SmolLM2 1.7B
by Hugging Face · smollm family
1.7B
parameters
text-generation code-generation summarization
SmolLM2 1.7B is Hugging Face's compact language model designed for on-device and edge deployment. Trained on 11 trillion tokens from a diverse mix of web data, code, and mathematics datasets, it delivers surprisingly strong performance for its size. Despite being one of the smallest models available, SmolLM2 1.7B outperforms other models in its class on reasoning, knowledge, and instruction-following tasks. Its tiny VRAM footprint makes it ideal for resource-constrained environments where larger models are impractical.
Quick Start with Ollama
ollama run 1.7b-instruct-q8_0 | Creator | Hugging Face |
| Parameters | 1.7B |
| Architecture | transformer-decoder |
| Context | 8K tokens |
| Released | Nov 2, 2024 |
| License | Apache 2.0 |
| Ollama | smollm2 |
Quantization Options
| Format | File Size | VRAM Required | Quality | Ollama Tag |
|---|---|---|---|---|
| Q4_K_M | 1 GB | 1.9 GB | | 1.7b-instruct-q4_K_M |
| Q8_0 rec | 1.8 GB | 2.7 GB | | 1.7b-instruct-q8_0 |
| F16 | 3.4 GB | 4.4 GB | | 1.7b-instruct-fp16 |
Compatible Hardware
Q8_0 requires 2.7 GB VRAM