Model comparison
Vikhr-Gemma-2B-instruct 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 1,708,582,848.
From the file· summed bytes, KV per layer
Side by side
| Vikhr-Gemma-2B-instruct | HyperCLOVAX-SEED-Text-Instruct-1.5B | |
|---|---|---|
| Parameters | 2.6B | 1.6B |
| Architecture | gemma2 | llama |
| Layers | 26 | — |
| Native context | 8,192 | — |
| Mixture of experts | no | no |
| Quantizations published | 46 | 1 |
| Smallest quantization | 0.78 GiB | 1.06 GiB |
| Q4_K_M | 1.59 GiB | 1.06 GiB |
| Licence | apache-2.0 | other |
KV cache by context
the term that decides long-context viability
| Context | Vikhr-Gemma-2B-instruct | HyperCLOVAX-SEED-Text-Instruct-1.5B | Ratio |
|---|---|---|---|
| 4,096 | 0.41 GiB | — | — |
| 8,192 | 0.63 GiB | — | — |
| 16,384 | 1.04 GiB | — | — |
| 32,768 | 1.85 GiB | — | — |
| 65,536 | 3.48 GiB | — | — |
| 131,072 | 6.73 GiB | — | — |