Model comparison
LFM2.5-230M vs SmolLM2-135M-Instruct
At Q4_K_M, SmolLM2-135M-Instruct is the smaller download — 105,454,144 bytes against 153,406,048. At long context the gap widens: LFM2.5-230M's KV cache at 32K is 1.9× smaller, which usually matters more than the difference in weights.
From the file· summed bytes, KV per layer
Side by side
| LFM2.5-230M | SmolLM2-135M-Instruct | |
|---|---|---|
| Parameters | 230M | 135M |
| Architecture | lfm2 | llama |
| Layers | 14 | 30 |
| Native context | 128,000 | 8,192 |
| Mixture of experts | no | no |
| Quantizations published | 25 | 41 |
| Smallest quantization | 0.10 GiB | 0.08 GiB |
| Q4_K_M | 0.14 GiB | 0.10 GiB |
| Licence | other | apache-2.0 |
KV cache by context
the term that decides long-context viability
| Context | LFM2.5-230M | SmolLM2-135M-Instruct | Ratio |
|---|---|---|---|
| 4,096 | 0.05 GiB | 0.09 GiB | 1.87× |
| 8,192 | 0.09 GiB | 0.18 GiB | 1.87× |
| 16,384 | 0.19 GiB | 0.35 GiB | 1.87× |
| 32,768 | 0.38 GiB | 0.70 GiB | 1.87× |
| 65,536 | 0.75 GiB | 1.41 GiB | 1.87× |
| 131,072 | 1.50 GiB | 2.81 GiB | 1.87× |