Model comparison
whisper-large-v3 vs whisper-large-v3-turbo
At Q4_K_M, whisper-large-v3-turbo is the smaller download — 536,069,728 bytes against 997,303,008.
From the file· summed bytes, KV per layer
Side by side
| whisper-large-v3 | whisper-large-v3-turbo | |
|---|---|---|
| Parameters | 1.5B | 809M |
| Architecture | whisper | whisper |
| Layers | 32 | 32 |
| Native context | — | — |
| Mixture of experts | no | no |
| Quantizations published | 18 | 31 |
| Smallest quantization | 0.49 GiB | 0.27 GiB |
| Q4_K_M | 0.93 GiB | 0.50 GiB |
| Licence | apache-2.0 | apache-2.0 |
KV cache by context
the term that decides long-context viability
| Context | whisper-large-v3 | whisper-large-v3-turbo | Ratio |
|---|---|---|---|
| 4,096 | — | — | — |
| 8,192 | — | — | — |
| 16,384 | — | — | — |
| 32,768 | — | — | — |
| 65,536 | — | — | — |
| 131,072 | — | — | — |