| [ |
| { |
| "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" |
| } |
| ] |
