Loading…
Loading…
Models / SmolLM3
1 model (3B) · ↓ 853.8K downloads a month · from $0.08/h on GTX 1660 SUPER right now. Each model's versions do the same job; quantized ones (AWQ, FP8, 4-bit) need less GPU memory and run on cheaper machines.
| Version | Runtime | Weights | GPU memory | Cheapest machine now |
|---|---|---|---|---|
| OriginalHuggingFaceTB/SmolLM3-3B-Base | vLLM | 6.2 GB | 9.7 GB | $0.10/h RTX 3060 |
| InstructHuggingFaceTB/SmolLM3-3B | vLLM | 6.2 GB | 9.7 GB | $0.10/h RTX 3060 |
| trellis 3inst 4bpw kernel (xv0y5ncu)xv0y5ncu/SmolLM3-3B-trellis-3inst-4bpw-kernel | vLLM | 1.9 GB | 5 GB | $0.10/h RTX 3060 |
| Main (inference4j)inference4j/smollm3-3b | ONNX Runtime GenAI | 2.0 GB | 4.4 GB | $0.08/h GTX 1660 SUPER |
Model details come from Hugging Face and are refreshed daily; prices follow the live offers. GPU memory is GridPup's estimate: the weights, room for an 8,192-token conversation (chat models) and the runtime's own memory. Large models are split across the GPUs of one machine ("4x RTX 4090").