Model comparison
GRM-Kerlin-3b vs Llama-3.2-1B-Instruct
At Q4_K_M, Llama-3.2-1B-Instruct is the smaller download — 807,690,688 bytes against 2,104,933,344.
From the file· summed bytes, KV per layer
Side by side
| GRM-Kerlin-3b | Llama-3.2-1B-Instruct | |
|---|---|---|
| Parameters | 3.4B | 1.2B |
| Architecture | qwen2 | llama |
| Layers | 36 | 16 |
| Native context | 32,768 | 131,072 |
| Mixture of experts | no | no |
| Quantizations published | 36 | 39 |
| Smallest quantization | 0.83 GiB | 0.39 GiB |
| Q4_K_M | 1.96 GiB | 0.75 GiB |
| Licence | apache-2.0 | — |
KV cache by context
the term that decides long-context viability
| Context | GRM-Kerlin-3b | Llama-3.2-1B-Instruct | Ratio |
|---|---|---|---|
| 4,096 | 0.14 GiB | 0.13 GiB | 1.13× |
| 8,192 | 0.28 GiB | 0.25 GiB | 1.13× |
| 16,384 | 0.56 GiB | 0.50 GiB | 1.13× |
| 32,768 | 1.13 GiB | 1.00 GiB | 1.13× |
| 65,536 | 2.25 GiB | 2.00 GiB | 1.13× |
| 131,072 | 4.50 GiB | 4.00 GiB | 1.13× |