Layer Embeddings. Text + image input, streaming text output. Runs on Apple Neural Engine at ~31 tok/s decode. Supports multi-turn conversations, image understanding, and reasoning.",
+      "thumbnail_url": "https://huggingface.co/mlboydaisuke/coreml-zoo/resolve/main/thumbnails/gemma4.jpg",
+      "demo": {
+        "template": "chat",
+        "config": {
+          "max_tokens": 1024,
+          "multimodal": true
+        }
+      },
+      "files": [
+        {
+          "name": "gemma4-e2b-coreml.zip",
+          "url": "https://huggingface.co/mlboydaisuke/gemma-4-E2B-coreml/resolve/main/gemma4-e2b-coreml.zip",
+          "archive": "zip",
+          "size_bytes": 2700000000,
+          "sha256": "TODO",
+          "compute_units": "cpuAndNeuralEngine",
+          "kind": "model"
+        }
+      ],
+      "requirements": {
+        "min_ios": "18.0",
+        "min_ram_mb": 1500,
+        "device_capabilities": [
+          "arm64"
+        ]
+      },
+      "license": {
+        "name": "Gemma",
+        "url": "https://ai.google.dev/gemma/terms"
+      },
+      "upstream": {
+        "name": "google/gemma-4-e2b",
+        "url": "https://huggingface.co/google/gemma-4-e2b",
+        "year": 2025
+      }
+    },
     {
       "id": "rmbg_1_4",
       "name": "RMBG-1.4",