Llama 3.2 1B Instruct
unsloth · released 2024-09-25 · llama3.2 license
Llama 3.2 1B Instruct is Meta's smallest on-device model (Dec-2023 cutoff): vendor IFEval 59.5; good for lightweight rewriting/summarization, very limited reasoning (MMLU 49.3).
Key specs
| Type | Local open-weight |
|---|---|
| Parameters | 1.24B total |
| Architecture | llama |
| Context window | 131K tokens |
| Knowledge cutoff | 2023-12-01 |
| Modalities | text |
| Recommended backends | — |
| Minimum viable rig | — |
Benchmark scores
| GPQA Diamond | — |
|---|---|
| SWE-bench Verified | — |
| AIME | — |
| MMLU-Pro | — |
| BFCL v3 (tool use) | — |
| Composite score | 5.95 |
| Community rating | No reviews yet |
VRAM & disk per quantization
| Quant | VRAM | Disk | RAM | Context |
|---|---|---|---|---|
| Q4_K_M | 2.2 GB | 0.7 GB | — | 131K |
Strengths & weaknesses
Strengths: Runs in highly constrained / on-device settings (~2.5GB), 128K context; Decent instruction following for 1B (IFEval 59.5)
Weaknesses: Very limited knowledge/reasoning (MMLU 49.3, GPQA ~chance); Not for complex coding/agentic tasks