use_case stringclasses 10
values | recommended stringclasses 10
values | size stringclasses 9
values | speed stringclasses 10
values | why stringclasses 10
values |
|---|---|---|---|---|
Ultra-fast classification | SmolLM2-135M-Instruct-mobile | 101MB | 59.7 t/s | Smallest, fastest, handles simple tasks |
On-device chat (budget phone) | Qwen2.5-0.5B-Instruct-mobile-int4 | 469MB | 23.2 t/s (phone) | Under 500MB, 23 t/s on phone, coherent |
On-device chat (mid-range) | Llama-3.2-1B-Instruct-Q4-mobile | 770MB | 5.4 t/s (phone) | 1B params, better quality, fits in 2GB RAM |
Code completion | Qwen2.5-0.5B-Coder-mobile | 1000MB | 8.0 t/s | Coder-tuned, small enough for mobile |
Math/reasoning | Qwen2.5-Math-1.5B-mobile | 3000MB | 15.7 t/s | Math-tuned, good at arithmetic |
Arabic chat | Gemma-2B-Arabic-mobile | 1555MB | 9.4 t/s | Arabic-capable (partial), gemma format |
Function calling | Llama-3.2-1B-FunctionCall-mobile | 1926MB | 6.0 t/s | Chat works, FC partial (33% success) |
Embedding/RAG | EmbeddingGemma-300M-mobile | 600MB | N/A | Embedding model, not chat |
Best quality | Llama-3.2-3B-Instruct-mobile | 6000MB | 4.8 t/s | 3B params, best quality, needs 4GB+ RAM |
Chinese language | Qwen2.5-0.5B-Chinese-mobile | 1000MB | 16.4 t/s | Chinese-tuned, verified working |
- Downloads last month
- 7