Model comparison
granite-speech-4.1-2b-nar vs Voxtral-Mini-4B-Realtime-2602
At Q4_K_M, granite-speech-4.1-2b-nar is the smaller download — 1,560,008,832 bytes against 2,830,493,984.
From the file· summed bytes, KV per layer
Side by side
| granite-speech-4.1-2b-nar | Voxtral-Mini-4B-Realtime-2602 | |
|---|---|---|
| Parameters | 2.3B | 4.4B |
| Architecture | granite_speech_nar | voxtral_realtime |
| Layers | 40 | 26 |
| Native context | 4,096 | 131,072 |
| Mixture of experts | no | no |
| Quantizations published | 8 | 8 |
| Smallest quantization | 1.45 GiB | 2.35 GiB |
| Q4_K_M | 1.45 GiB | 2.64 GiB |
| Licence | apache-2.0 | apache-2.0 |
KV cache by context
the term that decides long-context viability
| Context | granite-speech-4.1-2b-nar | Voxtral-Mini-4B-Realtime-2602 | Ratio |
|---|---|---|---|
| 4,096 | 0.31 GiB | — | — |
| 8,192 | 0.63 GiB | — | — |
| 16,384 | 1.25 GiB | — | — |
| 32,768 | 2.50 GiB | — | — |
| 65,536 | 5.00 GiB | — | — |
| 131,072 | 10.00 GiB | — | — |