Model comparison
Qwen2-VL-7B-Instruct vs Qwythos-9B-Claude-Mythos-5-1M
At Q4_K_M, Qwen2-VL-7B-Instruct is the smaller download — 4,683,072,416 bytes against 11,516,776,704. At long context the gap widens: Qwythos-9B-Claude-Mythos-5-1M's KV cache at 32K is 1.7× smaller, which usually matters more than the difference in weights.
From the file· summed bytes, KV per layer
Side by side
| Qwen2-VL-7B-Instruct | Qwythos-9B-Claude-Mythos-5-1M | |
|---|---|---|
| Parameters | 8.3B | 9.4B |
| Architecture | qwen2vl | qwen35 |
| Layers | 28 | 32 |
| Native context | 32,768 | 1,048,576 |
| Mixture of experts | no | no |
| Quantizations published | 28 | 16 |
| Smallest quantization | 2.59 GiB | 5.38 GiB |
| Q4_K_M | 4.36 GiB | 10.73 GiB |
| Licence | apache-2.0 | apache-2.0 |
KV cache by context
the term that decides long-context viability
| Context | Qwen2-VL-7B-Instruct | Qwythos-9B-Claude-Mythos-5-1M | Ratio |
|---|---|---|---|
| 4,096 | 0.22 GiB | 0.13 GiB | 1.75× |
| 8,192 | 0.44 GiB | 0.25 GiB | 1.75× |
| 16,384 | 0.88 GiB | 0.50 GiB | 1.75× |
| 32,768 | 1.75 GiB | 1.00 GiB | 1.75× |
| 65,536 | 3.50 GiB | 2.00 GiB | 1.75× |
| 131,072 | 7.00 GiB | 4.00 GiB | 1.75× |