Model comparison
MARTHA-LXVII.8B_QWEN-3.5-3.6_prune_9b-3.6_base vs llama-3-youko-8b
These two publish different quantization sets; the table below has the exact sizes. At long context the gap widens: MARTHA-LXVII.8B_QWEN-3.5-3.6_prune_9b-3.6_base's KV cache at 32K is 4.6× smaller, which usually matters more than the difference in weights.
From the file· summed bytes, KV per layer
Side by side
| MARTHA-LXVII.8B_QWEN-3.5-3.6_prune_9b-3.6_base | llama-3-youko-8b | |
|---|---|---|
| Parameters | 8.1B | 8.0B |
| Architecture | qwen35 | llama |
| Layers | 28 | 32 |
| Native context | 262,144 | 8,192 |
| Mixture of experts | no | no |
| Quantizations published | 36 | 2 |
| Smallest quantization | 2.35 GiB | 5.34 GiB |
| Q4_K_M | 4.76 GiB | — |
| Licence | apache-2.0 | llama3 |
KV cache by context
the term that decides long-context viability
| Context | MARTHA-LXVII.8B_QWEN-3.5-3.6_prune_9b-3.6_base | llama-3-youko-8b | Ratio |
|---|---|---|---|
| 4,096 | 0.11 GiB | 0.50 GiB | 4.57× |
| 8,192 | 0.22 GiB | 1.00 GiB | 4.57× |
| 16,384 | 0.44 GiB | 2.00 GiB | 4.57× |
| 32,768 | 0.88 GiB | 4.00 GiB | 4.57× |
| 65,536 | 1.75 GiB | 8.00 GiB | 4.57× |
| 131,072 | 3.50 GiB | 16.00 GiB | 4.57× |