Upload guide.json with huggingface_hub

guide.json

@@ -0,0 +1,72 @@
+[
+  {
+    "use_case": "Ultra-fast classification",
+    "recommended": "SmolLM2-135M-Instruct-mobile",
+    "size": "101MB",
+    "speed": "59.7 t/s",
+    "why": "Smallest, fastest, handles simple tasks"
+  },
+  {
+    "use_case": "On-device chat (budget phone)",
+    "recommended": "Qwen2.5-0.5B-Instruct-mobile-int4",
+    "size": "469MB",
+    "speed": "23.2 t/s (phone)",
+    "why": "Under 500MB, 23 t/s on phone, coherent"
+  },
+  {
+    "use_case": "On-device chat (mid-range)",
+    "recommended": "Llama-3.2-1B-Instruct-Q4-mobile",
+    "size": "770MB",
+    "speed": "5.4 t/s (phone)",
+    "why": "1B params, better quality, fits in 2GB RAM"
+  },
+  {
+    "use_case": "Code completion",
+    "recommended": "Qwen2.5-0.5B-Coder-mobile",
+    "size": "1000MB",
+    "speed": "8.0 t/s",
+    "why": "Coder-tuned, small enough for mobile"
+  },
+  {
+    "use_case": "Math/reasoning",
+    "recommended": "Qwen2.5-Math-1.5B-mobile",
+    "size": "3000MB",
+    "speed": "15.7 t/s",
+    "why": "Math-tuned, good at arithmetic"
+  },
+  {
+    "use_case": "Arabic chat",
+    "recommended": "Gemma-2B-Arabic-mobile",
+    "size": "1555MB",
+    "speed": "9.4 t/s",
+    "why": "Arabic-capable (partial), gemma format"
+  },
+  {
+    "use_case": "Function calling",
+    "recommended": "Llama-3.2-1B-FunctionCall-mobile",
+    "size": "1926MB",
+    "speed": "6.0 t/s",
+    "why": "Chat works, FC partial (33% success)"
+  },
+  {
+    "use_case": "Embedding/RAG",
+    "recommended": "EmbeddingGemma-300M-mobile",
+    "size": "600MB",
+    "speed": "N/A",
+    "why": "Embedding model, not chat"
+  },
+  {
+    "use_case": "Best quality",
+    "recommended": "Llama-3.2-3B-Instruct-mobile",
+    "size": "6000MB",
+    "speed": "4.8 t/s",
+    "why": "3B params, best quality, needs 4GB+ RAM"
+  },
+  {
+    "use_case": "Chinese language",
+    "recommended": "Qwen2.5-0.5B-Chinese-mobile",
+    "size": "1000MB",
+    "speed": "16.4 t/s",
+    "why": "Chinese-tuned, verified working"
+  }
+]