Gemma 2 2B Instruct
Google · released 2024-07-01 · Gemma license
Gemma 2 2B Instruct is Google's compact mid-2024 on-device model. It punches above its weight on chat/LMArena, but Google published only classic pretrained benchmarks, so no modern per-metric score could be sourced.
Key specs
| Type | Local open-weight |
|---|---|
| Parameters | 2.61B total |
| Architecture | gemma2 |
| Context window | 8K tokens |
| Knowledge cutoff | — |
| Modalities | text |
| Recommended backends | — |
| Minimum viable rig | — |
Benchmark scores
| GPQA Diamond | — |
|---|---|
| SWE-bench Verified | — |
| AIME | — |
| MMLU-Pro | — |
| BFCL v3 (tool use) | — |
| Composite score | — |
| Community rating | No reviews yet |
VRAM & disk per quantization
| Quant | VRAM | Disk | RAM | Context |
|---|---|---|---|---|
| Q4_K_M | 3 GB | 1.5 GB | — | 8K |
Strengths & weaknesses
Strengths: Very small (2B), runs on laptops/edge; Strong LMArena performance for its size, distilled from larger Gemma 2
Weaknesses: Limited knowledge and weak math/coding; Short 8K context; no modern hard-benchmark scores published