Model comparison
Qwen2-VL-2B-Instruct vs Unlimited-OCR
At Q4_K_M, Qwen2-VL-2B-Instruct is the smaller download — 986,046,944 bytes against 1,950,321,792.
From the file· summed bytes, KV per layer
Side by side
| Qwen2-VL-2B-Instruct | Unlimited-OCR | |
|---|---|---|
| Parameters | 2.2B | 3.3B |
| Architecture | qwen2vl | deepseek2-ocr |
| Layers | 28 | 12 |
| Native context | 32,768 | 32,768 |
| Mixture of experts | no | yes, 64 experts |
| Quantizations published | 25 | 21 |
| Smallest quantization | 0.56 GiB | 1.15 GiB |
| Q4_K_M | 0.92 GiB | 1.82 GiB |
| Licence | apache-2.0 | mit |
KV cache by context
the term that decides long-context viability
| Context | Qwen2-VL-2B-Instruct | Unlimited-OCR | Ratio |
|---|---|---|---|
| 4,096 | 0.11 GiB | — | — |
| 8,192 | 0.22 GiB | — | — |
| 16,384 | 0.44 GiB | — | — |
| 32,768 | 0.88 GiB | — | — |
| 65,536 | 1.75 GiB | — | — |
| 131,072 | 3.50 GiB | — | — |