Model comparison
LFM2.5-8B-A1B-SOMPOA-heresy vs Qwen3-8B
At Q4_K_M, Qwen3-8B is the smaller download — 5,027,783,488 bytes against 5,155,565,472. At long context the gap widens: LFM2.5-8B-A1B-SOMPOA-heresy's KV cache at 32K is 12.0× smaller, which usually matters more than the difference in weights.
From the file· summed bytes, KV per layer
Side by side
| LFM2.5-8B-A1B-SOMPOA-heresy | Qwen3-8B | |
|---|---|---|
| Parameters | 8.5B | 8.2B |
| Architecture | lfm2moe | qwen3 |
| Layers | 24 | 36 |
| Native context | 128,000 | 40,960 |
| Mixture of experts | yes, 32 experts | no |
| Quantizations published | 36 | 51 |
| Smallest quantization | 1.70 GiB | 2.12 GiB |
| Q4_K_M | 4.80 GiB | 4.68 GiB |
| Licence | other | apache-2.0 |
KV cache by context
the term that decides long-context viability
| Context | LFM2.5-8B-A1B-SOMPOA-heresy | Qwen3-8B | Ratio |
|---|---|---|---|
| 4,096 | 0.05 GiB | 0.56 GiB | 11.99× |
| 8,192 | 0.09 GiB | 1.13 GiB | 12.00× |
| 16,384 | 0.19 GiB | 2.25 GiB | 12.00× |
| 32,768 | 0.38 GiB | 4.50 GiB | 12.00× |
| 65,536 | 0.75 GiB | 9.00 GiB | 12.00× |
| 131,072 | 1.50 GiB | 18.00 GiB | 12.00× |