Can LFM2.5 2.6B run on the RTX 3070?
Yes. At 8K context the best fit is Q8_0, using about 3.4 GB of the 7.6 GB usable on the RTX 3070. Estimated speed ceiling: ~156 tokens/sec.
Memory by quant and context
| Quant | 4K ctx | 8K ctx | 32K ctx | 128K ctx | Max tok/s (est., 8K) |
|---|---|---|---|---|---|
| 2-bit (Q2_K) | 1.6 GB Fits | 1.7 GB Fits | 2.1 GB Fits | 3.6 GB Fits | 443 |
| 3-bit (Q3_K_M) | 1.9 GB Fits | 2.0 GB Fits | 2.4 GB Fits | 3.9 GB Fits | 340 |
| 4-bit (Q4_K_M) • | 2.3 GB Fits | 2.3 GB Fits | 2.7 GB Fits | 4.2 GB Fits | 268 |
| 5-bit (Q5_K_M) • | 2.5 GB Fits | 2.6 GB Fits | 3.0 GB Fits | 4.5 GB Fits | 231 |
| 6-bit (Q6_K) • | 2.8 GB Fits | 2.8 GB Fits | 3.2 GB Fits | 4.7 GB Fits | 202 |
| 8-bit (Q8_0) • | 3.4 GB Fits | 3.4 GB Fits | 3.8 GB Fits | 5.3 GB Fits | 156 |
| 16-bit (F16/BF16) • | 10.8 GB No | 10.8 GB No | 11.2 GB No | 12.7 GB No | — |
• real GGUF file size from LiquidAI/LFM2.5-2.6B-GGUF. Speed is a bandwidth ceiling (448 GB/s), not a benchmark. Methodology.
More
LFM2.5 2.6B on every GPU and Mac · Every model that runs on the RTX 3070 · Other models that fit 8 GB VRAM · Try other settings in the calculator
Data: Hugging Face Hub (LiquidAI/LFM2.5-2.6B), NVIDIA specs. Data as of 2026-11-17.