Model comparison
falcon-mamba-7b-instruct vs Qwen3.8-27B
At Q4_K_M, falcon-mamba-7b-instruct is the smaller download — 4,204,230,784 bytes against 16,810,714,336.
From the file· summed bytes, KV per layer
Side by side
| falcon-mamba-7b-instruct | Qwen3.8-27B | |
|---|---|---|
| Parameters | 7.3B | 27.8B |
| Architecture | mamba | qwen35 |
| Layers | 64 | 64 |
| Native context | — | 262,144 |
| Mixture of experts | no | no |
| Quantizations published | 1 | 22 |
| Smallest quantization | 3.92 GiB | 8.39 GiB |
| Q4_K_M | 3.92 GiB | 15.66 GiB |
| Licence | other | apache-2.0 |
KV cache by context
the term that decides long-context viability
| Context | falcon-mamba-7b-instruct | Qwen3.8-27B | Ratio |
|---|---|---|---|
| 4,096 | — | 0.25 GiB | — |
| 8,192 | — | 0.50 GiB | — |
| 16,384 | — | 1.00 GiB | — |
| 32,768 | — | 2.00 GiB | — |
| 65,536 | — | 4.00 GiB | — |
| 131,072 | — | 8.00 GiB | — |