Model comparison
Huihui-Qwen3.5-4B-Claude-4.6-Opus-abliterated vs Llama-3.1-8B-Instruct
At Q4_K_M, Huihui-Qwen3.5-4B-Claude-4.6-Opus-abliterated is the smaller download — 2,708,801,536 bytes against 4,920,739,168. At long context the gap widens: Huihui-Qwen3.5-4B-Claude-4.6-Opus-abliterated's KV cache at 32K is 4.0× smaller, which usually matters more than the difference in weights.
From the file· summed bytes, KV per layer
Side by side
| Huihui-Qwen3.5-4B-Claude-4.6-Opus-abliterated | Llama-3.1-8B-Instruct | |
|---|---|---|
| Parameters | 4.7B | 8.0B |
| Architecture | qwen35 | llama |
| Layers | 32 | 32 |
| Native context | 262,144 | 131,072 |
| Mixture of experts | no | no |
| Quantizations published | 36 | 45 |
| Smallest quantization | 1.27 GiB | 2.02 GiB |
| Q4_K_M | 2.52 GiB | 4.58 GiB |
| Licence | apache-2.0 | llama3.1 |
KV cache by context
the term that decides long-context viability
| Context | Huihui-Qwen3.5-4B-Claude-4.6-Opus-abliterated | Llama-3.1-8B-Instruct | Ratio |
|---|---|---|---|
| 4,096 | 0.13 GiB | 0.50 GiB | 4.00× |
| 8,192 | 0.25 GiB | 1.00 GiB | 4.00× |
| 16,384 | 0.50 GiB | 2.00 GiB | 4.00× |
| 32,768 | 1.00 GiB | 4.00 GiB | 4.00× |
| 65,536 | 2.00 GiB | 8.00 GiB | 4.00× |
| 131,072 | 4.00 GiB | 16.00 GiB | 4.00× |