Model comparison
Qwen3-VL-4B-Instruct-c_abliterated-v2 vs HyperCLOVAX-SEED-Text-Instruct-1.5B
At Q4_K_M, HyperCLOVAX-SEED-Text-Instruct-1.5B is the smaller download — 1,133,974,368 bytes against 2,497,282,848.
From the file· summed bytes, KV per layer
Side by side
| Qwen3-VL-4B-Instruct-c_abliterated-v2 | HyperCLOVAX-SEED-Text-Instruct-1.5B | |
|---|---|---|
| Parameters | 4.4B | 1.6B |
| Architecture | qwen3vl | llama |
| Layers | — | — |
| Native context | — | — |
| Mixture of experts | no | no |
| Quantizations published | 12 | 1 |
| Smallest quantization | 1.55 GiB | 1.06 GiB |
| Q4_K_M | 2.33 GiB | 1.06 GiB |
| Licence | apache-2.0 | other |
KV cache by context
the term that decides long-context viability
| Context | Qwen3-VL-4B-Instruct-c_abliterated-v2 | HyperCLOVAX-SEED-Text-Instruct-1.5B | Ratio |
|---|---|---|---|
| 4,096 | — | — | — |
| 8,192 | — | — | — |
| 16,384 | — | — | — |
| 32,768 | — | — | — |
| 65,536 | — | — | — |
| 131,072 | — | — | — |