Model comparison
oxy-1-small vs Qwen-AgentWorld-35B-A3B
These two publish different quantization sets; the table below has the exact sizes. At long context the gap widens: Qwen-AgentWorld-35B-A3B's KV cache at 32K is 9.6× smaller, which usually matters more than the difference in weights.
From the file· summed bytes, KV per layer
Side by side
| oxy-1-small | Qwen-AgentWorld-35B-A3B | |
|---|---|---|
| Parameters | 14.8B | 34.7B |
| Architecture | qwen2 | qwen35moe |
| Layers | 48 | 40 |
| Native context | 32,768 | 262,144 |
| Mixture of experts | no | yes, 256 experts |
| Quantizations published | 22 | 15 |
| Smallest quantization | 4.66 GiB | 10.71 GiB |
| Q4_K_M | 8.37 GiB | — |
| Licence | apache-2.0 | apache-2.0 |
KV cache by context
the term that decides long-context viability
| Context | oxy-1-small | Qwen-AgentWorld-35B-A3B | Ratio |
|---|---|---|---|
| 4,096 | 0.75 GiB | 0.08 GiB | 9.59× |
| 8,192 | 1.50 GiB | 0.16 GiB | 9.60× |
| 16,384 | 3.00 GiB | 0.31 GiB | 9.60× |
| 32,768 | 6.00 GiB | 0.63 GiB | 9.60× |
| 65,536 | 12.00 GiB | 1.25 GiB | 9.60× |
| 131,072 | 24.00 GiB | 2.50 GiB | 9.60× |