Model comparison
deepseek-math-7b-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 4,221,964,704.
From the file· summed bytes, KV per layer
Side by side
| deepseek-math-7b-instruct | HyperCLOVAX-SEED-Text-Instruct-1.5B | |
|---|---|---|
| Parameters | 6.9B | 1.6B |
| Architecture | llama | llama |
| Layers | 30 | — |
| Native context | 4,096 | — |
| Mixture of experts | no | no |
| Quantizations published | 38 | 1 |
| Smallest quantization | 2.53 GiB | 1.06 GiB |
| Q4_K_M | 3.93 GiB | 1.06 GiB |
| Licence | other | other |
KV cache by context
the term that decides long-context viability
| Context | deepseek-math-7b-instruct | HyperCLOVAX-SEED-Text-Instruct-1.5B | Ratio |
|---|---|---|---|
| 4,096 | 1.88 GiB | — | — |
| 8,192 | 3.75 GiB | — | — |
| 16,384 | 7.50 GiB | — | — |
| 32,768 | 15.00 GiB | — | — |
| 65,536 | 30.00 GiB | — | — |
| 131,072 | 60.00 GiB | — | — |